Markov Network Model Human Face Image Super-resolution Algorithm,

In recent years, face image as an important study object has been widely researched in the field of computer vision. With the increasing development of multimedia technology, there is a big demand for the quality of face image. It will increase the cost when we replace low-resolution sensors with high-resolution ones, and sensors are also subjected by the physical hardware limitations. So the use of super-resolution technology becomes a new choice. Super-Resolution (SR) technology is one kind of method which obtains the High Resolution (HR) image based on the signal processing technology. The SR can overcome the inherent resolution limitation of LR imaging systems, and the major advantage of the signal processing approach is that it may cost less and the existing LR imaging systems can be still utilized. The SR has been widely used in video, medical imaging, surveillance, remote sensing, military and other fields.We study Markov Network model of human face image super-resolution techniques. The MN model has two key issues: the solution of the observation function and the solution of the transfer function. First of all, this thesis proposes a face alignment method based on eye location, by which method realizing the location restrictions, speeding up the convergence rate of the Markov network, and effectively avoiding the interference. Then, in the search process of block matching, we compute the similarity measure in accordance with the distribution of high frequency components. In smooth region, we employ a simple gray block similarity measure method, while combine gray-scale with texture characteristics in the region of large gradient. To solve the transfer function, we use overlapping technique of sub-image blocks and the horizontal direction compatibility priority way to increase the relevance of the match. This method is also able to simplify the calculation of hidden layer nodes.Finally, we have developed a human face super-resolution prototype system based on Markov Network model on VC + +6.0 platform. The system can collect a training set of high-resolution face image, carry out face alignment and block similarity measurement. The experimental results are compared with other methods and the performance of the system is analyzed.