JPEG/JFIF是万维网(World Wide Web)上最普遍的被用来存储和传输照片的格式。它并不适合于线条绘图(drawing)和其他文字或图标(iconic)的图形,因为它的压缩方法用在这些类型的图形上,得到的结果并不好(PNG和GIF格式通常是用来存储这类的图形;GIF每个像素只有8比特,并不很适合于存储彩色照片,PNG可以无损地存储照片,但是文件太大的缺点让它不太适合在网络上传输)。
我们在网页中浏览大图时,如果图片够大,网速够慢,我们能够很清晰的看到一个现象。图片是由模糊到清晰慢慢呈现的。这个就是ProgressiveJpeg所展示的渐进式加载。如下图所示:
ProgressiveJpeg
网上有很多PS生成的方法,不过这不是最方便的方案,七牛可以对上传的图片进行直接转化。
interlace 是否支持渐进显示。取值1支持渐进显示,取值0不支持渐进显示(默认为0)。适用jpg目标格式,网速慢时,图片显示由模糊到清晰。
示例:
https://odum9helk.qnssl.com/resource/gogopher.jpg?imageMogr2/thumbnail/300x300/interlace/1
目前,在众多的开源图片加载库中,只有Fresco支持了ProgressiveJpeg格式图片的加载。
详见Fresco文档。
自此,对于ProgressiveJpeg,我们已经能上手了。
但Fresco有包体积过大的缺点,我们如果为了支持ProgressiveJpeg就受到了Fresco的其他限制。所以,我们从原理上了解一下ProgressiveJpeg格式,尝试写出一个轻量的库。
Jpeg
ProgressiveJpeg的编码格式非常复杂,但使用渐进式加载,我们并不需要破解它所有的奥秘。因为不论支不支持渐进式加载,一般的解码器(如Android中的BitmapFactory)一定能够解码出最终完整的Jpeg图片。
那么,为什么它们无法支持渐进式呢。原来一般的解码器解码图片文件时会把整个文件读完再解码,ProgressiveJpeg的图片中,包含了多Scan(包含了一张图片压缩信息)。因此,ProgressiveJpeg中的一部分数据便足以解码出一张完整的、相对模糊的图片。
了解到这里,我们便能够很容易地想到,其实渐进式加载的奥秘,其实就是在ProgressiveJpeg的数据流中找到合适的点。当我们读到这个点时,这个点之前的数据便可以被解析出一张图片。同时,我们继续读取数据流,找到下一个可解析点,就可以解析出一张更清晰的图片。
我们先来看看解析(寻找某个Scan)的过程。 首先,我创建了一个OutputStream将读到的数据写入其中,方便随时在读到合适的位置时,用它生成一个byte[]渲染成图片。
这个合适的位置通过上面的图表,其实是EOI或SOS时。
当寻找到这个点时,我们调用newScanOrImageEndFound();
将数据进行包装传到外部。
我们先来看看寻找EOI或SOS的代码
/**
* Progressively scans jpeg data and instructs caller when enough data is available to decode
* a partial image.
* <p/>
* <p> This class treats any sequence of bytes starting with 0xFFD8 as a valid jpeg image
* <p/>
* <p> Users should call parseMoreData method each time new chunk of data is received. The buffer
* passed as a parameter should include entire image data received so far.
*/
public class ProgressiveJpegParser {
/**
* Initial state of the parser. Next byte read by the parser should be 0xFF.
*/
private static final int READ_FIRST_JPEG_BYTE = 0;
/**
* Parser saw only one byte so far (0xFF). Next byte should be second byte of SOI marker
*/
private static final int READ_SECOND_JPEG_BYTE = 1;
/**
* Next byte is either entropy coded data or first byte of a marker. First byte of marker
* cannot appear in entropy coded data, unless it is followed by 0x00 escape byte.
*/
private static final int READ_MARKER_FIRST_BYTE_OR_ENTROPY_DATA = 2;
/**
* Last read byte is 0xFF, possible start of marker (possible, because next byte might be
* "escape byte" or 0xFF again)
*/
private static final int READ_MARKER_SECOND_BYTE = 3;
/**
* Last two bytes constitute a marker that indicates start of a segment, the following two bytes
* denote 16bit size of the segment
*/
private static final int READ_SIZE_FIRST_BYTE = 4;
/**
* Last three bytes are marker and first byte of segment size, after reading next byte, bytes
* constituting remaining part of segment will be skipped
*/
private static final int READ_SIZE_SECOND_BYTE = 5;
/**
* Parsed data is not a JPEG file
*/
private static final int NOT_A_JPEG = 6;
private static final int DIRECTLY_END = 7;
/**
* The buffer size in bytes to use.
