Java 解析Tiff深入研究
最近在读取客户发过来的tiff文件是，底层竟然报错了，错误：bandOffsets.length is wrong! 没办法，因为错误消息出现在tiff的read中，因此就对
底层序中tiff读取的代码进行了研究。
之前有一篇文章，我简单的介绍了Geotools读取Tiff的代码，Java 通过geotools读取tiff，其实通过深入研究发现，原来幕后的大佬竟然是imageio-ext中的TiffImageReader，
imageio做为Java开发的人员肯定都知道，而ImageIO-ext是imageio的扩展类，我们可以到github上看到它的源码，这是一个非常强大的库，对于Java处理各种栅格数据的读写非常有帮助！
借助这篇文章，我们需要先了解Tiff文件的具体结构，可以参考这篇文章，TIFF文件结构详解 https://blog.csdn.net/oYinHeZhiGuang/article/details/121710467 讲的很好！
下面我们来看下imageio-ext中的tiff读取代码，主要类TiffImageReader，我们来看下Java程序是如何读取tiff文件的。
构造方法：
public TIFFImageReader(ImageReaderSpi originatingProvider) { super(originatingProvider); }
这个类需要通过一个ImageReaderSpi来实例化，其实这种SPI的设计模式，Java的很多开源项目都在用到，这里我们通过TIFFImageReaderSpi这个类即可。
其次设置文件的路径，以及其它一些参数，通过该类的如下方法：
public void setInput(Object input, boolean seekForwardOnly, boolean ignoreMetadata)
这个方法，里面有input就是需要读取的文件，seekForwardOnly设置为true表示：只能从这个输入源按升序读取图像和元数据。ignoreMetadata设置为true表示读取忽略元数据
接下来就是对tiff元数据的读取，具体参见getImageMetadata(int imageIndex)这个方法：
public IIOMetadata getImageMetadata(int imageIndex) throws IIOException { seekToImage(imageIndex, true); TIFFImageMetadata im = new TIFFImageMetadata(imageMetadata.getRootIFD().getTagSetList()); Node root = imageMetadata.getAsTree(TIFFImageMetadata.nativeMetadataFormatName); im.setFromTree(TIFFImageMetadata.nativeMetadataFormatName, root); if (noData != null) { im.setNoData(new double[] {noData, noData}); } if (scales != null && offsets != null) { im.setScales(scales); im.setOffsets(offsets); } return im; }
其中的seekToImage(imageIndex, true)为最主要的逻辑处理，这个方法中，第一个参数，imageIndex为tiff多页中的第几个，第二参数设置标示该tiff页是否已经被解析过
private void seekToImage(int imageIndex, boolean optimized) throws IIOException { checkIndex(imageIndex); // TODO we should do this initialization just once!!! int index = locateImage(imageIndex); if (index != imageIndex) { throw new IndexOutOfBoundsException("imageIndex out of bounds!"); } final Integer i= Integer.valueOf(index); //optimized branch if(!optimized){ readMetadata(); initializeFromMetadata(); return; } // in case we have cache the info for this page if(pagesInfo.containsKey(i)){ // initialize from cachedinfo only if needed // TODO Improve if(imageMetadata == null || !initialized) {// this means the curindex has changed final PageInfo info = pagesInfo.get(i); final TIFFImageMetadata metadata = info.imageMetadata.get(); if (metadata != null) { initializeFromCachedInfo(info, metadata); return; } pagesInfo.put(i,null); } } readMetadata(); initializeFromMetadata(); }
这个方法当中，第一次加载tiff，通过readMetadata()和initializeFromMetadata()将tiff的元信息缓存起来，方便后面再次读取。
读取过程
主要是要结合Tiff的格式进行理解，大体主要是解析tiff头，然后获取到IFD(tiff的图像目录信息)，然后再依次去解析每个目录的具体内容，代码就不再这里罗列了。
这里主要说下，解析目录信息是获取tiff的元信息的过程，通常是解析每个tag的信息，解析代码TIFFIFD类的initialize(ImageInputStream stream, boolean ignoreUnknownFields, final boolean isBTIFF)方法中
public void initialize(ImageInputStream stream, boolean ignoreUnknownFields, final boolean isBTIFF) throws IOException { removeTIFFFields(); List tagSetList = getTagSetList(); final long numEntries; if(isBTIFF) numEntries= stream.readLong(); else numEntries= stream.readUnsignedShort(); for (int i = 0; i < numEntries; i++) { // Read tag number, value type, and value count. int tag = stream.readUnsignedShort(); int type = stream.readUnsignedShort(); int count; if(isBTIFF) { long count_=stream.readLong(); count = (int)count_; if(count!