Research On High-definition TV Back-end Image Processing Algorithm

In Full high-definition digital TV chip, the back-end image processing module is critical to determine image quality. Due to the time-varying characteristic, video has the redundancies on space and time. It is a promising research area to use motion estimation and compensation(MEMC) in video back-end image processing, as the intersection of the image static and the video dynamic characteristics. It can not only keep the original advantage in traditional image processing technology, but also improve image quality and robustness by fusing the useful motion information. However, this new area is just beginning, and the basic theory and technical research work are needed urgently.In this dissertation, in-depth and comprehensive research has been done on the major function modules of digital TV back-end processing system, based on MEMC. The research includes: three dimensional recursive search(3DRS) based MEMC, noise reduction, de-interlace, stabilization and frame rate up-conversion.The main research work is as follows:1. Algorithms for MEMC have been studied and some block matching methods have been discussed in this dissertation. Based on 3DRS, some improvements are made for settling the shortcomings of 3DRS, such as convergence condition setting, candidate vector update, scan sequence optimization and bi-direction ME. This improves the convergence rate. In addition, a MC method based on block erosion is proposed to solve the uncover problem in motion compensation(MC). Then, a 3DRS based frame rate up-conversion module is developed by combining with this method. Experimental results demonstrate that the proposed algorithm can improve image quality significantly by providing effective MEMC module.2. The noise reduction and de-interlace have been studied. A method which fuses MC technology is proposed. The key point is to restore the time correlation in a fixed space position. It can not only solve the identification problem for normal pixels and noise under the strong randomness of Gaussian noise situation, but also can solve the support problem for large direction and angle in traditional de-interlace. In addition, using both noise reduction and de-interlace in a MC module, can reduce the resource consumption and hardware cost effectively.3. The stabilization method has been studied. A novel global motion estimation(ME) approach based on projection is proposed, which can be applied to the global motion parameters in 3DRS ME module. During realization, using the motion vector cache generates the global motion vector, which can make good use of the existing resources and minimize consumption system. It can eliminate jitter between the image sequences to achieve stabilization significantly, and meet the need for low computation complexity in high-definition video real-time processing.