A fast reversible compression algorithm for Bayer color filter array images
King-Hong Chung, Yuk‐Hee Chan · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference · 2009
Abstract – Most digital cameras perform color demosaicing and compression sequentially to yield a color output. Recent reports indicate that the alternative compression-then-demosaicing approach outperforms the demosaicing-then-compression approach in terms of image quality and complexity. This paper presents a fast reversible Bayer image compression algorithm for the alternative approach. A statistic-based prediction is proposed to de-correlate the wavelet subband coefficients. By learning from experiences, the proposed predictor can improve its prediction performance adaptively. A context-based Golomb Rice code is then proposed to compress the subband residues. Simulation results show that, as compared with the existing lossless CFA image coding methods, the proposed algorithm can achieve a low bit-rate with lesser computation. 1. INTRODUCTION To reduce cost, most digital cameras acquire scenes using a single image sensor. In these cameras, a Bayer color filter array (CFA) [1], as shown in Fig. 1, is placed in front of the sensor such that the sensor samples only one of the three primary color components at each pixel. The mosaic-like CFA image, i.e. the raw sensor output, is first converted to a full color image, via color demosaicing [2-5], and then compressed for transmission or for storage. Recently, some reports [2,3] pointed out that such a demosaicing-then-compression approach is inefficient in a compression point of view as the demosaicing process introduces the redundancy which will be eventually removed in the later compression step. Accordingly, an alternative approach [2,4] which carries out compression prior to demosaicing has been proposed lately. Under this new workflow, a digital camera can have a higher quality color output and more power-efficient design as the critical yet computationally heavy processing steps like color demosaicing and post-processing can be carried out offline in a powerful personal computer. These advantages motivate the demand of compression techniques for CFA images. Generally, CFA image compression can be either lossy or lossless. Lossy compression results in a decompressed output different from the original [2,4-6]. It is rarely used in practice as the demosaicing to be carried out in the future is very sensitive to the corruption introduced in the compression. Lossless compression, on the contrary, provides a decompressed output exactly the same as the original. It is commonly used for some high-end photography applications like professional advertising where the original CFA image is required for producing the high quality full color image. Obviously, some lossless compression standards for grayscale images such as JPEG-LS [7] and JPEG 2000 [8] can be directly applied to compress CFA images. Nevertheless, they attain a fair compression performance only. Recently, two advanced lossless CFA image compression algorithms [9,10] were proposed. In [9] (LCMI), the Mallat packet transform is exploited to de-correlate the mosaic color data. The transform coefficients are then compressed by adaptive Golomb Rice code. As for CMBC[10], it uses a context matching technique to rank the neighboring pixels for predicting a pixel. This method generally provides a better compression performance as compared with the existing lossless CFA image compression schemes. However, it demands a relatively high computational complexity. In [9], it is found that applying a simple one-level 2D-wavelet transform to a mosaic CFA image is equivalent to separately transforming the full resolution green channel and the down sampled color difference images and then summing up the results. Based on this