Research On Image Sensor Preprocessing And Vector Quantization Of Image Coding Algorithm

With the rapid development of computer multimedia, information and network technologies, there have been more and more demands for using digital devices to produce images and videos, and storing, processing and transmitting these images in digital style. Due to the large amount of digital images, we need huge storage space, and the required bit rate is much higher than existing channel capacity. To improve the transmission efficiency and reduce the storage requirement, efficient encoding algorithms should be used to remove the residual information in images, and fewer bits should be used to describe images on restrictions of given distortion.This thesis researches on vector quantization of image compression algorithm based on the above premise. Vector quantization has long been established as an efficient lossy compression technique popular in image coding field due to its optimality in information theory and simplicity in practical applications. By finding the nearest code word's corresponding index to the decoder, vector quantization coding algorithm can usually provide high compression ratio and a simple table-look-up decoding operation. In conventional vector quantization, an image is divided into blocks that are all the same size. This uniform division could be redundant. In order to improve the efficiency of vector quantization algorithm, this paper proposes a novel vector quantization of image compression algorithm, which integrates multiple stage vector quantization and quad-tree segmentation. The simulation result show that the proposed algorithm consumes less time to encode, while achieves better image quality with higher PSNR at the same bit rates than full search algorithm.At last this thesis introduces a few commonly used color interpolate algorithms based on Bayer color filter array. These color interpolate algorithms are farther compared in terms of the image quality and the computational complexity. Experimental results show that bilinear interpolation algorithm has the lowest computational complexity and adaptive color plane interpolation has the best image quality.