blocks|key|2957172|text|下面是我用来检测文本块的另一种方法：|type|unstyled|depth|inlineStyleRanges|entityRanges|data|2957173|2957174|将图像转换为grayscale|ordered-list-item|2957175|Applied+threshold+(简单的二进制阈值，阈值为150+)|offset|length|2957176|2957177|+dilation可使图像中的线条变粗，从而使对象更紧凑，空白碎片更少。使用了较高的迭代次数值，因此膨胀非常繁重(13次迭代，也是为结果图像中对象的最佳results).|2957178|Identified轮廓精心挑选的，使用opencv+function.|2957179|Drew+a+text.|2957180|Optionally+(矩形)外接每个轮廓对象-它们中的每一个都框出一块findContours丢弃区域，这些区域不太可能是您正在搜索的对象(例如，文本块)，由于它们的大小，因为上面的算法还可以找到相交或嵌套的对象(比如第一张卡片的整个顶部区域)，其中一些对象可能对您的目的毫无兴趣。|2957181|2957182|下面是用python和pyopencv编写的代码，它应该很容易移植到C%2B%2B。|2957183|import+cv2

image+=+cv2.imread("card.png")
gray+=+cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)+#+grayscale
_,thresh+=+cv2.threshold(gray,150,255,cv2.THRESH_BINARY_INV)+#+threshold
kernel+=+cv2.getStructuringElement(cv2.MORPH_CROSS,(3,3))
dilated+=+cv2.dilate(thresh,kernel,iterations+=+13)+#+dilate
_,+contours,+hierarchy+=+cv2.findContours(dilated,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_NONE)+#+get+contours

#+for+each+contour+found,+draw+a+rectangle+around+it+on+original+image
for+contour+in+contours:
++++#+get+rectangle+bounding+contour
++++[x,y,w,h]+=+cv2.boundingRect(contour)

++++#+discard+areas+that+are+too+large
++++if+h>300+and+w>300:
++++++++continue

++++#+discard+areas+that+are+too+small
++++if+h<40+or+w<40:
++++++++continue

++++#+draw+rectangle+around+contour+on+original+image
++++cv2.rectangle(image,(x,y),(x%2Bw,y%2Bh),(255,0,255),2)

#+write+original+image+with+added+contours+to+disk++
cv2.imwrite("contoured.jpg",+image)+|code-block|syntax|javascript|2957184|原始图像是您帖子中的第一张图像。|2957185|经过预处理(灰度、阈值和扩张-所以在步骤3之后)图像看起来像这样：|2957186|​|2957187|📷|atomic|2957188|2957189|下面是生成的图像(最后一行中的“contoured.jpg”)；图像中对象的最终边界框如下所示：|2957190|2957191|2957192|2957193|您可以看到，左侧的文本块被检测为独立的块，与其周围环境分隔开来。|2957194|使用具有相同参数的相同脚本(除了如下所述的第二张图像的阈值类型已更改)，以下是其他两张卡的结果：|2957195|2957196|2957197|2957198|2957199|2957200|2957201|调整参数|2957202|参数(阈值、膨胀参数)针对此图像和此任务(查找文本块)进行了优化，如果需要，可以针对要查找的其他卡片图像或其他类型的对象进行调整。|2957203|对于阈值(步骤2)，我使用了黑色阈值。对于文本比背景亮的图像，例如文章中的第二张图像，应该使用白色阈值，因此请将文本类型替换为cv2.THRESH_BINARY)。对于第二张图像，我也使用了一个稍高一点的阈值(180)。改变用于扩张的阈值和迭代次数的参数将导致在界定图像中的对象时不同程度的敏感性。|style|CODE|2957204|查找其他对象类型的：|BOLD|2957205|例如，将第一张图像中的扩展减少到5次迭代，可以让我们对图像中的对象进行更精细的划分，大致找到图像中的所有单词(而不是文本块)：|2957206|2957207|2957208|2957209|知道了单词的大致大小，我在这里丢弃了太小(小于20像素的宽度或高度)或太大(大于100像素的宽度或高度)而不能忽略不太可能是单词的对象的区域，以便在上面的图像中获得结果。|2957210|entityMap|0|LINK|mutability|MUTABLE|url|http://docs.opencv.org/modules/imgproc/doc/miscellaneous_transformations.html?highlight=threshold#threshold|1|http://docs.opencv.org/modules/imgproc/doc/filtering.html?highlight=dilate#cv2.dilate|2|http://docs.opencv.org/modules/imgproc/doc/structural_analysis_and_shape_descriptors.html?highlight=findcontours#cv2.findContours|3|IMAGE|IMMUTABLE|imageUrl|https://ask.qcloudimg.com/http-save/yehe-900000/9630576f8d6a20cad161735fbf196e89.jpeg|imageAlt|4|https://ask.qcloudimg.com/http-save/yehe-900000/6fe5cf2a27e3676c3ec916b9d4a72c12.jpeg|5|https://ask.qcloudimg.com/http-save/yehe-900000/ae867928ee757feacaf0d19d57410d77.jpeg|6|https://ask.qcloudimg.com/http-save/yehe-900000/526a3300f6e3acec42d28cd1b8ffa42a.jpeg|7|https://ask.qcloudimg.com/http-save/yehe-900000/8ade3db2f8fdec69eb364cdda2947604.jpeg^0|0|0|0|8|9|0|0|0|1|8|1|0|0|0|10|C|2|0|0|0|0|0|0|0|0|1|3|0|0|0|0|0|1|4|0|0|0|0|0|0|1|5|0|0|0|0|1|6|0|0|0|0|1R|H|0|9|1|0|0|0|0|1|7|0|0|0^^$0|@$1|2|3|4|5|6|7|2X|8|@]|9|@]|A|$]]|$1|B|3|-4|5|6|7|2Y|8|@]|9|@]|A|$]]|$1|C|3|D|5|E|7|2Z|8|@]|9|@]|A|$]]|$1|F|3|G|5|E|7|30|8|@]|9|@$H|31|I|32|1|33]]|A|$]]|$1|J|3|-4|5|E|7|34|8|@]|9|@]|A|$]]|$1|K|3|L|5|E|7|35|8|@]|9|@$H|36|I|37|1|38]]|A|$]]|$1|M|3|N|5|E|7|39|8|@]|9|@]|A|$]]|$1|O|3|P|5|E|7|3A|8|@]|9|@]|A|$]]|$1|Q|3|R|5|E|7|3B|8|@]|9|@$H|3C|I|3D|1|3E]]|A|$]]|$1|S|3|-4|5|6|7|3F|8|@]|9|@]|A|$]]|$1|T|3|U|5|6|7|3G|8|@]|9|@]|A|$]]|$1|V|3|W|5|X|7|3H|8|@]|9|@]|A|$Y|Z]]|$1|10|3|11|5|6|7|3I|8|@]|9|@]|A|$]]|$1|12|3|13|5|6|7|3J|8|@]|9|@]|A|$]]|$1|14|3|15|5|6|7|3K|8|@]|9|@]|A|$]]|$1|16|3|17|5|18|7|3L|8|@]|9|@$H|3M|I|3N|1|3O]]|A|$]]|$1|19|3|15|5|6|7|3P|8|@]|9|@]|A|$]]|$1|1A|3|1B|5|6|7|3Q|8|@]|9|@]|A|$]]|$1|1C|3|15|5|6|7|3R|8|@]|9|@]|A|$]]|$1|1D|3|17|5|18|7|3S|8|@]|9|@$H|3T|I|3U|1|3V]]|A|$]]|$1|1E|3|15|5|6|7|3W|8|@]|9|@]|A|$]]|$1|1F|3|1G|5|6|7|3X|8|@]|9|@]|A|$]]|$1|1H|3|1I|5|6|7|3Y|8|@]|9|@]|A|$]]|$1|1J|3|15|5|6|7|3Z|8|@]|9|@]|A|$]]|$1|1K|3|17|5|18|7|40|8|@]|9|@$H|41|I|42|1|43]]|A|$]]|$1|1L|3|15|5|6|7|44|8|@]|9|@]|A|$]]|$1|1M|3|15|5|6|7|45|8|@]|9|@]|A|$]]|$1|1N|3|17|5|18|7|46|8|@]|9|@$H|47|I|48|1|49]]|A|$]]|$1|1O|3|15|5|6|7|4A|8|@]|9|@]|A|$]]|$1|1P|3|1Q|5|6|7|4B|8|@]|9|@]|A|$]]|$1|1R|3|1S|5|6|7|4C|8|@]|9|@]|A|$]]|$1|1T|3|1U|5|6|7|4D|8|@$H|4E|I|4F|1V|1W]]|9|@]|A|$]]|$1|1X|3|1Y|5|6|7|4G|8|@$H|4H|I|4I|1V|1Z]]|9|@]|A|$]]|$1|20|3|21|5|6|7|4J|8|@]|9|@]|A|$]]|$1|22|3|15|5|6|7|4K|8|@]|9|@]|A|$]]|$1|23|3|17|5|18|7|4L|8|@]|9|@$H|4M|I|4N|1|4O]]|A|$]]|$1|24|3|15|5|6|7|4P|8|@]|9|@]|A|$]]|$1|25|3|26|5|6|7|4Q|8|@]|9|@]|A|$]]|$1|27|3|-4|5|6|7|4R|8|@]|9|@]|A|$]]]|28|$29|$5|2A|2B|2C|A|$2D|2E]]|2F|$5|2A|2B|2C|A|$2D|2G]]|2H|$5|2A|2B|2C|A|$2D|2I]]|2J|$5|2K|2B|2L|A|$2M|2N|2O|-4]]|2P|$5|2K|2B|2L|A|$2M|2Q|2O|-4]]|2R|$5|2K|2B|2L|A|$2M|2S|2O|-4]]|2T|$5|2K|2B|2L|A|$2M|2U|2O|-4]]|2V|$5|2K|2B|2L|A|$2M|2W|2O|-4]]]]

Here is an alternative approach that I used to detect the text blocks:
<ol>
<li>Converted the image to grayscale</li>
<li>Applied <a href="http://docs.opencv.org/modules/imgproc/doc/miscellaneous_transformations.html?highlight=threshold#threshold" rel="noreferrer">threshold</a> (simple binary threshold, with a handpicked value of 150 as the threshold value)</li>
<li>Applied <a href="http://docs.opencv.org/modules/imgproc/doc/filtering.html?highlight=dilate#cv2.dilate" rel="noreferrer">dilation</a> to thicken lines in image, leading to more compact objects and less white space fragments. Used a high value for number of iterations, so dilation is very heavy (13 iterations, also handpicked for optimal results).</li>
<li>Identified contours of objects in resulted image using opencv <a href="http://docs.opencv.org/modules/imgproc/doc/structural_analysis_and_shape_descriptors.html?highlight=findcontours#cv2.findContours" rel="noreferrer">findContours</a> function.</li>
<li>Drew a <a href="http://docs.opencv.org/modules/imgproc/doc/structural_analysis_and_shape_descriptors.html?highlight=boundingrect#boundingrect" rel="noreferrer">bounding box</a> (rectangle) circumscribing each contoured object - each of them frames a block of text.</li>
<li>Optionally discarded areas that are unlikely to be the object you are searching for (e.g. text blocks) given their size, as the algorithm above can also find intersecting or nested objects (like the entire top area for the first card) some of which could be uninteresting for your purposes.</li>
</ol>
Below is the code written in python with pyopencv, it should easy to port to C++.
<pre><code>import cv2

image = cv2.imread(&quot;card.png&quot;)
gray = cv2.cvtColor(image,cv2.COLOR_BGR2GRAY) # grayscale
_,thresh = cv2.threshold(gray,150,255,cv2.THRESH_BINARY_INV) # threshold
kernel = cv2.getStructuringElement(cv2.MORPH_CROSS,(3,3))
dilated = cv2.dilate(thresh,kernel,iterations = 13) # dilate
_, contours, hierarchy = cv2.findContours(dilated,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_NONE) # get contours

