<title>DCT quantization matrices visually optimized for individual images</title>

Several image compression standards (JPEG, MPEG, H.261) are based on the Discrete Cosine Transform (DCT). These standards do not specify the actual DCT quantization matrix. Ahumada & Peterson and Peterson, Ahumada & Watson provide mathematical formulae to compute a perceptually lossless quantization matrix. Here I show how to compute a matrix that is optimized for a particular image. The method treats each DCT coefficient as an approximation to the local response of a visual `channel.' For a given quantization matrix, the DCT quantization errors are adjusted by contrast sensitivity, light adaptation, and contrast masking, and are pooled non-linearly over the blocks of the image. This yields an 8 X 8 `perceptual error matrix.' A second non-linear pooling over the perceptual error matrix yields total perceptual error. With this model we may estimate the quantization matrix for a particular image that yields minimum bit rate for a given total perceptual error, or minimum perceptual error for a given bit rate. Custom matrices for a number of images show clear improvement over image-independent matrices. Custom matrices are compatible with the JPEG standard, which requires transmission of the quantization matrix.