Research On Face Recognition Algorithms Based On Small Sample Conditions

With the in-depth study of image recognition technology and the rapid development of artificial intelligence technology,the field of face recognition has received great attention from researchers and has been widely used in many fields such as video surveillance,social networking,and face unlocking.However,face recognition technology still has great challenges due to factors such as illumination,posture,occlusion and expression changes,and sample size.Among them,the limited number of samples of face recognition in practical applications is one of the main factors affecting face recognition performance.In view of the small sample size and the low recognition rate of most algorithms under small sample conditions,how to increase the number of samples under small sample conditions and effectively improve the recognition rate of the algorithm is deeply studied.main tasks as follows:For the small sample face recognition field,the common problems of image recognition such as face pose change,facial expression change and illumination imbalance,and the fundamental factors of face sample deficiency under small sample conditions,through in-depth research,symmetrical face images are helpful to overcome The influence of illumination problems and face poses and expression changes,generating virtual face samples is an important way to increase the number of small samples,but the virtual face samples have the disadvantage of distortion and reduced recognition accuracy.In this regard,this paper proposes a local approximate symmetric face image generation method.The average of the two original face images is used as a virtual face sample,and the virtual half face is replaced with the original half face to synthesize a face.Because the method utilizes both a virtual face and a symmetry-based primitive face to generate a synthetic face,the synthesized face is close to the original face.The experimental results show that reasonable virtual samples can not only correctly simulate the changes of face posture and expression,but also help to improve the accuracy of face recognition.In face recognition in special environments such as counter-terrorism and criminal hunting,training samples are often very rare,so they do not represent test samples.Therefore,face recognition under small sample conditions is difficult to consider intra-class changes based on their inter-class changes,and residuals between images will result in erroneous face recognition results.In order to effectively recognize the face under the condition of small sample of each face image,this paper finds a linear projection matrix to make the intra-class reconstruction residual smaller,and in the transformed low-dimensional space,the inter-class reconstruction residual is larger..Through research,it is found that the image features obtained by singular value decomposition have the properties of stability,rotation invariance and image transformation invariance.In this regard,this paper proposes a new unsupervised dimensionality reduction algorithm,called singular value decomposition projection method(SP),which can better deal with the decision rules based on residuals in sparse representation classification.Through the effective combination with sparse representation algorithm,a series of experiments and results comparison analysis were carried out using synthetic database,AR database,PIE database and FERET database.The results show that the algorithm effectively improves the accuracy of face recognition under small sample conditions.At the same time,the algorithm has the characteristics of low computational cost,good robustness and uncorrelated sample characteristics,and maintains the simplicity and effectiveness of the unsupervised dimensionality reduction algorithm.