# for each contour found, draw a rectangle around it on original image
for contour in contours:
 # get rectangle bounding contour
 [x,y,w,h] = cv2.boundingRect(contour)

 # discard areas that are too large
 if h&gt;300 and w&gt;300:
 continue

 # discard areas that are too small
 if h&lt;40 or w&lt;40:
 continue

 # draw rectangle around contour on original image
 cv2.rectangle(image,(x,y),(x+w,y+h),(255,0,255),2)

# write original image with added contours to disk 
cv2.imwrite(&quot;contoured.jpg&quot;, image) 
</code></pre>
The original image is the first image in your post.
After preprocessing (grayscale, threshold and dilate - so after step 3) the image looked like this:
<img src="https://i.stack.imgur.com/LaONX.jpg" alt="Dilated image" />
Below is the resulted image (&quot;contoured.jpg&quot; in the last line); the final bounding boxes for the objects in the image look like this:
<img src="https://i.stack.imgur.com/SjNBg.jpg" alt="enter image description here" />
You can see the text block on the left is detected as a separate block, delimited from its surroundings.
Using the same script with the same parameters (except for thresholding type that was changed for the second image like described below), here are the results for the other 2 cards:
<img src="https://i.stack.imgur.com/pUxbi.jpg" alt="enter image description here" />
<img src="https://i.stack.imgur.com/DEP4t.jpg" alt="enter image description here" />
<h2>Tuning the parameters</h2>
The parameters (threshold value, dilation parameters) were optimized for this image and this task (finding text blocks) and can be adjusted, if needed, for other cards images or other types of objects to be found.
For thresholding (step 2), I used a black threshold. For images where text is lighter than the background, such as the second image in your post, a white threshold should be used, so replace thesholding type with <code>cv2.THRESH_BINARY</code>). For the second image I also used a slightly higher value for the threshold (180). Varying the parameters for the threshold value and the number of iterations for dilation will result in different degrees of sensitivity in delimiting objects in the image.
Finding other object types:
For example, decreasing the dilation to 5 iterations in the first image gives us a more fine delimitation of objects in the image, roughly finding all words in the image (rather than text blocks):
<img src="https://i.stack.imgur.com/HkP0o.jpg" alt="enter image description here" />
Knowing the rough size of a word, here I discarded areas that were too small (below 20 pixels width or height) or too large (above 100 pixels width or height) to ignore objects that are unlikely to be words, to get the results in the above image.

blocks|key|2744688|text|你可以试试由易初才和田英利开发的this+method。|type|unstyled|depth|inlineStyleRanges|entityRanges|offset|length|data|2744689|它们还共享一个软件(该软件基于Opencv-1.0，应在Windows平台下运行)。你可以使用的(虽然没有源代码)。它将生成图像中的所有文本边界框(以彩色阴影显示)。通过应用于您的示例图像，您将获得以下结果：|2744690|注意:为了使结果更可靠，您可以进一步合并相邻的框。|2744691|​|2744692|📷|atomic|2744693|2744694|2744695|2744696|如果你的最终目标是识别图像中的文本，你可以进一步检查gttext，这是一个自由软件和地面真实工具，用于彩色图像和文本。源代码也是可用的。|style|BOLD|2744697|有了这个，你就可以得到像这样的识别文本：|2744698|2744699|2744700|2744701|2744702|2744703|2744704|2744705|entityMap|0|LINK|mutability|MUTABLE|url|http://media-lab.engr.ccny.cuny.edu/~cyi/project_scenetextdetection.html|1|IMAGE|IMMUTABLE|imageUrl|https://ask.qcloudimg.com/http-save/yehe-900000/4527a9b0d3708739359bec54ae148462.png|imageAlt|2|https://ask.qcloudimg.com/http-save/yehe-900000/f4ff5d95776131430387e90457d52ab4.jpeg|3|https://ask.qcloudimg.com/http-save/yehe-900000/b48e39df67b80273dad2c14941243fbf.jpeg|4|https://code.google.com/p/gttext/|5|https://ask.qcloudimg.com/http-save/yehe-900000/c868176f9c5891d766d5634eb3434cd0.png|6|https://ask.qcloudimg.com/http-save/yehe-900000/a205705cc26ac9692ac6ce3b8f1d60d7.png|7|https://ask.qcloudimg.com/http-save/yehe-900000/1858d4cff27203fa2fa0bdead64bd183.jpeg|8|https://ask.qcloudimg.com/http-save/yehe-900000/baeeb82fa0c521d00b336537ae1eb64c.png|9|https://ask.qcloudimg.com/http-save/yehe-900000/fcdea0642fc77de045b0c584da1e873f.jpeg|10|https://ask.qcloudimg.com/http-save/yehe-900000/75af58e2b1a0bcf859c1fbd9226ce0c7.png^0|G|B|0|0|0|0|0|0|1|1|0|0|1|2|0|0|1|3|0|0|11|V|Q|6|4|0|0|K|0|0|1|0|0|1|0|1|5|0|0|1|0|1|6|0|0|1|0|1|7|0|0|1|0|1|8|0|0|1|0|1|9|0|0|1|0|1|A|0^^$0|@$1|2|3|4|5|6|7|1Y|8|@]|9|@$A|1Z|B|20|1|21]]|C|$]]|$1|D|3|E|5|6|7|22|8|@]|9|@]|C|$]]|$1|F|3|G|5|6|7|23|8|@]|9|@]|C|$]]|$1|H|3|I|5|6|7|24|8|@]|9|@]|C|$]]|$1|J|3|K|5|L|7|25|8|@]|9|@$A|26|B|27|1|28]]|C|$]]|$1|M|3|K|5|L|7|29|8|@]|9|@$A|2A|B|2B|1|2C]]|C|$]]|$1|N|3|K|5|L|7|2D|8|@]|9|@$A|2E|B|2F|1|2G]]|C|$]]|$1|O|3|-4|5|6|7|2H|8|@]|9|@]|C|$]]|$1|P|3|Q|5|6|7|2I|8|@$A|2J|B|2K|R|S]]|9|@$A|2L|B|2M|1|2N]]|C|$]]|$1|T|3|U|5|6|7|2O|8|@$A|2P|B|2Q|R|S]]|9|@]|C|$]]|$1|V|3|I|5|6|7|2R|8|@$A|2S|B|2T|R|S]]|9|@]|C|$]]|$1|W|3|K|5|L|7|2U|8|@$A|2V|B|2W|R|S]]|9|@$A|2X|B|2Y|1|2Z]]|C|$]]|$1|X|3|K|5|L|7|30|8|@$A|31|B|32|R|S]]|9|@$A|33|B|34|1|35]]|C|$]]|$1|Y|3|K|5|L|7|36|8|@$A|37|B|38|R|S]]|9|@$A|39|B|3A|1|3B]]|C|$]]|$1|Z|3|K|5|L|7|3C|8|@$A|3D|B|3E|R|S]]|9|@$A|3F|B|3G|1|3H]]|C|$]]|$1|10|3|K|5|L|7|3I|8|@$A|3J|B|3K|R|S]]|9|@$A|3L|B|3M|1|3N]]|C|$]]|$1|11|3|K|5|L|7|3O|8|@$A|3P|B|3Q|R|S]]|9|@$A|3R|B|3S|1|3T]]|C|$]]|$1|12|3|-4|5|6|7|3U|8|@]|9|@]|C|$]]]|13|$14|$5|15|16|17|C|$18|19]]|1A|$5|1B|16|1C|C|$1D|1E|1F|-4]]|1G|$5|1B|16|1C|C|$1D|1H|1F|-4]]|1I|$5|1B|16|1C|C|$1D|1J|1F|-4]]|1K|$5|15|16|17|C|$18|1L]]|1M|$5|1B|16|1C|C|$1D|1N|1F|-4]]|1O|$5|1B|16|1C|C|$1D|1P|1F|-4]]|1Q|$5|1B|16|1C|C|$1D|1R|1F|-4]]|1S|$5|1B|16|1C|C|$1D|1T|1F|-4]]|1U|$5|1B|16|1C|C|$1D|1V|1F|-4]]|1W|$5|1B|16|1C|C|$1D|1X|1F|-4]]]]

You can try <a href="http://media-lab.engr.ccny.cuny.edu/~cyi/project_scenetextdetection.html" rel="noreferrer">this method</a> that is developed by Chucai Yi and Yingli Tian.

They also share a software (which is based on Opencv-1.0 and it should run under Windows platform.) that you can use (though no source code available). It will generate all the text bounding boxes (shown in color shadows) in the image. By applying to your sample images, you will get the following results:

Note: to make the result more robust, you can further merge adjacent boxes together.

<img src="https://i.stack.imgur.com/HBnqK.png" width="420" height="240"> 
<img src="https://i.stack.imgur.com/ExRBy.jpg" width="420" height="240"> 
<img src="https://i.stack.imgur.com/za10Q.jpg" width="420" height="240"> 

<hr>

Update: If your ultimate goal is to recognize the texts in the image, you can further check out <a href="https://code.google.com/p/gttext/" rel="noreferrer">gttext</a>, which is an OCR free software and Ground Truthing tool for Color Images with Text. Source code is also available.

With this, you can get recognized texts like:

<img src="https://i.stack.imgur.com/VfDfJ.png" width="330" height="180"> 
<img src="https://i.stack.imgur.com/K8wrq.png" width="320" height="180"> 
<img src="https://i.stack.imgur.com/7h4SJ.jpg" width="330" height="180"> 
<img src="https://i.stack.imgur.com/9tZdn.png" width="230" height="180"> 
<img src="https://i.stack.imgur.com/8ipeJ.jpg" width="330" height="180">
<img src="https://i.stack.imgur.com/csb3p.png" width="320" height="180">

blocks|key|155275|text|您可以通过查找闭合边缘元素(灵感来自LPD)来检测文本：|type|unstyled|depth|inlineStyleRanges|entityRanges|data|155276|#include+"opencv2/opencv.hpp"

