Study On Single Sample Based Face Recognition

Face recognition has been increasingly applied in e-commerce, video surveillance,and access control systems. In recent decades, a great range of face recognition methodshave been proposed. Major classic algorithms are able to achieve high recognition rateunder the condition of multi-sample. However, in many practical applications, only oneface image for each person can be collected, such as passports, driving licenses, identitycards and so on, which gives rise to the single sample face recognition problem.Confronted with the single sample problem, many existing classic face recognitionmethods obtain unsatisfactory recognition rates, and some algorithms cannot even beutilized. Therefore, the study of the single training sample face recognition technologyhas great theoretical significance and application value.This paper first introduces the background and research significance of the singlesample face recognition, and then summarizes and analyzes the status and progress ofthe single sample face recognition research at home and abroad. By analyzing thedifference between the single sample face recognition and the multi-sample facerecognition, as well as the current mainstream face recognition technologies, the issuesof existing single sample face recognition are identified.Knowing that the current single sample recognition technologies neglect the localtexture features, apply the local image texture features in single sample face recognitiontechnology was proposed in this paper. After comparing Gabor, LBP and image gradienttexture extraction algorithm, and assessing both the recognition rate and recognitionspeed, we utilize the image gradient algorithm to extract texture features. Consideringthat the feature extraction is in the space of gradient, the method is very sensitive tonoise. This paper presents an image gradient algorithm based on B-spline, which selectB-spline functions for image filtering before use to extract local texture information,which can control the filter effect by adjusting order. The experimental result on theopen standard database shows that this algorithm for the single sample face recognitionis better than the traditional methods.Since the singe face recognition method hopes to abstract more face features fromone sample whenever possible, this paper considers to integrate the global imagefeatures and local features for single sample face recognition. This paper proposesG2DPCA algorithm which integrates the global image features extracted by2DPCA and local image gradient texture features extracted by image gradient extraction algorithm.At last, the experimental result on the open standard database shows that the algorithmin single sample face recognition is better than the traditional ones.