Improvement Of Face Recognition Algorithm Based On Regional Gabor Features

Face recognition is a very popular research topic in biological field,with the development of pattern recognition and artificial intelligence technology.In order to overcome a series of problems in face recognition,such as illumination,facial expression,partial occlusion,etc.After analyzing and comparing several common face recognition algorithms,it is found that the Local Gabor feature recognition algorithm can not only deal with all kinds of noise interference,but also ensure a certain recognition rate,and the recognition results are stable on different face database,and no big fluctuation will occur.Based on these existing advantages,we choose region Gabor feature recognition algorithm and make a series of improvements to this algorithm,so that it can be further extended to static face recognition.The contents and innovations involved in are as follows:1.Two-dimensional Gabor wavelet is used to extract the features of the image.Then,according to the distribution of the main organs of the face,the face is divided into regions,and the features of each region are fused in series,and then the results are classified and identified.2.Based on the characteristics of Gabor extraction in the region further introduces the Local Gabor Binary Pattern(LGBP).After extracting features from Gabor wavelets,the algorithm uses Local Binary Pattern(LBP)algorithm to extract two values of the extracted results,and effectively preserves the details of the image.Then,the results obtained from the two algorithms of region Gabor feature extraction and region LGBP feature extraction are connected in series,and the transformation of facial features from visual images to processing data is realized.3.It can be found in the process of further processing of the results of feature extraction.Because of the characteristics of Gabor wavelet,the feature dimension is greatly increased while the feature extraction is fine,and the noise contained in the feature increases exponentially,which causes huge burden to classification and recognition.In order to ensure the accuracy of the classification,it is necessary to reduce the dimension of the image before it is classified and recognized.In order to avoid the massive loss of image information in the process of dimensionality reduction,a large number of linear and nonlinear features are included in the image feature.By using the Principal Component Analysis combined with Linear Discriminant Analysis algorithm(PCA+LDA),Kernel Principal Component Analysis combined with Linear Discriminant Analysis algorithm(KPCA+LDA)for dimensionality reduction,then the dimension reduction results are further weighted fusion,using K-nearest neighbor algorithm to realize the algorithm,getting the final recognition rate.According to the above research content and innovation,through the establishment of comparative experiments,it is verified that the use of region feature fusion algorithm is better than only one feature extraction algorithm.Further,it is further explained that the improved algorithm from feature extraction,feature dimensionality reduction to the final classification and recognition process is effective for face recognition in complex environment,and can be further promoted.