std::vector<cv::Rect>+detectLetters(cv::Mat+img)
{
++++std::vector<cv::Rect>+boundRect;
++++cv::Mat+img_gray,+img_sobel,+img_threshold,+element;
++++cvtColor(img,+img_gray,+CV_BGR2GRAY);
++++cv::Sobel(img_gray,+img_sobel,+CV_8U,+1,+0,+3,+1,+0,+cv::BORDER_DEFAULT);
++++cv::threshold(img_sobel,+img_threshold,+0,+255,+CV_THRESH_OTSU%2BCV_THRESH_BINARY);
++++element+=+getStructuringElement(cv::MORPH_RECT,+cv::Size(17,+3)+);
++++cv::morphologyEx(img_threshold,+img_threshold,+CV_MOP_CLOSE,+element);+//Does+the+trick
++++std::vector<+std::vector<+cv::Point>+>+contours;
++++cv::findContours(img_threshold,+contours,+0,+1);+
++++std::vector<std::vector<cv::Point>+>+contours_poly(+contours.size()+);
++++for(+int+i+=+0;+i+<+contours.size();+i%2B%2B+)
++++++++if+(contours[i].size()>100)
++++++++{+
++++++++++++cv::approxPolyDP(+cv::Mat(contours[i]),+contours_poly[i],+3,+true+);
++++++++++++cv::Rect+appRect(+boundingRect(+cv::Mat(contours_poly[i])+));
++++++++++++if+(appRect.width>appRect.height)+
++++++++++++++++boundRect.push_back(appRect);
++++++++}
++++return+boundRect;
}|code-block|syntax|javascript|155277|用法：|155278|int+main(int+argc,char**+argv)
{
++++//Read
++++cv::Mat+img1=cv::imread("side_1.jpg");
++++cv::Mat+img2=cv::imread("side_2.jpg");
++++//Detect
++++std::vector<cv::Rect>+letterBBoxes1=detectLetters(img1);
++++std::vector<cv::Rect>+letterBBoxes2=detectLetters(img2);
++++//Display
++++for(int+i=0;+i<+letterBBoxes1.size();+i%2B%2B)
++++++++cv::rectangle(img1,letterBBoxes1[i],cv::Scalar(0,255,0),3,8,0);
++++cv::imwrite(+"imgOut1.jpg",+img1);++
++++for(int+i=0;+i<+letterBBoxes2.size();+i%2B%2B)
++++++++cv::rectangle(img2,letterBBoxes2[i],cv::Scalar(0,255,0),3,8,0);
++++cv::imwrite(+"imgOut2.jpg",+img2);++
++++return+0;
}|155279|结果：|155280|a.元素=+getStructuringElement(cv::MORPH_RECT，cv::Size(17，3)+)；|155281|📷|atomic|offset|length|155282|155283|​|155284|b.元素=+getStructuringElement(cv::MORPH_RECT，cv::Size(30，30)+)；|155285|155286|155287|155288|对于提到的另一个图像，结果是相似的。|155289|entityMap|0|IMAGE|mutability|IMMUTABLE|imageUrl|https://ask.qcloudimg.com/http-save/yehe-900000/9f81d19c465b1cf751cccfbaa40ca287.jpeg|imageAlt|1|https://ask.qcloudimg.com/http-save/yehe-900000/29f3607c605316dbcbdff66e58506543.jpeg|2|https://ask.qcloudimg.com/http-save/yehe-900000/673aac6160ecf57b19dd7cdbe0731021.jpeg|3|https://ask.qcloudimg.com/http-save/yehe-900000/6c4e96f3d1489b4c95ffd1ddd98e1951.jpeg^0|0|0|0|0|0|0|0|1|0|0|0|1|1|0|0|0|0|1|2|0|0|1|3|0|0|0^^$0|@$1|2|3|4|5|6|7|1I|8|@]|9|@]|A|$]]|$1|B|3|C|5|D|7|1J|8|@]|9|@]|A|$E|F]]|$1|G|3|H|5|6|7|1K|8|@]|9|@]|A|$]]|$1|I|3|J|5|D|7|1L|8|@]|9|@]|A|$E|F]]|$1|K|3|L|5|6|7|1M|8|@]|9|@]|A|$]]|$1|M|3|N|5|6|7|1N|8|@]|9|@]|A|$]]|$1|O|3|P|5|Q|7|1O|8|@]|9|@$R|1P|S|1Q|1|1R]]|A|$]]|$1|T|3|P|5|Q|7|1S|8|@]|9|@$R|1T|S|1U|1|1V]]|A|$]]|$1|U|3|V|5|6|7|1W|8|@]|9|@]|A|$]]|$1|W|3|X|5|6|7|1X|8|@]|9|@]|A|$]]|$1|Y|3|P|5|Q|7|1Y|8|@]|9|@$R|1Z|S|20|1|21]]|A|$]]|$1|Z|3|P|5|Q|7|22|8|@]|9|@$R|23|S|24|1|25]]|A|$]]|$1|10|3|V|5|6|7|26|8|@]|9|@]|A|$]]|$1|11|3|12|5|6|7|27|8|@]|9|@]|A|$]]|$1|13|3|-4|5|6|7|28|8|@]|9|@]|A|$]]]|14|$15|$5|16|17|18|A|$19|1A|1B|-4]]|1C|$5|16|17|18|A|$19|1D|1B|-4]]|1E|$5|16|17|18|A|$19|1F|1B|-4]]|1G|$5|16|17|18|A|$19|1H|1B|-4]]]]

You can detect text by finding close edge elements (inspired from a LPD):

<pre><code>#include "opencv2/opencv.hpp"

std::vector&lt;cv::Rect&gt; detectLetters(cv::Mat img)
{
 std::vector&lt;cv::Rect&gt; boundRect;
 cv::Mat img_gray, img_sobel, img_threshold, element;
 cvtColor(img, img_gray, CV_BGR2GRAY);
 cv::Sobel(img_gray, img_sobel, CV_8U, 1, 0, 3, 1, 0, cv::BORDER_DEFAULT);
 cv::threshold(img_sobel, img_threshold, 0, 255, CV_THRESH_OTSU+CV_THRESH_BINARY);
 element = getStructuringElement(cv::MORPH_RECT, cv::Size(17, 3) );
 cv::morphologyEx(img_threshold, img_threshold, CV_MOP_CLOSE, element); //Does the trick
 std::vector&lt; std::vector&lt; cv::Point&gt; &gt; contours;
 cv::findContours(img_threshold, contours, 0, 1); 
 std::vector&lt;std::vector&lt;cv::Point&gt; &gt; contours_poly( contours.size() );
 for( int i = 0; i &lt; contours.size(); i++ )
 if (contours[i].size()&gt;100)
 { 
 cv::approxPolyDP( cv::Mat(contours[i]), contours_poly[i], 3, true );
 cv::Rect appRect( boundingRect( cv::Mat(contours_poly[i]) ));
 if (appRect.width&gt;appRect.height) 
 boundRect.push_back(appRect);
 }
 return boundRect;
}
</code></pre>

Usage:

<pre><code>int main(int argc,char** argv)
{
 //Read
 cv::Mat img1=cv::imread("side_1.jpg");
 cv::Mat img2=cv::imread("side_2.jpg");
 //Detect
 std::vector&lt;cv::Rect&gt; letterBBoxes1=detectLetters(img1);
 std::vector&lt;cv::Rect&gt; letterBBoxes2=detectLetters(img2);
 //Display
 for(int i=0; i&lt; letterBBoxes1.size(); i++)
 cv::rectangle(img1,letterBBoxes1[i],cv::Scalar(0,255,0),3,8,0);
 cv::imwrite( "imgOut1.jpg", img1); 
 for(int i=0; i&lt; letterBBoxes2.size(); i++)
 cv::rectangle(img2,letterBBoxes2[i],cv::Scalar(0,255,0),3,8,0);
 cv::imwrite( "imgOut2.jpg", img2); 
 return 0;
}
</code></pre>

Results:

a. element = getStructuringElement(cv::MORPH_RECT, cv::Size(17, 3) );
<img src="https://i.stack.imgur.com/VmMJZ.jpg" alt="imgOut1">
<img src="https://i.stack.imgur.com/0s0U2.jpg" alt="imgOut2">

b. element = getStructuringElement(cv::MORPH_RECT, cv::Size(30, 30) );
<img src="https://i.stack.imgur.com/4oBM3.jpg" alt="imgOut1">
<img src="https://i.stack.imgur.com/DRW5p.jpg" alt="imgOut2">

Results are similar for the other image mentioned.

blocks|key|156475|text|我在下面的程序中使用了基于梯度的方法。添加了生成的图像。请注意，我正在使用图像的缩小版本进行处理。|type|unstyled|depth|inlineStyleRanges|entityRanges|data|156476|c%2B%2B版本|156477|The+MIT+License+(MIT)

Copyright+(c)+2014+Dhanushka+Dangampola

Permission+is+hereby+granted,+free+of+charge,+to+any+person+obtaining+a+copy
of+this+software+and+associated+documentation+files+(the+"Software"),+to+deal
in+the+Software+without+restriction,+including+without+limitation+the+rights
to+use,+copy,+modify,+merge,+publish,+distribute,+sublicense,+and/or+sell
copies+of+the+Software,+and+to+permit+persons+to+whom+the+Software+is
furnished+to+do+so,+subject+to+the+following+conditions:

The+above+copyright+notice+and+this+permission+notice+shall+be+included+in
all+copies+or+substantial+portions+of+the+Software.

THE+SOFTWARE+IS+PROVIDED+"AS+IS",+WITHOUT+WARRANTY+OF+ANY+KIND,+EXPRESS+OR
IMPLIED,+INCLUDING+BUT+NOT+LIMITED+TO+THE+WARRANTIES+OF+MERCHANTABILITY,
FITNESS+FOR+A+PARTICULAR+PURPOSE+AND+NONINFRINGEMENT.+IN+NO+EVENT+SHALL+THE
AUTHORS+OR+COPYRIGHT+HOLDERS+BE+LIABLE+FOR+ANY+CLAIM,+DAMAGES+OR+OTHER
LIABILITY,+WHETHER+IN+AN+ACTION+OF+CONTRACT,+TORT+OR+OTHERWISE,+ARISING+FROM,
OUT+OF+OR+IN+CONNECTION+WITH+THE+SOFTWARE+OR+THE+USE+OR+OTHER+DEALINGS+IN
THE+SOFTWARE.

#include+"stdafx.h"

#include+<opencv2/core/core.hpp>
#include+<opencv2/highgui/highgui.hpp>
#include+<opencv2/imgproc/imgproc.hpp>
#include+<iostream>

using+namespace+cv;
using+namespace+std;

#define+INPUT_FILE++++++++++++++"1.jpg"
#define+OUTPUT_FOLDER_PATH++++++string("")

int+_tmain(int+argc,+_TCHAR*+argv[])
{
++++Mat+large+=+imread(INPUT_FILE);
++++Mat+rgb;
++++//+downsample+and+use+it+for+processing
++++pyrDown(large,+rgb);
++++Mat+small;
++++cvtColor(rgb,+small,+CV_BGR2GRAY);
++++//+morphological+gradient
++++Mat+grad;
++++Mat+morphKernel+=+getStructuringElement(MORPH_ELLIPSE,+Size(3,+3));
++++morphologyEx(small,+grad,+MORPH_GRADIENT,+morphKernel);
++++//+binarize
++++Mat+bw;
++++threshold(grad,+bw,+0.0,+255.0,+THRESH_BINARY+%7C+THRESH_OTSU);
++++//+connect+horizontally+oriented+regions
++++Mat+connected;
++++morphKernel+=+getStructuringElement(MORPH_RECT,+Size(9,+1));
++++morphologyEx(bw,+connected,+MORPH_CLOSE,+morphKernel);
++++//+find+contours
++++Mat+mask+=+Mat::zeros(bw.size(),+CV_8UC1);
++++vector<vector<Point>>+contours;
++++vector<Vec4i>+hierarchy;
++++findContours(connected,+contours,+hierarchy,+CV_RETR_CCOMP,+CV_CHAIN_APPROX_SIMPLE,+Point(0,+0));
++++//+filter+contours
++++for(int+idx+=+0;+idx+>=+0;+idx+=+hierarchy[idx][0])
++++{
++++++++Rect+rect+=+boundingRect(contours[idx]);
++++++++Mat+maskROI(mask,+rect);
++++++++maskROI+=+Scalar(0,+0,+0);
++++++++//+fill+the+contour
++++++++drawContours(mask,+contours,+idx,+Scalar(255,+255,+255),+CV_FILLED);
++++++++//+ratio+of+non-zero+pixels+in+the+filled+region
++++++++double+r+=+(double)countNonZero(maskROI)/(rect.width*rect.height);

