Vector quantization for compression of noisy imagery

摘要

Compression of noisy imagery usually consists of two stages, prefiltering followed by encoding. In this paper we present a technique based on vector quantization, which combines noise reduction and compression into one step. The idea is to generate a codebook, consisting only of clean image data, which is then used for quantization of the noisy imagery. Simulations performed shows that this approach can efficiently handle images corrupted by noise, and compared to MPEG-4 encoding, this technique, in spite of its simplicity, is the better choice when dealing with high levels of noise.