Research On Face Recognition Methods Under Difficult Lighting Condition

As the technology further mature and the improvement of social perception, patternrecognition has been widely used in government, army, bank, social welfare, electroniccommerce, security and other fields. Face recognition technology is one of the morepopular research topics in pattern recognition. Although there have developed a lot ofapplication technology, face as a recognition feature still brings certain difficulty. In facerecognition process, the image is subject to many kinds of complex factors, such asexpression, posture, angle, illumination and so on. Therefore, the research topic has not amost rapid and effective identification technology been recognized. The accuracy of facerecognition is still doesn’t meet the actual demand.This thesis main viewpoint is multi-feature information fusion. There include featureextraction algorithm, the related feature fusion algorithm and fast classification techniques.It focuses on the research of an adaptive face feature extraction method and the weightedfusion process of large amounts data. Then, we research the method of fusion a fastclassification recognition method and a feature extraction algorithm that is suitable forvariety light images. The specific research results are divided into the following twoaspects.First, a research on the local adaptive threshold ternary pattern for face recognition isproposed. For the problems of LTP threshold optimization and universality, this thesisproposes a face recognition method based on adaptive threshold LTP. Each neighborhoodpixels’ discrete ratio is define as the adaptive LTP threshold. Then, according to theinformation entropy weight, the layered coding characteristics are fuse. The experimentalresults show that this method is more robust to illumination changes and noise.In addition, the face recognition precision is effectively improved than before.Second, a near-infrared face recognition method by fusion of E-GV-LBP and FKNNis proposed. In order to realize the face recognition with complex illumination changesand improve the recognition efficiency, there proposes a new face recognition algorithmby fusion of E-GV-LBP and the improved FKNN. Firstly, using Gabor wavelet transformset for images. Secondly, extracting encoded LBP features what contains space, scale anddirection information at the same time. Finally, using the improved fast KNN algorithmbased on the spatial domain classify the extracted feature data. The experimental results show that no matter under near-infrared or visible light, the face recognition accuracywith new method are improved and the recognition time is reduced simultaneously.