++++++++if+(r+>+.45+/*+assume+at+least+45%25+of+the+area+is+filled+if+it+contains+text+*/
++++++++++++&&+
++++++++++++(rect.height+>+8+&&+rect.width+>+8)+/*+constraints+on+region+size+*/
++++++++++++/*+these+two+conditions+alone+are+not+very+robust.+better+to+use+something+
++++++++++++like+the+number+of+significant+peaks+in+a+horizontal+projection+as+a+third+condition+*/
++++++++++++)
++++++++{
++++++++++++rectangle(rgb,+rect,+Scalar(0,+255,+0),+2);
++++++++}
++++}
++++imwrite(OUTPUT_FOLDER_PATH+%2B+string("rgb.jpg"),+rgb);

++++return+0;
}|code-block|syntax|javascript|156478|python版本|156479|The+MIT+License+(MIT)

Copyright+(c)+2017+Dhanushka+Dangampola

Permission+is+hereby+granted,+free+of+charge,+to+any+person+obtaining+a+copy
of+this+software+and+associated+documentation+files+(the+"Software"),+to+deal
in+the+Software+without+restriction,+including+without+limitation+the+rights
to+use,+copy,+modify,+merge,+publish,+distribute,+sublicense,+and/or+sell
copies+of+the+Software,+and+to+permit+persons+to+whom+the+Software+is
furnished+to+do+so,+subject+to+the+following+conditions:

The+above+copyright+notice+and+this+permission+notice+shall+be+included+in
all+copies+or+substantial+portions+of+the+Software.

THE+SOFTWARE+IS+PROVIDED+"AS+IS",+WITHOUT+WARRANTY+OF+ANY+KIND,+EXPRESS+OR
IMPLIED,+INCLUDING+BUT+NOT+LIMITED+TO+THE+WARRANTIES+OF+MERCHANTABILITY,
FITNESS+FOR+A+PARTICULAR+PURPOSE+AND+NONINFRINGEMENT.+IN+NO+EVENT+SHALL+THE
AUTHORS+OR+COPYRIGHT+HOLDERS+BE+LIABLE+FOR+ANY+CLAIM,+DAMAGES+OR+OTHER
LIABILITY,+WHETHER+IN+AN+ACTION+OF+CONTRACT,+TORT+OR+OTHERWISE,+ARISING+FROM,
OUT+OF+OR+IN+CONNECTION+WITH+THE+SOFTWARE+OR+THE+USE+OR+OTHER+DEALINGS+IN
THE+SOFTWARE.

import+cv2
import+numpy+as+np

large+=+cv2.imread('1.jpg')
rgb+=+cv2.pyrDown(large)
small+=+cv2.cvtColor(rgb,+cv2.COLOR_BGR2GRAY)

kernel+=+cv2.getStructuringElement(cv2.MORPH_ELLIPSE,+(3,+3))
grad+=+cv2.morphologyEx(small,+cv2.MORPH_GRADIENT,+kernel)

_,+bw+=+cv2.threshold(grad,+0.0,+255.0,+cv2.THRESH_BINARY+%7C+cv2.THRESH_OTSU)

kernel+=+cv2.getStructuringElement(cv2.MORPH_RECT,+(9,+1))
connected+=+cv2.morphologyEx(bw,+cv2.MORPH_CLOSE,+kernel)
#+using+RETR_EXTERNAL+instead+of+RETR_CCOMP
contours,+hierarchy+=+cv2.findContours(connected.copy(),+cv2.RETR_EXTERNAL,+cv2.CHAIN_APPROX_NONE)
#For+opencv+3%2B+comment+the+previous+line+and+uncomment+the+following+line
#_,+contours,+hierarchy+=+cv2.findContours(connected.copy(),+cv2.RETR_EXTERNAL,+cv2.CHAIN_APPROX_NONE)

mask+=+np.zeros(bw.shape,+dtype=np.uint8)

for+idx+in+range(len(contours)):
++++x,+y,+w,+h+=+cv2.boundingRect(contours[idx])
++++mask[y:y%2Bh,+x:x%2Bw]+=+0
++++cv2.drawContours(mask,+contours,+idx,+(255,+255,+255),+-1)
++++r+=+float(cv2.countNonZero(mask[y:y%2Bh,+x:x%2Bw]))+/+(w+*+h)

++++if+r+>+0.45+and+w+>+8+and+h+>+8:
++++++++cv2.rectangle(rgb,+(x,+y),+(x%2Bw-1,+y%2Bh-1),+(0,+255,+0),+2)

cv2.imshow('rects',+rgb)|156480|​|156481|📷|atomic|offset|length|156482|156483|156484|156485|entityMap|0|IMAGE|mutability|IMMUTABLE|imageUrl|https://ask.qcloudimg.com/http-save/yehe-900000/03a4ab51a944081a3bcbd3412da2565f.jpeg|imageAlt|1|https://ask.qcloudimg.com/http-save/yehe-900000/c6d34ef57d2c3ea044bd1f3af35143ab.jpeg|2|https://ask.qcloudimg.com/http-save/yehe-900000/deadcac6153a91752b281a1d14e55923.jpeg^0|0|0|0|0|0|0|0|1|0|0|0|1|1|0|0|1|2|0|0^^$0|@$1|2|3|4|5|6|7|19|8|@]|9|@]|A|$]]|$1|B|3|C|5|6|7|1A|8|@]|9|@]|A|$]]|$1|D|3|E|5|F|7|1B|8|@]|9|@]|A|$G|H]]|$1|I|3|J|5|6|7|1C|8|@]|9|@]|A|$]]|$1|K|3|L|5|F|7|1D|8|@]|9|@]|A|$G|H]]|$1|M|3|N|5|6|7|1E|8|@]|9|@]|A|$]]|$1|O|3|P|5|Q|7|1F|8|@]|9|@$R|1G|S|1H|1|1I]]|A|$]]|$1|T|3|P|5|Q|7|1J|8|@]|9|@$R|1K|S|1L|1|1M]]|A|$]]|$1|U|3|P|5|Q|7|1N|8|@]|9|@$R|1O|S|1P|1|1Q]]|A|$]]|$1|V|3|N|5|6|7|1R|8|@]|9|@]|A|$]]|$1|W|3|-4|5|6|7|1S|8|@]|9|@]|A|$]]]|X|$Y|$5|Z|10|11|A|$12|13|14|-4]]|15|$5|Z|10|11|A|$12|16|14|-4]]|17|$5|Z|10|11|A|$12|18|14|-4]]]]

I used a gradient based method in the program below. Added the resulting images. Please note that I'm using a scaled down version of the image for processing.

c++ version

<pre><code>The MIT License (MIT)

Copyright (c) 2014 Dhanushka Dangampola

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.

#include "stdafx.h"

#include &lt;opencv2/core/core.hpp&gt;
#include &lt;opencv2/highgui/highgui.hpp&gt;
#include &lt;opencv2/imgproc/imgproc.hpp&gt;
#include &lt;iostream&gt;

using namespace cv;
using namespace std;

#define INPUT_FILE "1.jpg"
#define OUTPUT_FOLDER_PATH string("")

int _tmain(int argc, _TCHAR* argv[])
{
 Mat large = imread(INPUT_FILE);
 Mat rgb;
 // downsample and use it for processing
 pyrDown(large, rgb);
 Mat small;
 cvtColor(rgb, small, CV_BGR2GRAY);
 // morphological gradient
 Mat grad;
 Mat morphKernel = getStructuringElement(MORPH_ELLIPSE, Size(3, 3));
 morphologyEx(small, grad, MORPH_GRADIENT, morphKernel);
 // binarize
 Mat bw;
 threshold(grad, bw, 0.0, 255.0, THRESH_BINARY | THRESH_OTSU);
 // connect horizontally oriented regions
 Mat connected;
 morphKernel = getStructuringElement(MORPH_RECT, Size(9, 1));
 morphologyEx(bw, connected, MORPH_CLOSE, morphKernel);
 // find contours
 Mat mask = Mat::zeros(bw.size(), CV_8UC1);
 vector&lt;vector&lt;Point&gt;&gt; contours;
 vector&lt;Vec4i&gt; hierarchy;
 findContours(connected, contours, hierarchy, CV_RETR_CCOMP, CV_CHAIN_APPROX_SIMPLE, Point(0, 0));
 // filter contours
 for(int idx = 0; idx &gt;= 0; idx = hierarchy[idx][0])
 {
 Rect rect = boundingRect(contours[idx]);
 Mat maskROI(mask, rect);
 maskROI = Scalar(0, 0, 0);
 // fill the contour
 drawContours(mask, contours, idx, Scalar(255, 255, 255), CV_FILLED);
 // ratio of non-zero pixels in the filled region
 double r = (double)countNonZero(maskROI)/(rect.width*rect.height);

 if (r &gt; .45 /* assume at least 45% of the area is filled if it contains text */
 &amp;&amp; 
 (rect.height &gt; 8 &amp;&amp; rect.width &gt; 8) /* constraints on region size */
 /* these two conditions alone are not very robust. better to use something 
 like the number of significant peaks in a horizontal projection as a third condition */
 )
 {
 rectangle(rgb, rect, Scalar(0, 255, 0), 2);
 }
 }
 imwrite(OUTPUT_FOLDER_PATH + string("rgb.jpg"), rgb);

 return 0;
}
</code></pre>

python version

<pre><code>The MIT License (MIT)

Copyright (c) 2017 Dhanushka Dangampola

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in
all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
THE SOFTWARE.

import cv2
import numpy as np

large = cv2.imread('1.jpg')
rgb = cv2.pyrDown(large)
small = cv2.cvtColor(rgb, cv2.COLOR_BGR2GRAY)

kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
grad = cv2.morphologyEx(small, cv2.MORPH_GRADIENT, kernel)

