A perceptual model for JPEG applications based on block classification, texture masking, and luminance masking

A perceptual model based on the texture masking and luminance masking properties of the human visual system is presented in this paper. The model computes a local multiplier map for scaling of the JPEG quantization matrix. The result is that fewer bits are used to represent the perceptually less important areas of the image. The texture masking model is based on a block classification algorithm to differentiate between the plain, edge, and texture blocks. An adaptive luminance masking scheme is used to adjust the luminance masking strategy depending on the image's mean luminance value. An adaptive JPEG coder based on the perceptual model is implemented. Experimental results show that the adaptive coder provides savings in bit-rate over baseline JPEG, with no overall loss in perceptual quality according to a subjective test.