Face Recognition Research Based On Improved Skin Color Detection Algorithm And Improved CNN

In image processing and pattern recognition technology,face recognition technology has always been a research hotspot.In the current face recognition technology,the face recognition algorithm based on skin color detection and convolution neural network has created great contributions to the research and application of face recognition.Skin color detection play a role in fast detection of skin color information,so in face recognition,it can be used to quickly detect the face under a some backgrounds,and then the detected face will be identified by recognition agorithm.Convolutional neural network has embodied a bigger advance in face recognition agortithm.Convolutional neural network can automatically extract image features,and its recognition accuracy is relatively high.Therefore,the application of skin color detection and convolution neural network in face recognition is worth studying.In face recognition,facial expressions and gestures are the main interference factors affecting face recognition.Face recognition mainly relies on extracting facial features for recognition.Generally,when facial expressions and gestures change,more complete facial features can still be extracted from images,but it is difficult to extract more complete facial features if face occurs occlusion condition.Considering the influence of these interference factors on face recognition,this paper wages relevant researchs.The main work includes the following three aspects.(1)The related concepts and implementation methods of some algorithms commonly used in face recognition are elaborated,and some of them are analyzed,problems of which are also introduced.On the basis of these problems,some improvements' ways of related argorithms are poropsed,hence a face recognition algorithm which can further improve the recognition effect is proposed.(2)By combining pseudo-color algorithm,BP neural network and traditional skin color detection agorithm,an improved skin color detection algorithm based on pseudo-color and BP neural network is created.The algorithm is used to detect the main information areas of face in gray and color images,aming to improve the efficiency of face recognition.The input data of BP neural network adopts the fused P CA feature,namely LTP-PCA feature,and the optimization of BP neural network is realized by the combination of genetic algorithm and an improved quasi-Newton algorithm.From the simulation results,we can found that the improved skin color dete-ction algorithm can detect the main information region of face accurately.(3)An improved RPCA algorithm of low-rank reconstructed images is proposed and used it to denoise face image.Then a PCAL convolution neural network is constructed by combining the convolution filter constructed by PCA with LeNet-5 network.Eventually,a face recognition algorithm based on convolution neural network with denoise process function is proposed by combining the improved RPCA algorithm with PCAL convolution neural network.To compare other face recognition algorithms of convolutional neural networks,the algorithm is proved that effects of recognition are better in the no-occlusion and occlusion environment.