_, bw = cv2.threshold(grad, 0.0, 255.0, cv2.THRESH_BINARY | cv2.THRESH_OTSU)

kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (9, 1))
connected = cv2.morphologyEx(bw, cv2.MORPH_CLOSE, kernel)
# using RETR_EXTERNAL instead of RETR_CCOMP
contours, hierarchy = cv2.findContours(connected.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
#For opencv 3+ comment the previous line and uncomment the following line
#_, contours, hierarchy = cv2.findContours(connected.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)

mask = np.zeros(bw.shape, dtype=np.uint8)

for idx in range(len(contours)):
 x, y, w, h = cv2.boundingRect(contours[idx])
 mask[y:y+h, x:x+w] = 0
 cv2.drawContours(mask, contours, idx, (255, 255, 255), -1)
 r = float(cv2.countNonZero(mask[y:y+h, x:x+w])) / (w * h)

 if r &gt; 0.45 and w &gt; 8 and h &gt; 8:
 cv2.rectangle(rgb, (x, y), (x+w-1, y+h-1), (0, 255, 0), 2)

cv2.imshow('rects', rgb)
</code></pre>

<img src="https://i.stack.imgur.com/rHeQ9.jpg" alt="enter image description here">
<img src="https://i.stack.imgur.com/jUTUK.jpg" alt="enter image description here">
<img src="https://i.stack.imgur.com/9jxCD.jpg" alt="enter image description here">

blocks|key|2745087|text|@dhanushka的方法显示了最大的希望，但我想使用Python，所以为了好玩，我继续翻译它：|type|unstyled|depth|inlineStyleRanges|entityRanges|data|2745088|import+cv2
import+numpy+as+np
from+cv2+import+boundingRect,+countNonZero,+cvtColor,+drawContours,+findContours,+getStructuringElement,+imread,+morphologyEx,+pyrDown,+rectangle,+threshold

large+=+imread(image_path)
#+downsample+and+use+it+for+processing
rgb+=+pyrDown(large)
#+apply+grayscale
small+=+cvtColor(rgb,+cv2.COLOR_BGR2GRAY)
#+morphological+gradient
morph_kernel+=+getStructuringElement(cv2.MORPH_ELLIPSE,+(3,+3))
grad+=+morphologyEx(small,+cv2.MORPH_GRADIENT,+morph_kernel)
#+binarize
_,+bw+=+threshold(src=grad,+thresh=0,+maxval=255,+type=cv2.THRESH_BINARY%2Bcv2.THRESH_OTSU)
morph_kernel+=+getStructuringElement(cv2.MORPH_RECT,+(9,+1))
#+connect+horizontally+oriented+regions
connected+=+morphologyEx(bw,+cv2.MORPH_CLOSE,+morph_kernel)
mask+=+np.zeros(bw.shape,+np.uint8)
#+find+contours
im2,+contours,+hierarchy+=+findContours(connected,+cv2.RETR_CCOMP,+cv2.CHAIN_APPROX_SIMPLE)
#+filter+contours
for+idx+in+range(0,+len(hierarchy[0])):
++++rect+=+x,+y,+rect_width,+rect_height+=+boundingRect(contours[idx])
++++#+fill+the+contour
++++mask+=+drawContours(mask,+contours,+idx,+(255,+255,+2555),+cv2.FILLED)
++++#+ratio+of+non-zero+pixels+in+the+filled+region
++++r+=+float(countNonZero(mask))+/+(rect_width+*+rect_height)
++++if+r+>+0.45+and+rect_height+>+8+and+rect_width+>+8:
++++++++rgb+=+rectangle(rgb,+(x,+y%2Brect_height),+(x%2Brect_width,+y),+(0,255,0),3)|code-block|syntax|javascript|2745089|现在显示图像：|2745090|from+PIL+import+Image
Image.fromarray(rgb).show()|2745091|不是最有Pythonic风格的脚本，但我尝试尽可能地与原始C%2B%2B代码相似，以便读者能够理解。|2745092|它几乎和原版一样好用。我很乐意阅读如何改进/修复它，使其与原始结果完全相似的建议。|2745093|​|2745094|📷|atomic|offset|length|2745095|2745096|2745097|2745098|2745099|2745100|2745101|2745102|entityMap|0|IMAGE|mutability|IMMUTABLE|imageUrl|https://ask.qcloudimg.com/http-save/yehe-900000/9a735d0c0f99cce7dda2d3a34021599d.jpeg|imageAlt|1|https://ask.qcloudimg.com/http-save/yehe-900000/7ca35942e59bde3d71d6435c15ce2b56.jpeg|2|https://ask.qcloudimg.com/http-save/yehe-900000/82fc4445ef6b9f2839601a38b54e1f25.jpeg^0|0|0|0|0|0|0|0|0|1|0|0|0|0|0|1|1|0|0|0|0|1|2|0|0^^$0|@$1|2|3|4|5|6|7|1F|8|@]|9|@]|A|$]]|$1|B|3|C|5|D|7|1G|8|@]|9|@]|A|$E|F]]|$1|G|3|H|5|6|7|1H|8|@]|9|@]|A|$]]|$1|I|3|J|5|D|7|1I|8|@]|9|@]|A|$E|F]]|$1|K|3|L|5|6|7|1J|8|@]|9|@]|A|$]]|$1|M|3|N|5|6|7|1K|8|@]|9|@]|A|$]]|$1|O|3|P|5|6|7|1L|8|@]|9|@]|A|$]]|$1|Q|3|R|5|S|7|1M|8|@]|9|@$T|1N|U|1O|1|1P]]|A|$]]|$1|V|3|P|5|6|7|1Q|8|@]|9|@]|A|$]]|$1|W|3|P|5|6|7|1R|8|@]|9|@]|A|$]]|$1|X|3|R|5|S|7|1S|8|@]|9|@$T|1T|U|1U|1|1V]]|A|$]]|$1|Y|3|P|5|6|7|1W|8|@]|9|@]|A|$]]|$1|Z|3|P|5|6|7|1X|8|@]|9|@]|A|$]]|$1|10|3|R|5|S|7|1Y|8|@]|9|@$T|1Z|U|20|1|21]]|A|$]]|$1|11|3|P|5|6|7|22|8|@]|9|@]|A|$]]|$1|12|3|-4|5|6|7|23|8|@]|9|@]|A|$]]]|13|$14|$5|15|16|17|A|$18|19|1A|-4]]|1B|$5|15|16|17|A|$18|1C|1A|-4]]|1D|$5|15|16|17|A|$18|1E|1A|-4]]]]

@dhanushka's approach showed the most promise but I wanted to play around in Python so went ahead and translated it for fun:

<pre><code>import cv2
import numpy as np
from cv2 import boundingRect, countNonZero, cvtColor, drawContours, findContours, getStructuringElement, imread, morphologyEx, pyrDown, rectangle, threshold

large = imread(image_path)
# downsample and use it for processing
rgb = pyrDown(large)
# apply grayscale
small = cvtColor(rgb, cv2.COLOR_BGR2GRAY)
# morphological gradient
morph_kernel = getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
grad = morphologyEx(small, cv2.MORPH_GRADIENT, morph_kernel)
# binarize
_, bw = threshold(src=grad, thresh=0, maxval=255, type=cv2.THRESH_BINARY+cv2.THRESH_OTSU)
morph_kernel = getStructuringElement(cv2.MORPH_RECT, (9, 1))
# connect horizontally oriented regions
connected = morphologyEx(bw, cv2.MORPH_CLOSE, morph_kernel)
mask = np.zeros(bw.shape, np.uint8)
# find contours
im2, contours, hierarchy = findContours(connected, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
# filter contours
for idx in range(0, len(hierarchy[0])):
 rect = x, y, rect_width, rect_height = boundingRect(contours[idx])
 # fill the contour
 mask = drawContours(mask, contours, idx, (255, 255, 2555), cv2.FILLED)
 # ratio of non-zero pixels in the filled region
 r = float(countNonZero(mask)) / (rect_width * rect_height)
 if r &gt; 0.45 and rect_height &gt; 8 and rect_width &gt; 8:
 rgb = rectangle(rgb, (x, y+rect_height), (x+rect_width, y), (0,255,0),3)
</code></pre>

Now to display the image:

<pre><code>from PIL import Image
Image.fromarray(rgb).show()
</code></pre>

Not the most Pythonic of scripts but I tried to resemble the original C++ code as closely as possible for readers to follow.

It works almost as well as the original. I'll be happy to read suggestions how it could be improved/fixed to resemble the original results fully.

<a href="https://i.stack.imgur.com/AaTsS.jpg" rel="noreferrer"><img src="https://i.stack.imgur.com/AaTsS.jpg" alt="enter image description here"></a>

<a href="https://i.stack.imgur.com/es2j9.jpg" rel="noreferrer"><img src="https://i.stack.imgur.com/es2j9.jpg" alt="enter image description here"></a>

<a href="https://i.stack.imgur.com/5fap7.jpg" rel="noreferrer"><img src="https://i.stack.imgur.com/5fap7.jpg" alt="enter image description here"></a>

blocks|key|156558|text|以上代码JAVA版本:谢谢@William|type|unstyled|depth|inlineStyleRanges|entityRanges|data|156559|public+static+List<Rect>+detectLetters(Mat+img){++++
++++List<Rect>+boundRect=new+ArrayList<>();

++++Mat+img_gray+=new+Mat(),+img_sobel=new+Mat(),+img_threshold=new+Mat(),+element=new+Mat();
++++Imgproc.cvtColor(img,+img_gray,+Imgproc.COLOR_RGB2GRAY);
++++Imgproc.Sobel(img_gray,+img_sobel,+CvType.CV_8U,+1,+0,+3,+1,+0,+Core.BORDER_DEFAULT);
++++//at+src,+Mat+dst,+double+thresh,+double+maxval,+int+type
++++Imgproc.threshold(img_sobel,+img_threshold,+0,+255,+8);
++++element=Imgproc.getStructuringElement(Imgproc.MORPH_RECT,+new+Size(15,5));
++++Imgproc.morphologyEx(img_threshold,+img_threshold,+Imgproc.MORPH_CLOSE,+element);
++++List<MatOfPoint>+contours+=+new+ArrayList<MatOfPoint>();
++++Mat+hierarchy+=+new+Mat();
++++Imgproc.findContours(img_threshold,+contours,hierarchy,+0,+1);

++++List<MatOfPoint>+contours_poly+=+new+ArrayList<MatOfPoint>(contours.size());

+++++for(+int+i+=+0;+i+<+contours.size();+i%2B%2B+){+++++++++++++

+++++++++MatOfPoint2f++mMOP2f1=new+MatOfPoint2f();
+++++++++MatOfPoint2f++mMOP2f2=new+MatOfPoint2f();

+++++++++contours.get(i).convertTo(mMOP2f1,+CvType.CV_32FC2);
+++++++++Imgproc.approxPolyDP(mMOP2f1,+mMOP2f2,+2,+true);+
+++++++++mMOP2f2.convertTo(contours.get(i),+CvType.CV_32S);


++++++++++++Rect+appRect+=+Imgproc.boundingRect(contours.get(i));
++++++++++++if+(appRect.width>appRect.height)+{
++++++++++++++++boundRect.add(appRect);
++++++++++++}
+++++}

++++return+boundRect;
}|code-block|syntax|javascript|156560|并在实践中使用以下代码：|156561|++++++++System.loadLibrary(Core.NATIVE_LIBRARY_NAME);
++++++++Mat+img1=Imgcodecs.imread("abc.png");
++++++++List<Rect>+letterBBoxes1=Utils.detectLetters(img1);

++++++++for(int+i=0;+i<+letterBBoxes1.size();+i%2B%2B)
++++++++++++Imgproc.rectangle(img1,letterBBoxes1.get(i).br(),+letterBBoxes1.get(i).tl(),new+Scalar(0,255,0),3,8,0);+++++++++
++++++++Imgcodecs.imwrite("abc1.png",+img1);|156562|entityMap^0|0|0|0|0^^$0|@$1|2|3|4|5|6|7|M|8|@]|9|@]|A|$]]|$1|B|3|C|5|D|7|N|8|@]|9|@]|A|$E|F]]|$1|G|3|H|5|6|7|O|8|@]|9|@]|A|$]]|$1|I|3|J|5|D|7|P|8|@]|9|@]|A|$E|F]]|$1|K|3|-4|5|6|7|Q|8|@]|9|@]|A|$]]]|L|$]]

Above Code JAVA version: 
Thanks @William

<pre><code>public static List&lt;Rect&gt; detectLetters(Mat img){ 
 List&lt;Rect&gt; boundRect=new ArrayList&lt;&gt;();