*/
private static final int BUFFER_SIZE = 32 * 1024;
private int mBufferSize = BUFFER_SIZE;
private int mParserState;
private int mLastByteRead;
public interface OnImageDataListener {
void onImageDataReady(byte[] datas);
}
private OnImageDataListener mOnImageDataListener;
public ProgressiveJpegParser() {
mLastByteRead = 0;
mParserState = READ_FIRST_JPEG_BYTE;
}
public void setOnImageDataListener(OnImageDataListener listener) {
mOnImageDataListener = listener;
}
private ByteArrayOutputStream mBaos ;
private void writeToBaos(ByteArrayOutputStream outputStream, int nextByte) {
outputStream.write(nextByte);
}
private void writeToBaos(InputStream inputStream, ByteArrayOutputStream outputStream, int length)
throws IOException {
byte[] buffer;
int readNum = 0;
while (length > mBufferSize) {
buffer = new byte[mBufferSize];
int perReadNum = 0;
while (perReadNum < mBufferSize) {
perReadNum += inputStream.read(buffer, 0, mBufferSize - perReadNum);
readNum += perReadNum;
}
}
buffer = new byte[length - readNum];
while (readNum < length) {
readNum += inputStream.read(buffer, 0, length - readNum);
}
outputStream.write(buffer);
}
private boolean writeToBaos(InputStream inputStream, ByteArrayOutputStream outputStream)
throws IOException {
final byte[] bytes = new byte[mBufferSize];
int count;
while ((count = inputStream.read(bytes, 0, mBufferSize)) != -1) {
outputStream.write(bytes, 0, count);
}
return true;
}
/**
* Parses more data from inputStream.
*
* @param inputStream instance of buffered pooled byte buffer input stream
*/
public boolean doParseMoreData(final InputStream inputStream, ByteArrayOutputStream outputStream) {
mBaos = outputStream;
try {
int nextByte;
while ((nextByte = inputStream.read()) != -1) {
writeToBaos(outputStream, nextByte);
switch (mParserState) {
case READ_FIRST_JPEG_BYTE:
if (nextByte == JfifUtil.MARKER_FIRST_BYTE) {
mParserState = READ_SECOND_JPEG_BYTE;
} else {
mParserState = NOT_A_JPEG;
}
break;
case READ_SECOND_JPEG_BYTE:
if (nextByte == JfifUtil.MARKER_SOI) {
mParserState = READ_MARKER_FIRST_BYTE_OR_ENTROPY_DATA;
} else {
mParserState = NOT_A_JPEG;
}
break;
case READ_MARKER_FIRST_BYTE_OR_ENTROPY_DATA:
if (nextByte == JfifUtil.MARKER_FIRST_BYTE) {
mParserState = READ_MARKER_SECOND_BYTE;
}
break;
case READ_MARKER_SECOND_BYTE:
if (nextByte == JfifUtil.MARKER_FIRST_BYTE) {
mParserState = READ_MARKER_SECOND_BYTE;
} else if (nextByte == JfifUtil.MARKER_ESCAPE_BYTE) {
mParserState = READ_MARKER_FIRST_BYTE_OR_ENTROPY_DATA;
} else {
if (nextByte == JfifUtil.MARKER_SOS || nextByte == JfifUtil.MARKER_EOI) {
newScanOrImageEndFound();
}
if (doesMarkerStartSegment(nextByte)) {
mParserState = READ_SIZE_FIRST_BYTE;
} else {
mParserState = READ_MARKER_FIRST_BYTE_OR_ENTROPY_DATA;
}
}
break;
case READ_SIZE_FIRST_BYTE:
mParserState = READ_SIZE_SECOND_BYTE;
break;
case READ_SIZE_SECOND_BYTE:
final int size = (mLastByteRead << 8) + nextByte;
// We need to jump after the end of the segment - skip size-2 next bytes.
