Add a new parameter to JpegImage.getData to indicate the source of the image data (issue 9513)

The purpose of this patch is to provide a better default behaviour when `JpegImage` is used to parse standalone JPEG images with CMYK colour spaces.
Since the issue that the patch concerns is somewhat of a special-case, the implementation utilizes the already existing decode support in an attempt to minimize the impact w.r.t. code size.

*Please note:* It's always possible for the user of `JpegImage` to control image inversion, and thus override the new behaviour, by simply passing a custom `decodeTransform` array upon initialization.

Author: Snuffleupagus

examples/image_decoders/jpeg_viewer.js

@@ -58,7 +58,11 @@ var jpegImage = new pdfjsImageDecoders.JpegImage();
 jpegImage.parse(typedArrayImage);
 
 var width = jpegImage.width, height = jpegImage.height;
-var jpegData = jpegImage.getData(width, height, /* forceRGB = */ true);
+var jpegData = jpegImage.getData({
+  width,
+  height,
+  forceRGB: true,
+});
 
 // Render the JPEG image on a <canvas>.
 //

src/core/jpeg_stream.js

@@ -97,8 +97,12 @@ let JpegStream = (function JpegStreamClosure() {
     const jpegImage = new JpegImage(jpegOptions);
 
     jpegImage.parse(this.bytes);
-    let data = jpegImage.getData(this.drawWidth, this.drawHeight,
-                                 this.forceRGB);
+    let data = jpegImage.getData({
+      width: this.drawWidth,
+      height: this.drawHeight,
+      forceRGB: this.forceRGB,
+      isSourcePDF: true,
+    });
     this.buffer = data;
     this.bufferLength = data.length;
     this.eof = true;

src/core/jpg.js

@@ -975,7 +975,7 @@ var JpegImage = (function JpegImageClosure() {
       this.numComponents = this.components.length;
     },
 
-    _getLinearizedBlockData: function getLinearizedBlockData(width, height) {
+    _getLinearizedBlockData(width, height, isSourcePDF = false) {
       var scaleX = this.width / width, scaleY = this.height / height;
 
       var component, componentScaleX, componentScaleY, blocksPerScanline;
@@ -1013,7 +1013,24 @@ var JpegImage = (function JpegImageClosure() {
       }
 
       // decodeTransform contains pairs of multiplier (-256..256) and additive
-      const transform = this._decodeTransform;
+      let transform = this._decodeTransform;
+
+      // In PDF files, JPEG images with CMYK colour spaces are usually inverted
+      // (this can be observed by extracting the raw image data).
+      // Since the conversion algorithms (see below) were written primarily for
+      // the PDF use-cases, attempting to use `JpegImage` to parse standalone
+      // JPEG (CMYK) images may thus result in inverted images (see issue 9513).
+      //
+      // Unfortunately it's not (always) possible to tell, from the image data
+      // alone, if it needs to be inverted. Thus in an attempt to provide better
+      // out-of-box behaviour when `JpegImage` is used standalone, default to
+      // inverting JPEG (CMYK) images if and only if the image data does *not*
+      // come from a PDF file and no `decodeTransform` was passed by the user.
+      if (!transform && numComponents === 4 && !isSourcePDF) {
+        transform = new Int32Array([
+          -256, 255, -256, 255, -256, 255, -256, 255]);
+      }
+
       if (transform) {
         for (i = 0; i < dataLength;) {
           for (j = 0, k = 0; j < numComponents; j++, i++, k += 2) {
@@ -1162,14 +1179,14 @@ var JpegImage = (function JpegImageClosure() {
       return data.subarray(0, offset);
     },
 
-    getData: function getData(width, height, forceRGBoutput) {
+    getData({ width, height, forceRGB = false, isSourcePDF = false, }) {
       if (this.numComponents > 4) {
         throw new JpegError('Unsupported color mode');
       }
-      // type of data: Uint8Array(width * height * numComponents)
-      var data = this._getLinearizedBlockData(width, height);
+      // Type of data: Uint8ClampedArray(width * height * numComponents)
+      var data = this._getLinearizedBlockData(width, height, isSourcePDF);
 
-      if (this.numComponents === 1 && forceRGBoutput) {
+      if (this.numComponents === 1 && forceRGB) {
         var dataLength = data.length;
         var rgbData = new Uint8ClampedArray(dataLength * 3);
         var offset = 0;
@@ -1184,11 +1201,11 @@ var JpegImage = (function JpegImageClosure() {
         return this._convertYccToRgb(data);
       } else if (this.numComponents === 4) {
         if (this._isColorConversionNeeded) {
-          if (forceRGBoutput) {
+          if (forceRGB) {
             return this._convertYcckToRgb(data);
           }
           return this._convertYcckToCmyk(data);
-        } else if (forceRGBoutput) {
+        } else if (forceRGB) {
           return this._convertCmykToRgb(data);
         }
       }