 Mat img_gray =new Mat(), img_sobel=new Mat(), img_threshold=new Mat(), element=new Mat();
 Imgproc.cvtColor(img, img_gray, Imgproc.COLOR_RGB2GRAY);
 Imgproc.Sobel(img_gray, img_sobel, CvType.CV_8U, 1, 0, 3, 1, 0, Core.BORDER_DEFAULT);
 //at src, Mat dst, double thresh, double maxval, int type
 Imgproc.threshold(img_sobel, img_threshold, 0, 255, 8);
 element=Imgproc.getStructuringElement(Imgproc.MORPH_RECT, new Size(15,5));
 Imgproc.morphologyEx(img_threshold, img_threshold, Imgproc.MORPH_CLOSE, element);
 List&lt;MatOfPoint&gt; contours = new ArrayList&lt;MatOfPoint&gt;();
 Mat hierarchy = new Mat();
 Imgproc.findContours(img_threshold, contours,hierarchy, 0, 1);

 List&lt;MatOfPoint&gt; contours_poly = new ArrayList&lt;MatOfPoint&gt;(contours.size());

 for( int i = 0; i &lt; contours.size(); i++ ){ 

 MatOfPoint2f mMOP2f1=new MatOfPoint2f();
 MatOfPoint2f mMOP2f2=new MatOfPoint2f();

 contours.get(i).convertTo(mMOP2f1, CvType.CV_32FC2);
 Imgproc.approxPolyDP(mMOP2f1, mMOP2f2, 2, true); 
 mMOP2f2.convertTo(contours.get(i), CvType.CV_32S);


 Rect appRect = Imgproc.boundingRect(contours.get(i));
 if (appRect.width&gt;appRect.height) {
 boundRect.add(appRect);
 }
 }

 return boundRect;
}
</code></pre>

And use this code in practice : 

<pre><code> System.loadLibrary(Core.NATIVE_LIBRARY_NAME);
 Mat img1=Imgcodecs.imread("abc.png");
 List&lt;Rect&gt; letterBBoxes1=Utils.detectLetters(img1);

 for(int i=0; i&lt; letterBBoxes1.size(); i++)
 Imgproc.rectangle(img1,letterBBoxes1.get(i).br(), letterBBoxes1.get(i).tl(),new Scalar(0,255,0),3,8,0); 
 Imgcodecs.imwrite("abc1.png", img1);
</code></pre>

blocks|key|159031|text|@dhanushka解决方案的Python实现：|type|unstyled|depth|inlineStyleRanges|entityRanges|data|159032|def+process_rgb(rgb):
++++hasText+=+False
++++gray+=+cv2.cvtColor(rgb,+cv2.COLOR_BGR2GRAY)
++++morphKernel+=+cv2.getStructuringElement(cv2.MORPH_ELLIPSE,+(3,3))
++++grad+=+cv2.morphologyEx(gray,+cv2.MORPH_GRADIENT,+morphKernel)
++++#+binarize
++++_,+bw+=+cv2.threshold(grad,+0.0,+255.0,+cv2.THRESH_BINARY+%7C+cv2.THRESH_OTSU)
++++#+connect+horizontally+oriented+regions
++++morphKernel+=+cv2.getStructuringElement(cv2.MORPH_RECT,+(9,+1))
++++connected+=+cv2.morphologyEx(bw,+cv2.MORPH_CLOSE,+morphKernel)
++++#+find+contours
++++mask+=+np.zeros(bw.shape[:2],+dtype="uint8")
++++_,contours,+hierarchy+=+cv2.findContours(connected,+cv2.RETR_CCOMP,+cv2.CHAIN_APPROX_SIMPLE)
++++#+filter+contours
++++idx+=+0
++++while+idx+>=+0:
++++++++x,y,w,h+=+cv2.boundingRect(contours[idx])
++++++++#+fill+the+contour
++++++++cv2.drawContours(mask,+contours,+idx,+(255,+255,+255),+cv2.FILLED)
++++++++#+ratio+of+non-zero+pixels+in+the+filled+region
++++++++r+=+cv2.contourArea(contours[idx])/(w*h)
++++++++if(r+>+0.45+and+h+>+5+and+w+>+5+and+w+>+h):
++++++++++++cv2.rectangle(rgb,+(x,y),+(x%2Bw,y%2Bh),+(0,+255,+0),+2)
++++++++++++hasText+=+True
++++++++idx+=+hierarchy[0][idx][0]
++++return+hasText,+rgb|code-block|syntax|javascript|159033|entityMap^0|0|0^^$0|@$1|2|3|4|5|6|7|I|8|@]|9|@]|A|$]]|$1|B|3|C|5|D|7|J|8|@]|9|@]|A|$E|F]]|$1|G|3|-4|5|6|7|K|8|@]|9|@]|A|$]]]|H|$]]

Python Implementation for @dhanushka's solution:

<pre><code>def process_rgb(rgb):
 hasText = False
 gray = cv2.cvtColor(rgb, cv2.COLOR_BGR2GRAY)
 morphKernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3,3))
 grad = cv2.morphologyEx(gray, cv2.MORPH_GRADIENT, morphKernel)
 # binarize
 _, bw = cv2.threshold(grad, 0.0, 255.0, cv2.THRESH_BINARY | cv2.THRESH_OTSU)
 # connect horizontally oriented regions
 morphKernel = cv2.getStructuringElement(cv2.MORPH_RECT, (9, 1))
 connected = cv2.morphologyEx(bw, cv2.MORPH_CLOSE, morphKernel)
 # find contours
 mask = np.zeros(bw.shape[:2], dtype="uint8")
 _,contours, hierarchy = cv2.findContours(connected, cv2.RETR_CCOMP, cv2.CHAIN_APPROX_SIMPLE)
 # filter contours
 idx = 0
 while idx &gt;= 0:
 x,y,w,h = cv2.boundingRect(contours[idx])
 # fill the contour
 cv2.drawContours(mask, contours, idx, (255, 255, 255), cv2.FILLED)
 # ratio of non-zero pixels in the filled region
 r = cv2.contourArea(contours[idx])/(w*h)
 if(r &gt; 0.45 and h &gt; 5 and w &gt; 5 and w &gt; h):
 cv2.rectangle(rgb, (x,y), (x+w,y+h), (0, 255, 0), 2)
 hasText = True
 idx = hierarchy[0][idx][0]
 return hasText, rgb
</code></pre>

blocks|key|2957779|text|这是来自dhanushka的使用OpenCVSharp的answer的C#版本|type|unstyled|depth|inlineStyleRanges|entityRanges|offset|length|data|2957780|++++++++Mat+large+=+new+Mat(INPUT_FILE);
++++++++Mat+rgb+=+new+Mat(),+small+=+new+Mat(),+grad+=+new+Mat(),+bw+=+new+Mat(),+connected+=+new+Mat();

++++++++//+downsample+and+use+it+for+processing
++++++++Cv2.PyrDown(large,+rgb);
++++++++Cv2.CvtColor(rgb,+small,+ColorConversionCodes.BGR2GRAY);

++++++++//+morphological+gradient
++++++++var+morphKernel+=+Cv2.GetStructuringElement(MorphShapes.Ellipse,+new+OpenCvSharp.Size(3,+3));
++++++++Cv2.MorphologyEx(small,+grad,+MorphTypes.Gradient,+morphKernel);

++++++++//+binarize
++++++++Cv2.Threshold(grad,+bw,+0,+255,+ThresholdTypes.Binary+%7C+ThresholdTypes.Otsu);

++++++++//+connect+horizontally+oriented+regions
++++++++morphKernel+=+Cv2.GetStructuringElement(MorphShapes.Rect,+new+OpenCvSharp.Size(9,+1));
++++++++Cv2.MorphologyEx(bw,+connected,+MorphTypes.Close,+morphKernel);

++++++++//+find+contours
++++++++var+mask+=+new+Mat(Mat.Zeros(bw.Size(),+MatType.CV_8UC1),+Range.All);
++++++++Cv2.FindContours(connected,+out+OpenCvSharp.Point[][]+contours,+out+HierarchyIndex[]+hierarchy,+RetrievalModes.CComp,+ContourApproximationModes.ApproxSimple,+new+OpenCvSharp.Point(0,+0));

++++++++//+filter+contours
++++++++var+idx+=+0;
++++++++foreach+(var+hierarchyItem+in+hierarchy)
++++++++{
++++++++++++idx+=+hierarchyItem.Next;
++++++++++++if+(idx+<+0)
++++++++++++++++break;
++++++++++++OpenCvSharp.Rect+rect+=+Cv2.BoundingRect(contours[idx]);
++++++++++++var+maskROI+=+new+Mat(mask,+rect);
++++++++++++maskROI.SetTo(new+Scalar(0,+0,+0));

++++++++++++//+fill+the+contour
++++++++++++Cv2.DrawContours(mask,+contours,+idx,+Scalar.White,+-1);

++++++++++++//+ratio+of+non-zero+pixels+in+the+filled+region
++++++++++++double+r+=+(double)Cv2.CountNonZero(maskROI)+/+(rect.Width+*+rect.Height);
++++++++++++if+(r+>+.45+/*+assume+at+least+45%25+of+the+area+is+filled+if+it+contains+text+*/
+++++++++++++++++&&
++++++++++++(rect.Height+>+8+&&+rect.Width+>+8)+/*+constraints+on+region+size+*/
++++++++++++/*+these+two+conditions+alone+are+not+very+robust.+better+to+use+something+
++++++++++++like+the+number+of+significant+peaks+in+a+horizontal+projection+as+a+third+condition+*/
++++++++++++)
++++++++++++{
++++++++++++++++Cv2.Rectangle(rgb,+rect,+new+Scalar(0,+255,+0),+2);
++++++++++++}
++++++++}

++++++++rgb.SaveImage(Path.Combine(AppDomain.CurrentDomain.BaseDirectory,+"rgb.jpg"));|code-block|syntax|javascript|2957781|entityMap|0|LINK|mutability|MUTABLE|url|https://github.com/shimat/opencvsharp|1|https://stackoverflow.com/a/23672571/2187991^0|G|B|0|S|6|1|0|0^^$0|@$1|2|3|4|5|6|7|S|8|@]|9|@$A|T|B|U|1|V]|$A|W|B|X|1|Y]]|C|$]]|$1|D|3|E|5|F|7|Z|8|@]|9|@]|C|$G|H]]|$1|I|3|-4|5|6|7|10|8|@]|9|@]|C|$]]]|J|$K|$5|L|M|N|C|$O|P]]|Q|$5|L|M|N|C|$O|R]]]]

This is a C# version of the <a href="https://stackoverflow.com/a/23672571/2187991">answer</a> from dhanushka using <a href="https://github.com/shimat/opencvsharp" rel="nofollow noreferrer">OpenCVSharp</a>

<pre><code> Mat large = new Mat(INPUT_FILE);
 Mat rgb = new Mat(), small = new Mat(), grad = new Mat(), bw = new Mat(), connected = new Mat();

 // downsample and use it for processing
 Cv2.PyrDown(large, rgb);
 Cv2.CvtColor(rgb, small, ColorConversionCodes.BGR2GRAY);

 // morphological gradient
 var morphKernel = Cv2.GetStructuringElement(MorphShapes.Ellipse, new OpenCvSharp.Size(3, 3));
 Cv2.MorphologyEx(small, grad, MorphTypes.Gradient, morphKernel);

 // binarize
 Cv2.Threshold(grad, bw, 0, 255, ThresholdTypes.Binary | ThresholdTypes.Otsu);