// We might want to skip more data than is available to read, in which case we will
// consume entire data in inputStream and exit this function before entering another
// iteration of the loop.
final int bytesToSkip = size - 2;
// StreamUtil.skip(inputStream, bytesToSkip);
// Todo by lsy: Save the skip data in Buffer
writeToBaos(inputStream, outputStream, bytesToSkip);
mParserState = READ_MARKER_FIRST_BYTE_OR_ENTROPY_DATA;
break;
case NOT_A_JPEG:
writeToBaos(inputStream, outputStream);
break;
case DIRECTLY_END:
writeToBaos(inputStream, outputStream);
break;
default:
break;
}
mLastByteRead = nextByte;
}
} catch (IOException ioe) {
// does not happen, input stream returned by pooled byte buffer does not throw IOExceptions
}
return true;
}
/**
* Not every marker is followed by associated segment
*/
private static boolean doesMarkerStartSegment(int markerSecondByte) {
if (markerSecondByte == JfifUtil.MARKER_TEM) {
return false;
}
if (markerSecondByte >= JfifUtil.MARKER_RST0 && markerSecondByte <= JfifUtil.MARKER_RST7) {
return false;
}
return markerSecondByte != JfifUtil.MARKER_EOI && markerSecondByte != JfifUtil.MARKER_SOI;
}
/**
* Util for obtaining information from JPEG file.
*/
public class JfifUtil {
/**
* Definitions of jpeg markers as well as overall description of jpeg file format can be found
* here: <a href="http://www.w3.org/Graphics/JPEG/itu-t81.pdf">Recommendation T.81</a>
*/
public static final int MARKER_FIRST_BYTE = 0xFF;
public static final int MARKER_ESCAPE_BYTE = 0x00;
public static final int MARKER_SOI = 0xD8;
public static final int MARKER_TEM = 0x01;
public static final int MARKER_EOI = 0xD9;
public static final int MARKER_SOS = 0xDA;
public static final int MARKER_APP1 = 0xE1;
public static final int MARKER_SOFn = 0xC0;
public static final int MARKER_RST0 = 0xD0;
public static final int MARKER_RST7 = 0xD7;
public static final int APP1_EXIF_MAGIC = 0x45786966;
private JfifUtil() {
}
}
接近三百行的代码,比较难以阅读。但对照上面的格式说明,细心读一读会发现,我们就是在寻找上面所说的格式,然后在找到格式后,调用newScanOrImageEndFound();
。
上面代码中,我们将读到的所有字节都写入了mBaos中。所以,在newScanOrImageEndFound();
中我们将mBaos的数据拿出来做处理。
private void newScanOrImageEndFound() throws IOException {
if (mOnImageDataListener != null) {
byte[] bytes = mBaos.toByteArray();
byte[] tailBytes = new byte[] {(byte) JfifUtil.MARKER_FIRST_BYTE, (byte) JfifUtil.MARKER_EOI};
byte[] finalBytes = new byte[bytes.length ];
System.arraycopy(bytes , 0 , finalBytes ,0, bytes.length-2);
System.arraycopy(tailBytes , 0 , finalBytes ,bytes.length-2, tailBytes.length);
mOnImageDataListener.onImageDataReady(finalBytes);
}
}
包装数据也非常简单,由于我们发现的数据是以SOS或EOI结尾的,但我们要欺骗BitmapFactory现在给它的就是完整的数据。所以我们将SOS或EOI的结尾一律替换为EOI的结尾。这类似于告诉BitmapFactory当前的byte[]已经是一张完整的图片啦。
最后,我们在外面调用这个方法。
mDecodeThread = new Thread(new Runnable() {
@Override
public void run() {
ProgressiveJpegParser parser = new ProgressiveJpegParser();
parser.setOnImageDataListener(
new ProgressiveJpegParser.OnImageDataListener() {
@Override
public void onImageDataReady(byte[] datas) {
mBitmap = BitmapFactory.decodeByteArray(datas, 0, datas.length);
try {
Thread.sleep(1000);
} catch (InterruptedException e) {
e.printStackTrace();
}
mUiHandler.sendEmptyMessage(BITMAP_READY);
}
}
);
AssetManager assetManager = getAssets();
InputStream sourceInput = null;
try {
sourceInput = assetManager.open("jpeg_test.jpg");
} catch (IOException e) {
e.printStackTrace();
}
ByteArrayOutputStream outputStream = new ByteArrayOutputStream();
parser.doParseMoreData(sourceInput, outputStream);
}
});
将准备好的byte[]送给BitmapFactory去解析,得到Bitmap就可以显示了。
通过这种方法,我们就可以在Android设备上也展现出渐进式加载的效果。是不是很cooool。
但是,这个方法因为会不断地产生byte[]其实非常吃内存。在实际使用中,我们可以考虑限制渐进图片的粒度。比如我只显示前两张,就不再寻找过渡图了,这些优化就不在此赘述。
以上。