Research On Face Image Restoration And Recognition Algorithm

There are many factors that degrade the face image during their obtaining,so the presence of image degradation, such as face noise and face bluring isunavoidable. However, in the flied of face detection and face recognition,theclear face images are needed. Therefore, face restoration become the basicsubject of face recognition which we research. Face recognition is a challengingproblem, which has great theoretical significance and practical value. How toovercome face pose variation and Optical noise impact using computer andperform the fast and exact face recognition have been a research focus in the fieldof image processing and pattern recognition.This paper introduces and analyses the present situation of the facerestoration and face detection and recognition’s algorithm. This paper takes thevery low resolution image and spontaneous image as the research object, andproposes a research topic and the feasibility of approach for very low resolutionface restoration and face recognition. The main works of this paper is as follows:1. Very Low Resolution Face Image Super-Resolution Based on DCTIn this paper, we introduced the VLR problem and the principle of DCT. Onthis basis, we proposed the improved DCT algorithm. A detailed description ofthe process of our algorithm was given. Then we completed the experimentalvalidation and results analysis.2. Face Detection Based on MB-LBPIn this paper, we simply introduced the definition of MB-LBP and itsuniform pattern. Then we described the process of constructing classifiers byadaboost. Finally, we proposed the face detection method which is based onMB-LBP and eye tracking and completed the experimental validation and resultsanalysis.3. Face recognition Based on MB-LBP On the basis of the previous results, the problem of face recognitionmulti-class classification was analyzed. This paper applies MB-LBP and adaboostalgorithm to face recognition using a new cascade classifier, and our finalexperimental results have compared with the common LBP algorithm, PCAalgorithm.