 // connect horizontally oriented regions
 morphKernel = Cv2.GetStructuringElement(MorphShapes.Rect, new OpenCvSharp.Size(9, 1));
 Cv2.MorphologyEx(bw, connected, MorphTypes.Close, morphKernel);

 // find contours
 var mask = new Mat(Mat.Zeros(bw.Size(), MatType.CV_8UC1), Range.All);
 Cv2.FindContours(connected, out OpenCvSharp.Point[][] contours, out HierarchyIndex[] hierarchy, RetrievalModes.CComp, ContourApproximationModes.ApproxSimple, new OpenCvSharp.Point(0, 0));

 // filter contours
 var idx = 0;
 foreach (var hierarchyItem in hierarchy)
 {
 idx = hierarchyItem.Next;
 if (idx &lt; 0)
 break;
 OpenCvSharp.Rect rect = Cv2.BoundingRect(contours[idx]);
 var maskROI = new Mat(mask, rect);
 maskROI.SetTo(new Scalar(0, 0, 0));

 // fill the contour
 Cv2.DrawContours(mask, contours, idx, Scalar.White, -1);

 // ratio of non-zero pixels in the filled region
 double r = (double)Cv2.CountNonZero(maskROI) / (rect.Width * rect.Height);
 if (r &gt; .45 /* assume at least 45% of the area is filled if it contains text */
 &amp;&amp;
 (rect.Height &gt; 8 &amp;&amp; rect.Width &gt; 8) /* constraints on region size */
 /* these two conditions alone are not very robust. better to use something 
 like the number of significant peaks in a horizontal projection as a third condition */
 )
 {
 Cv2.Rectangle(rgb, rect, new Scalar(0, 255, 0), 2);
 }
 }

 rgb.SaveImage(Path.Combine(AppDomain.CurrentDomain.BaseDirectory, "rgb.jpg"));
</code></pre>

blocks|key|155414|text|这是dhanushka使用EmguCV的answer的VB.NET版本。|type|unstyled|depth|inlineStyleRanges|entityRanges|offset|length|data|155415|与使用OpenCVSharp的C#版本相比，EmguCV中的一些函数和结构需要不同的考虑|155416|Imports+Emgu.CV
Imports+Emgu.CV.Structure
Imports+Emgu.CV.CvEnum
Imports+Emgu.CV.Util

++++++++Dim+input_file+As+String+=+"C:\your_input_image.png"
++++++++Dim+large+As+Mat+=+New+Mat(input_file)
++++++++Dim+rgb+As+New+Mat
++++++++Dim+small+As+New+Mat
++++++++Dim+grad+As+New+Mat
++++++++Dim+bw+As+New+Mat
++++++++Dim+connected+As+New+Mat
++++++++Dim+morphanchor+As+New+Point(0,+0)

++++++++'//downsample+and+use+it+for+processing
++++++++CvInvoke.PyrDown(large,+rgb)
++++++++CvInvoke.CvtColor(rgb,+small,+ColorConversion.Bgr2Gray)

++++++++'//morphological+gradient
++++++++Dim+morphKernel+As+Mat+=+CvInvoke.GetStructuringElement(ElementShape.Ellipse,+New+Size(3,+3),+morphanchor)
++++++++CvInvoke.MorphologyEx(small,+grad,+MorphOp.Gradient,+morphKernel,+New+Point(0,+0),+1,+BorderType.Isolated,+New+MCvScalar(0))

++++++++'//+binarize
++++++++CvInvoke.Threshold(grad,+bw,+0,+255,+ThresholdType.Binary+Or+ThresholdType.Otsu)

++++++++'//+connect+horizontally+oriented+regions
++++++++morphKernel+=+CvInvoke.GetStructuringElement(ElementShape.Rectangle,+New+Size(9,+1),+morphanchor)
++++++++CvInvoke.MorphologyEx(bw,+connected,+MorphOp.Close,+morphKernel,+morphanchor,+1,+BorderType.Isolated,+New+MCvScalar(0))

++++++++'//+find+contours
++++++++Dim+mask+As+Mat+=+Mat.Zeros(bw.Size.Height,+bw.Size.Width,+DepthType.Cv8U,+1)++''+MatType.CV_8UC1
++++++++Dim+contours+As+New+VectorOfVectorOfPoint
++++++++Dim+hierarchy+As+New+Mat

++++++++CvInvoke.FindContours(connected,+contours,+hierarchy,+RetrType.Ccomp,+ChainApproxMethod.ChainApproxSimple,+Nothing)

++++++++'//+filter+contours
++++++++Dim+idx+As+Integer
++++++++Dim+rect+As+Rectangle
++++++++Dim+maskROI+As+Mat
++++++++Dim+r+As+Double
++++++++For+Each+hierarchyItem+In+hierarchy.GetData
++++++++++++rect+=+CvInvoke.BoundingRectangle(contours(idx))
++++++++++++maskROI+=+New+Mat(mask,+rect)
++++++++++++maskROI.SetTo(New+MCvScalar(0,+0,+0))

++++++++++++'//+fill+the+contour
++++++++++++CvInvoke.DrawContours(mask,+contours,+idx,+New+MCvScalar(255),+-1)

++++++++++++'//+ratio+of+non-zero+pixels+in+the+filled+region
++++++++++++r+=+CvInvoke.CountNonZero(maskROI)+/+(rect.Width+*+rect.Height)

++++++++++++'/*+assume+at+least+45%25+of+the+area+Is+filled+if+it+contains+text+*/
++++++++++++'/*+constraints+on+region+size+*/
++++++++++++'/*+these+two+conditions+alone+are+Not+very+robust.+better+to+use+something+
++++++++++++'Like+the+number+of+significant+peaks+in+a+horizontal+projection+as+a+third+condition+*/
++++++++++++If+r+>+0.45+AndAlso+rect.Height+>+8+AndAlso+rect.Width+>+8+Then
++++++++++++++++'draw+green+rectangle
++++++++++++++++CvInvoke.Rectangle(rgb,+rect,+New+MCvScalar(0,+255,+0),+2)
++++++++++++End+If
++++++++++++idx+%2B=+1
++++++++Next
++++++++rgb.Save(IO.Path.Combine(Application.StartupPath,+"rgb.jpg"))|code-block|syntax|javascript|155417|entityMap|0|LINK|mutability|MUTABLE|url|http://www.emgu.com/wiki/index.php/Main_Page|1|https://stackoverflow.com/a/23672571/2187991^0|D|6|0|K|6|1|0|0|0^^$0|@$1|2|3|4|5|6|7|U|8|@]|9|@$A|V|B|W|1|X]|$A|Y|B|Z|1|10]]|C|$]]|$1|D|3|E|5|6|7|11|8|@]|9|@]|C|$]]|$1|F|3|G|5|H|7|12|8|@]|9|@]|C|$I|J]]|$1|K|3|-4|5|6|7|13|8|@]|9|@]|C|$]]]|L|$M|$5|N|O|P|C|$Q|R]]|S|$5|N|O|P|C|$Q|T]]]]

this is a VB.NET version of the <a href="https://stackoverflow.com/a/23672571/2187991">answer</a> from dhanushka using <a href="http://www.emgu.com/wiki/index.php/Main_Page" rel="nofollow noreferrer">EmguCV</a>.

A few functions and structures in EmguCV need different consideration than the C# version with OpenCVSharp

<pre><code>Imports Emgu.CV
Imports Emgu.CV.Structure
Imports Emgu.CV.CvEnum
Imports Emgu.CV.Util

 Dim input_file As String = "C:\your_input_image.png"
 Dim large As Mat = New Mat(input_file)
 Dim rgb As New Mat
 Dim small As New Mat
 Dim grad As New Mat
 Dim bw As New Mat
 Dim connected As New Mat
 Dim morphanchor As New Point(0, 0)

 '//downsample and use it for processing
 CvInvoke.PyrDown(large, rgb)
 CvInvoke.CvtColor(rgb, small, ColorConversion.Bgr2Gray)

 '//morphological gradient
 Dim morphKernel As Mat = CvInvoke.GetStructuringElement(ElementShape.Ellipse, New Size(3, 3), morphanchor)
 CvInvoke.MorphologyEx(small, grad, MorphOp.Gradient, morphKernel, New Point(0, 0), 1, BorderType.Isolated, New MCvScalar(0))

 '// binarize
 CvInvoke.Threshold(grad, bw, 0, 255, ThresholdType.Binary Or ThresholdType.Otsu)

 '// connect horizontally oriented regions
 morphKernel = CvInvoke.GetStructuringElement(ElementShape.Rectangle, New Size(9, 1), morphanchor)
 CvInvoke.MorphologyEx(bw, connected, MorphOp.Close, morphKernel, morphanchor, 1, BorderType.Isolated, New MCvScalar(0))

 '// find contours
 Dim mask As Mat = Mat.Zeros(bw.Size.Height, bw.Size.Width, DepthType.Cv8U, 1) '' MatType.CV_8UC1
 Dim contours As New VectorOfVectorOfPoint
 Dim hierarchy As New Mat

 CvInvoke.FindContours(connected, contours, hierarchy, RetrType.Ccomp, ChainApproxMethod.ChainApproxSimple, Nothing)

 '// filter contours
 Dim idx As Integer
 Dim rect As Rectangle
 Dim maskROI As Mat
 Dim r As Double
 For Each hierarchyItem In hierarchy.GetData
 rect = CvInvoke.BoundingRectangle(contours(idx))
 maskROI = New Mat(mask, rect)
 maskROI.SetTo(New MCvScalar(0, 0, 0))

 '// fill the contour
 CvInvoke.DrawContours(mask, contours, idx, New MCvScalar(255), -1)

 '// ratio of non-zero pixels in the filled region
 r = CvInvoke.CountNonZero(maskROI) / (rect.Width * rect.Height)

 '/* assume at least 45% of the area Is filled if it contains text */
 '/* constraints on region size */
 '/* these two conditions alone are Not very robust. better to use something 
 'Like the number of significant peaks in a horizontal projection as a third condition */
 If r &gt; 0.45 AndAlso rect.Height &gt; 8 AndAlso rect.Width &gt; 8 Then
 'draw green rectangle
 CvInvoke.Rectangle(rgb, rect, New MCvScalar(0, 255, 0), 2)
 End If
 idx += 1
 Next
 rgb.Save(IO.Path.Combine(Application.StartupPath, "rgb.jpg"))
</code></pre>

blocks|key|2745267|text|您可以利用python实现SWTloc。|type|unstyled|depth|inlineStyleRanges|entityRanges|offset|length|data|2745268|完全公开:我是这个库的作者|style|BOLD|2745269|要做到这点：|2745270|第一张和第二张图片|2745271|请注意，这里的text_mode是'lb_df'，它代表浅背景暗前景，即此图像中的文本颜色将比背景颜色更暗|2745272|from+swtloc+import+SWTLocalizer
from+swtloc.utils+import+imgshowN,+imgshow

swtl+=+SWTLocalizer()
#+Stroke+Width+Transform
swtl.swttransform(imgpaths='img1.jpg',+text_mode+=+'lb_df',
++++++++++++++++++save_results=True,+save_rootpath+=+'swtres/',
++++++++++++++++++minrsw+=+3,+maxrsw+=+20,+max_angledev+=+np.pi/3)
imgshow(swtl.swtlabelled_pruned13C)

#+Grouping
respacket=swtl.get_grouped(lookup_radii_multiplier=0.9,+ht_ratio=3.0)
grouped_annot_bubble+=+respacket[2]
maskviz+=+respacket[4]
maskcomb++=+respacket[5]

