Bayer pattern compression by prediction errors vector quantization

Most digital cameras acquire data through a Bayer Colour Filter Array (CFA) placed on sensors where each pixel element records intensity information of only one colour component. The colour image is then produced through a pipeline of image processing algorithms which restores the subsampled components. In the last few years the wide diffusion of Digital Still Cameras (DSC) and mobile imaging devices disposes to develop new coding techniques able to save resources needed to store and to transmit Bayer pattern data. This paper introduces an innovative coding method that allows achieving compression by Vector Quantization (VQ) applied to prediction errors, among adjacent pixel of Bayer Pattern source, computed by a Differential Pulse Code Modulation (DPCM)-like algorithm. The proposed method allows a visually lossless compression of Bayer data and it requires less memory and transmission bandwidth than classic "Bayer-oriented" compression methods.