Noise Filtering for Subband Image Coding Using Vector Quantization

Abstract

In this paper, we propose an adaptive postfilter for the subband coding (SBC) of images using vector quantization. This filter is to remove the inevitable channel noise after a compressed image is transmitted across a channel. We first proposed two simple schemes to detect the edges and channel noises in the subbands based on image and subband local characteristics. Then the noise-corrupted pixels at uniform, edge and texture region are adaptively lowpass filtered, while the error-free pixels are unchanged. After noise filtering, the image is reconstructed from all its subbands. Since the proposed filter does not assume any knowledge about quantization, it can be used for any SBC schemes. In addition, the filter does not degrade image quality at high frequency if high frequency signal is not corrupted by noise. The experimental results show that the proposed filter always improves coding performance when channel bit error rate is greater than 0·01%. The largest gain (5·7 dB) occurs when bit error rate is 1%. In addition, the proposed filter outperforms many well-known filters in both performance and computation complexity.