Perceptual Coding of Image Signals

Abstract

High-quality coding of images at low bit rates depends critically on matching the characteristics of the coding algorithm to the properties of the human visual system. We describe two recent examples of perceptually optimized codmg algorithms. For transparent coding of images m a 2-D sub-band coding framework, we use a model in which the quantization threshold for just noticeable distortion is a function of three parameters: center frequency (sub-band number), background intensity and background texture. Use of this model with variable rate DPCM coding of sub-bands leads to transparent coding of color images at rates ranging from 0.2 to 2.0 hits per pixel, depending on input signal. For high-quality coding of sparse images, such as highfrequency sub-bands in 3-D coding of video signals, we use a geometrical vector quantizer whose codevectors are edgerelated shapes rather than traditional designs based on minimum mean squared error. In the coding of teleconference scenes, average bit rates for high-frequency sub-bands are in the range of 0.01 to 0.3 bits per pixel. Combinations of the above principles are expected to provide efficient coder designs for a broad spectrum of applications, including image transmission for ISDN and HDTV.