A Novel Face Recognition Method Using HOG Features Deriving From Multi-Layer Pyramid Feature Blocks

Face recognition,as one of the great challenges and long-standing academic issues,has been applied in the direction of security,financial,legal constraints and many other fields of intelligent access control system.As a kind of non-contact,long distance,strongly concealed biological recognition measure,face recognition technology can quickly identify human beings in real scene.However,in the actual use of face recognition,the face images are often obtained from complex environment,the images may be filled with changes in light,facial expression variations or occlusions,etc.In these cases,it is a great challenge and difficulty for computers to recognize different face images.In this paper,by the unique advantages of Histograms Oriented Gradients(HOG)descriptors,I propose a novel face recognition algorithm based on Multi-Layer Pyramid Feature Blocks(MPFB).First of all,the input images should be standardly preprocessed face images.In order to simulate the observation process of human eyes at different scales,this paper proceeds down sampling and up sampling.The key parts of the faces such as eyes,eyebrows,nose,nose,mouths contain a lot of featured information.Besides,these parts also contain structural and global information for face recognition.In order to highlight these key parts and the contained information,at the same time with the purpose of enhancing the robustness to occlusions and facial expression transformations,this paper proposes the concept of feature blocks and uses the fusion of feature block descriptors as a more effective way to express a face image.At each layer of the pyramid,I conduct a number of combinations that are related to the numbers,sizes and positions of feature blocks and took the best combination to extract features with the purpose of obtaining more typical descriptors.Then,I fuse feature blocks descriptors of each picture to create more representative and more integral descriptors.Besides,the multi-layer pyramid was used to construct the adjacency graph for Locality Preserving Projections algorithm to reduce the dimensionality of descriptors and make the classification process less prone to over-fitting.Finally,by employing the nearest neighbor classifier,experimental results on well-known face databases illustrate the preponderance and robustness of this approach.