Research On Image Interpolation And Lossless Compression Based On Bayer Pattern

This paper mainly discusses Bayer data compression and CFA demosaicing. CFA demosaicing is necessary in a single-chip digital camera before fully RGB color image. Advanced CFA demosaicing algorithms are some kind of property of camera manufacture, also are used popularly in the camera.CFA demosiacing has two steps, first image interpolation then image enhancement. A wide discussion about most of demosaising algorithms today is present. Spatial smoothness and color correlation are two mainly factors which effect reconstructed image quality. But in spite of performance of these methods, the distortions such as false color, edge blurring, which usually generate in the edges and details of the reconstructed image, are unavoidable. In this paper, adaptive CFA demosaicing methods is proposed. Using data-adaptive filtering concept, the proposed method fully uitilizes the inter-channel correlation, decreases the errors in the experiment and the distortion in the color channels effectively. Experiment verify that the proposed method outperforms the existing state-of-the-art methods both visually and in terms of peak signal-to-noise ratio, at a notably lower computational cost. The average PSNR of red, green and blue reach 38.52, 42.24, 38.83dB respectively, and the improvement to most methods is around 4dB~6dB.The problem of Bayer data compression is quite recent. Most methods directly use traditional coding techniques after processing Bayer image simply, but most of these do not offer the same performances when images captured by digital sensors in CFA format are processed. The kernel of Bayer data compression is to preprocess Bayer data and choose the coding techniques. This paper proposes a new lossless compression method suitable for Bayer data compression. For the characteristics of the Bayer data, a low-complexity and high-efficient compression method is proposed. The algorithm on one hand uses color models reducing the correlation between the plane the other hand, on the other hand uses prediction techniques reducing the relevance of the interplane. Finally, the predicted error is encoded using Golomb-Rice coding. The experimental results obtained by processing real Bayer data are quite promissing, achieving an average bpp of 4.07. It saves 2.431bpp, 0.845bpp compared with JPEG-LS and JPEG2000 respectively.