Researches On Face Hallucination Algorithms Based On Interpolation And PCA

With the development of video surveillance technology, face detection and recognition have the great prospects in security systems or criminal identificaition. And it has become a research focus in current pattern recognition and artificial intelligence. However, if the resolution of the face images collected by cameras is not high enough, it affects the success rate of the face detection and recognition. Image super-resolution technology can generate a high resolution image from its low resolution input. For the request of face detection and recognition, face image super-resolution (or called "face hallucination") emerged. It's the particular application of image super-resolution in face images.The research focus of this paper is the super-resolution algorithms and the face hallucination technology.Firstly, the thesis researches several typical super-resolution algorithms based on edge-adaptive interpolation and implements them, whose principles of the resolution enhancement are carefully analyzed. The advantages and disadvantages of each algorithm are vertified by experiment. Secondly, several basic methods of face hallucination are detailed described. This paper introduced the framework of super-resolution algorithm based on Bayesian probability, described a variety of ways of prior probability model, and introduced the three constraints on face hallucination.For some broadly concerned and discussed issues about the reality and errors of face hallucination technology, the paper launched an in-depth, detailed discussion and offered its own views and conclusions by analysis. A novel face hallucination method based on interpolation and principal component analysis is presented. The method considers that a high-resolution face image is composed of the global face image and local high-frequency details. At first, the low-resolution input image is reconstructed by the new edge-directed interpolation algorithm to get the global face image. Then, the structural information of the global face image is extracted by PCA. With the help of the training database, we can synthesize the local details according the structural information. Finally, the global face image and local details are combined to the final face super-resolution results. Experiments show that this method can achieve face hallucination effectively.With the assistance of the OpenCV computer vision library, we developed an application to realize the proposed algorithm in the Visual C++enviroment.Finally, the conclusions summarized the work of this paper and pointed out the work needing to deal with in the further research.