Application of Fuzzy Neural Network Predictor to Mosaic Image Coding

In this paper, we propose an effective sequential lossless compression scheme which performs sequentially pixel by pixel in raster-scan order, also so called one-pass scheme as opposed to multiple-pass scheme. The proposed method has two major components: fuzzy neural network predictor and context based adaptive arithmetic coder. The proposed predictor which is based on T-S fuzzy neural network system effectively exploits structural redundancies between mosaic-like color components, and its residuals are well-modeled by a two-side geometry distribution centered at zero. An advantage of the proposed compression method is the clean separation between prediction and entropy coding of prediction errors. Any entropy coder, be it Huffman or arithmetic, static or adaptive, binary or m-ary, can be easily interfaced with our predictor. Consequently, we can encode mosaic images at low complexity level. Compared with existing methods of lossless compression for Bayer raw data, the performance of proposed method is apparently the best.