A low complexity image compression algorithm for Bayer color filter array

Abstract

Digital image in their raw form requires an excessive amount of storage capacity. Image compression is a process of reducing the cost of storage and transmission of image data. The compression algorithm reduces the file size so that it requires less storage or transmission bandwidth. This work presents a new color transformation and compression algorithm for the Bayer color filter array (CFA) images. In a full color image, each pixel contains R, G, and B components. A CFA image contains single channel information in each pixel position, demosaicking is required to construct a full color image. For each pixel, demosaicking constructs the missing two-color information by using information from neighbouring pixels. After demosaicking, each pixel contains R, G, and B information, and a full color image is constructed. Conventional CFA compression occurs after the demosaicking. However, the Bayer CFA image can be compressed before demosaicking which is called compression-first method, and the algorithm proposed in this research follows the compression-first or direct compression method. The compression-first method applies the compression algorithm directly onto the CFA data and shifts demosaicking to the other end of the transmission and storage process. The advantage of the compression-first method is that it requires three time less transmission bandwidth for each pixel than conventional compression.

Compression-first method of CFA data produces spatial redundancy, artifacts, and false high frequencies. The process requires a color transformation with less correlation among the color components than that Bayer RGB color space. This work analyzes correlation coefficient, standard deviation, entropy, and intensity range of the Bayer RGB color components. The analysis provides two efficient color transformations in terms of features of color transformation. The proposed color components show lesser correlation coefficient than occurs with the Bayer RGB color components. Color transformations reduce both the spatial and spectral redundancies of the Bayer CFA image. After color transformation, the components are independently encoded using differential pulse-code modulation (DPCM) in raster order fashion. The residue error of DPCM is mapped to a positive integer for the adaptive Golomb rice code. The compression algorithm includes both the adaptive Golomb rice and Unary coding to generate bit stream. Extensive simulation analysis is performed on both simulated CFA and real CFA datasets. This analysis is extended for the WCE (wireless capsule endoscopic) images. The compression algorithm is also realized with a simulated WCE CFA dataset. The results show that the proposed algorithm requires less bits per pixel than the conventional CFA compression. The algorithm also outperforms recent works on CFA compression algorithms for both real and simulated CFA datasets.