=count_) throw new IllegalArgumentException("unable to use long number of values"); } else count = (int)stream.readUnsignedInt(); // Get the associated TIFFTag. TIFFTag tiffTag = getTag(tag, tagSetList); // Ignore unknown fields. if(ignoreUnknownFields && tiffTag == null) { // Skip the value/offset so as to leave the stream // position at the start of the next IFD entry. if(isBTIFF) stream.skipBytes(8); else stream.skipBytes(4); // XXX Warning message ... // Continue with the next IFD entry. continue; } long nextTagOffset; if(isBTIFF){ nextTagOffset = stream.getStreamPosition() + 8; int sizeOfType = TIFFTag.getSizeOfType(type); if (count*sizeOfType > 8) { long value = stream.readLong(); stream.seek(value); } } else{ nextTagOffset = stream.getStreamPosition() + 4; int sizeOfType = TIFFTag.getSizeOfType(type); if (count*sizeOfType > 4) { long value = stream.readUnsignedInt(); stream.seek(value); } } if (tag == BaselineTIFFTagSet.TAG_STRIP_BYTE_COUNTS || tag == BaselineTIFFTagSet.TAG_TILE_BYTE_COUNTS || tag == BaselineTIFFTagSet.TAG_JPEG_INTERCHANGE_FORMAT_LENGTH) { this.stripOrTileByteCountsPosition = stream.getStreamPosition(); if (LAZY_LOADING) { type = type == TIFFTag.TIFF_LONG ? TIFFTag.TIFF_LAZY_LONG : TIFFTag.TIFF_LAZY_LONG8; } } else if (tag == BaselineTIFFTagSet.TAG_STRIP_OFFSETS || tag == BaselineTIFFTagSet.TAG_TILE_OFFSETS || tag == BaselineTIFFTagSet.TAG_JPEG_INTERCHANGE_FORMAT) { this.stripOrTileOffsetsPosition = stream.getStreamPosition(); if (LAZY_LOADING) { type = type == TIFFTag.TIFF_LONG ? TIFFTag.TIFF_LAZY_LONG : TIFFTag.TIFF_LAZY_LONG8; } } Object obj = null; try { switch (type) { case TIFFTag.TIFF_BYTE: case TIFFTag.TIFF_SBYTE: case TIFFTag.TIFF_UNDEFINED: case TIFFTag.TIFF_ASCII: byte[] bvalues = new byte[count]; stream.readFully(bvalues, 0, count); if (type == TIFFTag.TIFF_ASCII) { // Can be multiple strings final List<String> v = new ArrayList<String>(); boolean inString = false; int prevIndex = 0; for (int index = 0; index <= count; index++) { if (index < count && bvalues[index] != 0) { if (!inString) { // start of string prevIndex = index; inString = true; } } else { // null or special case at end of string if (inString) { // end of string final String s = new String(bvalues, prevIndex,index - prevIndex); v.add(s); inString = false; } } } count = v.size(); String[] strings; if(count != 0) { strings = new String[count]; for (int c = 0 ; c < count; c++) { strings[c] = v.get(c); } } else { // This case has been observed when the value of // 'count' recorded in the field is non-zero but // the value portion contains all nulls. count = 1; strings = new String[] {""}; } obj = strings; } else { obj = bvalues; } break; case TIFFTag.TIFF_SHORT: char[] cvalues = new char[count]; for (int j = 0; j < count; j++) { cvalues[j] = (char)(stream.readUnsignedShort()); } obj = cvalues; break; case TIFFTag.TIFF_LONG: case TIFFTag.TIFF_IFD_POINTER: long[] lvalues = new long[count]; for (int j = 0; j < count; j++) { lvalues[j] = stream.readUnsignedInt(); } obj = lvalues; break; case TIFFTag.TIFF_RATIONAL: long[][] llvalues = new long[count][2]; for (int j = 0; j < count; j++) { llvalues[j][0] = stream.readUnsignedInt(); llvalues[j][1] = stream.readUnsignedInt(); } obj = llvalues; break; case TIFFTag.TIFF_SSHORT: short[] svalues = new short[count]; for (int j = 0; j < count; j++) { svalues[j] = stream.readShort(); } obj = svalues; break; case