Image Processing Reference
Figure 12-3 Edge-detected and gridded layer.
Compression Efficiency for Shaped Coding
Pixels that will not be visible because they fall outside the object's boundary or outside the
shape-coded boundary are processed to set them to the average value for the whole mac-
roblock. They will be replaced by the DC coefficient value for that macroblock. Setting
them to black or white regardless of the average value of the macroblock would require a
lot of additional energy in the DCT coefficients to transform the pixel value from the aver-
age for that block to the default value we've specified. Figure 12-4 shows the breakdown
into three kinds of alpha-coded block.