Quantization Issues of Wavelet Transform Implementation and optimization with Genetic Algorithm

In this paper, we are motivated by the quantization of filter coefficients in the wavelet transform for image compression. First, filter coefficients are quantized in coefficient level which is based on fixed point representation. However, if coefficients are quantized, this change affects filter properties. Thus, a genetic algorithm based approach for filter coefficient quantization preserving filters properties is proposed. The genetic algorithm run twice for the quantization of high pass and low pass filter coefficients. To overcome this problem, an extension of the second approach considering the preservation of filter bank properties is proposed. It has been found that the better peak to signal noise (PSNR) is given by the filter bank quantization level. Furthermore, it gives the lowest number of non-zero bit.