jpeg decoder: quality regression in chroma upsampling

[pengvado]

Recent versions of imagemagick produce lower-quality decoding of subsampled jpegs than older versions did. I noticed the problem in 6.8.4-6 2013-04-03 Q16, and tracked the regression to svn-r7275. That commit message ("Change IsMagickTrue() to more sensible name IsStringTrue()") implies it was only supposed to be a refactoring, but it in fact changed the default upsampling algorithm.

Patch to revert to the previous behavior, i.e. keeping libjpeg defaults:

Code: Select all

--- a/coders/jpeg.c
+++ b/coders/jpeg.c
@@ -1126,7 +1126,8 @@ static Image *ReadJPEGImage(const ImageInfo *image_info,
       jpeg_info.desired_number_of_colors=(int) StringToUnsignedLong(option);
     }
   option=GetImageOption(image_info,"jpeg:block-smoothing");
-  jpeg_info.do_block_smoothing=IsStringTrue(option);
+  if (option != (const char *) NULL)
+    jpeg_info.do_block_smoothing=IsStringTrue(option);
   jpeg_info.dct_method=JDCT_FLOAT;
   option=GetImageOption(image_info,"jpeg:dct-method");
   if (option != (const char *) NULL)
@@ -1159,7 +1160,8 @@ static Image *ReadJPEGImage(const ImageInfo *image_info,
       }
     }
   option=GetImageOption(image_info,"jpeg:fancy-upsampling");
-  jpeg_info.do_fancy_upsampling=IsStringTrue(option);
+  if (option != (const char *) NULL)
+    jpeg_info.do_fancy_upsampling=IsStringTrue(option);
   (void) jpeg_start_decompress(&jpeg_info);
   image->columns=jpeg_info.output_width;
   image->rows=jpeg_info.output_height;

[magick]

Thanks for the patch, we'll get it into the Subversion trunk by sometime tomorrow.

[anthony]

Shouldn't that be GetImageArtifact() instead of GetImageOption() ????