#+Saving+the+results
_=cv2.imwrite('img1_processed.jpg',+swtl.swtlabelled_pruned13C)
imgshowN([maskcomb,+grouped_annot_bubble],+savepath='grouped_img1.jpg')|code-block|syntax|javascript|2745273|2745274|​|2745275|📷|atomic|2745276|2745277|2745278|2745279|2745280|2745281|2745282|2745283|第三张图片|2745284|请注意，这里的text_mode是'db_lf'，它代表暗背景光前景，即此图像中的文本将采用比背景更亮的颜色|2745285|from+swtloc+import+SWTLocalizer
from+swtloc.utils+import+imgshowN,+imgshow

swtl+=+SWTLocalizer()
#+Stroke+Width+Transform
swtl.swttransform(imgpaths=imgpaths[1],+text_mode+=+'db_lf',
++++++++++++++save_results=True,+save_rootpath+=+'swtres/',
++++++++++++++minrsw+=+3,+maxrsw+=+20,+max_angledev+=+np.pi/3)
imgshow(swtl.swtlabelled_pruned13C)

#+Grouping
respacket=swtl.get_grouped(lookup_radii_multiplier=0.9,+ht_ratio=3.0)
grouped_annot_bubble+=+respacket[2]
maskviz+=+respacket[4]
maskcomb++=+respacket[5]

#+Saving+the+results
_=cv2.imwrite('img1_processed.jpg',+swtl.swtlabelled_pruned13C)
imgshowN([maskcomb,+grouped_annot_bubble],+savepath='grouped_img1.jpg')|2745286|2745287|2745288|2745289|2745290|2745291|您还会注意到，完成的分组不是很准确，要获得所需的结果，因为图像可能会发生变化，请尝试在swtl.get_grouped()函数中调优分组参数。|CODE|2745292|entityMap|0|LINK|mutability|MUTABLE|url|https://github.com/ag-ds-bubble/swtloc|1|IMAGE|IMMUTABLE|imageUrl|https://ask.qcloudimg.com/http-save/yehe-900000/c6944a3e972ba8681395f1362e9eeb2b.jpeg|imageAlt|2|https://ask.qcloudimg.com/http-save/yehe-900000/eb4f8b046922818cd415e287220f9f53.jpeg|3|https://ask.qcloudimg.com/http-save/yehe-900000/3ff640e09e86569f728527727dcd8999.jpeg|4|https://ask.qcloudimg.com/http-save/yehe-900000/2e7a6e41a3c258f6fffc4e079c0211dd.jpeg|5|https://ask.qcloudimg.com/http-save/yehe-900000/749a16ad3cfd069317bc66197f9705c9.jpeg|6|https://ask.qcloudimg.com/http-save/yehe-900000/d2e3fabb53497fc64b369c89c80cda5b.jpeg^0|D|6|0|0|0|A|0|0|0|0|0|0|0|0|1|1|0|0|1|2|0|0|0|0|0|1|3|0|0|1|4|0|0|0|0|0|0|0|0|1|5|0|0|1|6|0|0|17|I|0^^$0|@$1|2|3|4|5|6|7|27|8|@]|9|@$A|28|B|29|1|2A]]|C|$]]|$1|D|3|E|5|6|7|2B|8|@$A|2C|B|2D|F|G]]|9|@]|C|$]]|$1|H|3|I|5|6|7|2E|8|@]|9|@]|C|$]]|$1|J|3|K|5|6|7|2F|8|@]|9|@]|C|$]]|$1|L|3|M|5|6|7|2G|8|@]|9|@]|C|$]]|$1|N|3|O|5|P|7|2H|8|@]|9|@]|C|$Q|R]]|$1|S|3|-4|5|6|7|2I|8|@]|9|@]|C|$]]|$1|T|3|U|5|6|7|2J|8|@]|9|@]|C|$]]|$1|V|3|W|5|X|7|2K|8|@]|9|@$A|2L|B|2M|1|2N]]|C|$]]|$1|Y|3|W|5|X|7|2O|8|@]|9|@$A|2P|B|2Q|1|2R]]|C|$]]|$1|Z|3|U|5|6|7|2S|8|@]|9|@]|C|$]]|$1|10|3|-4|5|6|7|2T|8|@]|9|@]|C|$]]|$1|11|3|U|5|6|7|2U|8|@]|9|@]|C|$]]|$1|12|3|W|5|X|7|2V|8|@]|9|@$A|2W|B|2X|1|2Y]]|C|$]]|$1|13|3|W|5|X|7|2Z|8|@]|9|@$A|30|B|31|1|32]]|C|$]]|$1|14|3|U|5|6|7|33|8|@]|9|@]|C|$]]|$1|15|3|16|5|6|7|34|8|@]|9|@]|C|$]]|$1|17|3|18|5|6|7|35|8|@]|9|@]|C|$]]|$1|19|3|1A|5|P|7|36|8|@]|9|@]|C|$Q|R]]|$1|1B|3|-4|5|6|7|37|8|@]|9|@]|C|$]]|$1|1C|3|U|5|6|7|38|8|@]|9|@]|C|$]]|$1|1D|3|W|5|X|7|39|8|@]|9|@$A|3A|B|3B|1|3C]]|C|$]]|$1|1E|3|W|5|X|7|3D|8|@]|9|@$A|3E|B|3F|1|3G]]|C|$]]|$1|1F|3|U|5|6|7|3H|8|@]|9|@]|C|$]]|$1|1G|3|1H|5|6|7|3I|8|@$A|3J|B|3K|F|1I]]|9|@]|C|$]]|$1|1J|3|-4|5|6|7|3L|8|@]|9|@]|C|$]]]|1K|$1L|$5|1M|1N|1O|C|$1P|1Q]]|1R|$5|1S|1N|1T|C|$1U|1V|1W|-4]]|1X|$5|1S|1N|1T|C|$1U|1Y|1W|-4]]|1Z|$5|1S|1N|1T|C|$1U|20|1W|-4]]|21|$5|1S|1N|1T|C|$1U|22|1W|-4]]|23|$5|1S|1N|1T|C|$1U|24|1W|-4]]|25|$5|1S|1N|1T|C|$1U|26|1W|-4]]]]

You can utilize a python implementation <a href="https://github.com/ag-ds-bubble/swtloc" rel="nofollow noreferrer">SWTloc</a>.
Full Disclosure : I am the author of this library
To do that :-
<h1>First and Second Image</h1>
Notice that the text_mode here is 'lb_df', which stands for Light Background Dark Foreground i.e the text in this image is going to be in darker color than the background
<pre><code>from swtloc import SWTLocalizer
from swtloc.utils import imgshowN, imgshow

swtl = SWTLocalizer()
# Stroke Width Transform
swtl.swttransform(imgpaths='img1.jpg', text_mode = 'lb_df',
 save_results=True, save_rootpath = 'swtres/',
 minrsw = 3, maxrsw = 20, max_angledev = np.pi/3)
imgshow(swtl.swtlabelled_pruned13C)

# Grouping
respacket=swtl.get_grouped(lookup_radii_multiplier=0.9, ht_ratio=3.0)
grouped_annot_bubble = respacket[2]
maskviz = respacket[4]
maskcomb = respacket[5]

# Saving the results
_=cv2.imwrite('img1_processed.jpg', swtl.swtlabelled_pruned13C)
imgshowN([maskcomb, grouped_annot_bubble], savepath='grouped_img1.jpg')
</code></pre>
<hr />
<a href="https://i.stack.imgur.com/nvEV3.jpg" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/nvEV3.jpg" alt="enter image description here" /></a>
<a href="https://i.stack.imgur.com/jT0Mb.jpg" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/jT0Mb.jpg" alt="enter image description here" /></a>
<hr />
<a href="https://i.stack.imgur.com/epKw1.jpg" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/epKw1.jpg" alt="enter image description here" /></a>
<a href="https://i.stack.imgur.com/P4GS4.jpg" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/P4GS4.jpg" alt="enter image description here" /></a>
<h1>Third Image</h1>
Notice that the text_mode here is 'db_lf', which stands for Dark Background Light Foreground i.e the text in this image is going to be in lighter color than the background
<pre><code>from swtloc import SWTLocalizer
from swtloc.utils import imgshowN, imgshow

swtl = SWTLocalizer()
# Stroke Width Transform
swtl.swttransform(imgpaths=imgpaths[1], text_mode = 'db_lf',
 save_results=True, save_rootpath = 'swtres/',
 minrsw = 3, maxrsw = 20, max_angledev = np.pi/3)
imgshow(swtl.swtlabelled_pruned13C)

# Grouping
respacket=swtl.get_grouped(lookup_radii_multiplier=0.9, ht_ratio=3.0)
grouped_annot_bubble = respacket[2]
maskviz = respacket[4]
maskcomb = respacket[5]

# Saving the results
_=cv2.imwrite('img1_processed.jpg', swtl.swtlabelled_pruned13C)
imgshowN([maskcomb, grouped_annot_bubble], savepath='grouped_img1.jpg')
</code></pre>
<hr />
<a href="https://i.stack.imgur.com/J4wIb.jpg" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/J4wIb.jpg" alt="enter image description here" /></a>
<a href="https://i.stack.imgur.com/5Vy2z.jpg" rel="nofollow noreferrer"><img src="https://i.stack.imgur.com/5Vy2z.jpg" alt="enter image description here" /></a>
You will also notice that the grouping done is not so accurate, to get the desired results as the images might vary, try to tune the grouping parameters in <code>swtl.get_grouped()</code> function.

I am trying to find the bounding boxes of text in an image and am currently using this approach:

<pre><code>// calculate the local variances of the grayscale image
Mat t_mean, t_mean_2;
Mat grayF;
outImg_gray.convertTo(grayF, CV_32F);
int winSize = 35;
blur(grayF, t_mean, cv::Size(winSize,winSize));
blur(grayF.mul(grayF), t_mean_2, cv::Size(winSize,winSize));
Mat varMat = t_mean_2 - t_mean.mul(t_mean);
varMat.convertTo(varMat, CV_8U);

// threshold the high variance regions
Mat varMatRegions = varMat &gt; 100;
</code></pre>

When given an image like this:

<img src="https://i.stack.imgur.com/VfDfJ.png" alt="enter image description here">

Then when I show <code>varMatRegions</code> I get this image:

<img src="https://i.stack.imgur.com/64wu4.png" alt="enter image description here">

As you can see it somewhat combines the left block of text with the header of the card, for most cards this method works great but on busier cards it can cause problems. 

The reason it is bad for those contours to connect is that it makes the bounding box of the contour nearly take up the entire card.

Can anyone suggest a different way I can find the text to ensure proper detection of text?

200 points to whoever can find the text in the card above the these two.

<img src="https://i.stack.imgur.com/7h4SJ.jpg" alt="enter image description here">
<img src="https://i.stack.imgur.com/8ipeJ.jpg" alt="enter image description here">

Extracting text OpenCV

Windows

Python

我正在尝试查找图像中的文本边界框，目前正在使用此方法：// calculate the local variances of the grayscale imageMat t_mean, t_mean_2;Mat grayF;outImg_gray.convertTo(grayF, CV_32F);int winSiz...

问提取文本OpenCV
EN

社区

活动

资源

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问提取文本OpenCVEN

社区

活动

资源

关于

腾讯云开发者

热门产品

热门推荐

更多推荐

问提取文本OpenCV
EN