Matrix Quantization of Homomorphically Processed Images

This paper discusses the design of a tree-search matrix quantizer to be used to encode homomorphically processed images. An image obtained by optical means is modeled as a product of the illumtination and reflectance components. The illumination part is lowpass and the reflectance part which reveals the edge and texture details, is highpass. The homomorphic system as applied to image processing consists of a logarithmic point transformation followed by linear filtering. Homomorphic processing achieves a simultaneous dynamic range reduction and contrast enhancenment. A matrix quantizer operating on such processed images is shown to achieve bit rates as low as 0.16 bpp with acceptable quality.