Wavelet filter banr selection based on power spectral density

Filter banr design for wavelet compression schemes is very important. Although, JPEG2000 standard adopted bi-orthogonal 9/7-tap Daubechies wavelet, it is experimentally shown that this banr is well suited for natural images, which they have small amount of energy in the high frequency in horizontal and vertical directions. In this paper, we introduce a method for filter banr design according to power spectral density function of input source. By this method we propose a new efficient 9/7 wavelet filter banr for image compression applications, whose coefficients are adaptively chosen according to power spectral density function of the input image. The experimental results for images by substituting our filter banr in the JPEG2000 standard coding mode show that little gain was obtained by for natural images, since they have small amount of energy in the high frequency bands in horizontal and vertical directions. However, 0.5–1 dB improvement of PSNR has been obtained for images such as Fingerprint, which have a large amount of energy in high frequency range.