Research On Face Recognition Technology Under Occlusion

Compared with the traditional user identity information recognition method,face recognition can more accurately identify the user's identity information.However,there are various occlusions in the face recognition process.The existence of these occlusions will have a serious impact on the accuracy of face recognition.The face recognition for occlusion is as follows:Firstly,through the learning of the occlusion face recognition process,the three main modules in the face recognition process were mastered,namely preprocessing,feature extraction and face recognition.The most important one was the feature extraction stage.The selected methods were different.There were also differences in the extracted features.Secondly,using the sparse coding method,the experiment was carried out in the AR face image database and the experimental results were analyzed.It was concluded that the recognition rate would decrease with the increase of the occlusion area in the face image.When the occlusion area exceeds 50%,the recognition rate it would drop rapidly,and a robust sparse coding method was introduced for this problem.Then the experiment and analysis of the data reveal different occlusion areas,different dimensions and three different occlusions(glass,scarf and glasses scarf mixed occlusion)In the case of the robust sparse coding method,the recognition rate was increased by 5% to 10%.Then,the theoretical analysis of the dynamic graph regularization method shows that it had a good recognition effect on the time series,so the face was segmented and rearranged from left to right and top to bottom,and the obtained sequence information was obtained.Just like a time series,it was bold to try to apply the dynamic graph normalization method to face recognition in occlusion conditions.The dynamic graph regularization method was used to perform similar experiments in the AR face image database.The analysis data showed that the recognition rate of the dynamic graph regularization was higher than the sparse coding method in different contexts,but the recognition speed was slow.Finally,through the above two sets of experiments,the advantages and disadvantages of the robust sparse coding method and the dynamic graph regularization method were obtained.Finally,the two methods were elastically combined,and an improved algorithm called RSD was proposed and the improved algorithm was implemented.The experimental verification showed that when the number of face image samples exceeds a certain value,the face image of the training set had an occlusion test set.The face image was not occluded,and the training set face image had no occlusion test set.The face image had occlusion and training.When the face image of the test set was occluded,the recognition rate of the improved RSD algorithm was 3%~5% higher than that of the other two methods alone,which verified the effectiveness of the improved RSD algorithm.Figure 47;Table 5;Reference 50.