TIFFTag.TIFF_SLONG: int[] ivalues = new int[count]; for (int j = 0; j < count; j++) { ivalues[j] = stream.readInt(); } obj = ivalues; break; case TIFFTag.TIFF_SRATIONAL: int[][] iivalues = new int[count][2]; for (int j = 0; j < count; j++) { iivalues[j][0] = stream.readInt(); iivalues[j][1] = stream.readInt(); } obj = iivalues; break; case TIFFTag.TIFF_FLOAT: float[] fvalues = new float[count]; for (int j = 0; j < count; j++) { fvalues[j] = stream.readFloat(); } obj = fvalues; break; case TIFFTag.TIFF_DOUBLE: double[] dvalues = new double[count]; for (int j = 0; j < count; j++) { dvalues[j] = stream.readDouble(); } obj = dvalues; break; case TIFFTag.TIFF_LONG8: case TIFFTag.TIFF_SLONG8: case TIFFTag.TIFF_IFD8: long[] lBvalues = new long[count]; for (int j = 0; j < count; j++) { lBvalues[j] = stream.readLong(); } obj = lBvalues; break; case TIFFTag.TIFF_LAZY_LONG8: case TIFFTag.TIFF_LAZY_LONG: obj = new TIFFLazyData(stream, type, count); break; default: // XXX Warning break; } } catch(EOFException eofe) { // The TIFF 6.0 fields have tag numbers less than or equal // to 532 (ReferenceBlackWhite) or equal to 33432 (Copyright). // If there is an error reading a baseline tag, then re-throw // the exception and fail; otherwise continue with the next // field. if(BaselineTIFFTagSet.getInstance().getTag(tag) == null) { throw eofe; } } if (tiffTag == null) { // XXX Warning: unknown tag } else if (!tiffTag.isDataTypeOK(type)) { // XXX Warning: bad data type } else if (tiffTag.isIFDPointer() && obj != null) { stream.mark(); stream.seek(((long[])obj)[0]); List tagSets = new ArrayList(1); tagSets.add(tiffTag.getTagSet()); TIFFIFD subIFD = new TIFFIFD(tagSets); // XXX Use same ignore policy for sub-IFD fields? subIFD.initialize(stream, ignoreUnknownFields); obj = subIFD; stream.reset(); } if (tiffTag == null) { tiffTag = new TIFFTag(null, tag, 1 << type, null); } // Add the field if its contents have been initialized which // will not be the case if an EOF was ignored above. if(obj != null) { TIFFField f = new TIFFField(tiffTag, type, count, obj); addTIFFField(f); } stream.seek(nextTagOffset); } this.lastPosition = stream.getStreamPosition(); }
Tiff常用的Tag标签类有BaseLineTiffTagSet、FaxTiffTagSet、GeoTiffTagSet、EXIFPTiffTagSet、PrivateTIFFTagSet等。
其中的GeoTiffTagSet用于geotiff的额外存储信息，在这里说明下，Geotiff是Tiff格式对Gis数据的一种存储支持，而PrivateTIFFTagSet是对gdal的支持，增加了NODATA、MEATADATA的信息。
对于文章开头提的关于bandOffsets.length is wrong!，主要原因出现在getImageTypes(int imageIndex)这个方法的下面这个实现中。
ImageTypeSpecifier itsRaw =
TIFFDecompressor.getRawImageTypeSpecifier
(photometricInterpretation,
compression,
samplesPerPixel,
bitsPerSample,
sampleFormat,
extraSamples,
colorMap);
最终我们在ImageTypeSpecifier这个类的Interleaved(ColorSpace colorSpace,int[] bandOffsets,int dataType,boolean hasAlpha,boolean isAlphaPremultiplied) 方法中发现问题。
public Interleaved(ColorSpace colorSpace, int[] bandOffsets, int dataType, boolean hasAlpha, boolean isAlphaPremultiplied) { if (colorSpace == null) { throw new IllegalArgumentException("colorSpace == null!"); } if (bandOffsets == null) { throw new IllegalArgumentException("bandOffsets == null!"); } int numBands = colorSpace.getNumComponents() + (hasAlpha ? 1 : 0); if (bandOffsets.length != numBands) { throw new IllegalArgumentException ("bandOffsets.length is wrong!"); }
我们发现只有当我们的图像偏移数量和我们的通道数不一致的时候，就会报这个错误！
总结
通过研究这个问题，基本上梳理了Java基于ImageIO-ext读取tiff的过程，基本跟tiff的数据结构对应起来。
