Study On Demosaic Techniques In Video Preprocessing

With the rapid development of video technology and the popularization of digital products, the collection and disposal of image and its sequence which is of high quality, become as an active field, in which demosaic is a necessary key technique.Except for discussing the principle of image interpolation, this paper mainly analyzes the principle and practicability of four image demosaic interpolation algorithms, and compares them, then puts forward demosaic interpolation algorithm using adaptive filter.The main task of researching image interpolation algorithm is seeking scientific interpolation method and finding the balance between the complexity of interpolation and the quality of interpolation image. Through analyzing the image sequence of CFA model, we can get the following conclusions:1. Three coloury channels R, G, B have close relativity;2. Brightness channel, G, has higher sampling rate than chroma channels, R and B. According to this, we know that variable G saves more image detail. Because of the low sampling rate of channels R and B, they will overlap in the high frequency region, such as region's rim, and produce artificial color. So, the interpolation of color channels R and B is the most important factor to improve the quality of image.Considering the real time demand of applied systems, and thinking through computing complexity and recovering image quality, we put forward an interpolation algorithm based on the adaptive filter technique. Firstly we pick up high frequency information from variables R, B variable as modifying factor to finish G variable interpolation, and then estimates the lost variables R, B, using the assumption that tonality is smooth. Finally, we deal with the filter in chromatic aberration variables (R-B) and (B-G) to ensure chromatic change smoothly, and remove the artificial color in the last operation. The outstanding advantageous of this algorithm is that it can satisfy the real time demand of different systems, through selecting different SNR.According to the results of this experiment, our algorithm is a demosaic algorithm with high efficiency. It can greatly decrease the rim fuzzy effect, and the artificial mark caused by some algorithms available. It is a practicable image demosaic interpolation, as it is low in complexity, has better efficiency in dealing with image and selective quality, and can satisfy the real time demand of different systems.