Face Recognition Based On Improved SIFT Deep Belief Network

Face recognition technology has many advantages over other biometric technologies.Face recognition technology is affected by uncontrollable factors such as illumination,rotation,posture and expression changes,resulting in a lower recognition rate.It was proposed to combine the improved SIFT algorithm with the deep belief network to solve the above problems.1)Aiming at the problem that the rectangular region in SIFT algorithm is easily disturbed by rotation variation,the region segmentation Haar-SIFT deep belief network algorithm was proposed for face recognition.The SIFT algorithm was improved by the rotation invariance of the circular region.The extracted feature vector was reduced by haar wavelet feature.Deep belief network was used to classify to enhance the antiinterference ability of the algorithm on rotation and posture,and improved the effect.2)Aiming at the redundancy problem of face feature data extracted by SIFT algorithm,sparse coding is used to optimize SIFT algorithm.Sparse coding can eliminate a large number of redundant feature information.The problem of SIFT feature data redundancy,low matching rate and long matching time was solved.Finally,the deep belief network was used to identify and classify,the simulation results showed that the improved algorithm was robust to the interference of expression changes.3)Because the region segmentation Haar-SIFT-DBN model had a good effect on pose recognition.The sparse SIFT algorithm was more accurate when the facial expressions were different.Therefore,the two algorithms were combined.Using local variance similarity method to optimize sparse coding to improve the description of image detail features,and proposed a face recognition model based on region segmentation sparse detail SIFT deep learning,which improved the recognition accuracy.The improved SIFT algorithm and deep belief network could extract face features more effectively and accurately,which improved the accuracy and matching rate of face recognition and provided new ideas for the face recognition technology.Figure 33;Table 1;Reference 60...