Face identification method based on gradient sparse representation

Abstract

The invention belongs to the technical field of image processing and pattern recognition, and discloses a face identification method based on gradient sparse representation. In recent years, owing to excellent recognition effect and wide application prospect, the face identification algorithm based on sparse representation gains more and more attention. However, the face identification algorithm based on sparse representation requires a complete training set, which is hardly satisfied in practical application; and the face identification algorithm based on sparse representation needs to solve the 1<1> minimization problem, which consumes plenty of time. In consideration of insensitivity of image gradient to uniform illuminance, the method introduces image gradient under the framework of sparse representation, and meanwhile adopts X-direction gradient, Y-direction gradient and image pixel value of a gray level image to identify a face image. Therefore, the method relaxes the requirement on the completeness of the training sample set to a great extent, and a better identification effect can be obtained only by selecting a few training samples from each type. Besides, the method solves the sparse representation factor of a testing face image on a training face image set by minimization of the 1<2> norm, so the method is fast and has higher application value.

Description

technical field [0001] The invention belongs to the technical field of face recognition, and in particular relates to a face recognition method based on gradient sparse representation and a face image recognition method. Background technique [0002] Face recognition algorithms based on whole images have been intensively studied in recent decades. At the same time, numerous feature extraction algorithms have been proposed, the most popular of which is: Principal Component Analysis (PCA) [1] , Linear Discriminant Analysis (Linear Discriminant Analysis, LDA) [2] , Locality Preserving Projections (LPP) [3] ; The goal of PCA is to find an orthogonal projection matrix that maximizes the trace of the feature covariance matrix; LDA seeks the optimal projection direction so that the distance between samples from different classes is as large as possible after projection, and at the same time from the same class The distance between the samples after projection is as small as poss...

AI Technical Summary

Problems solved by technology

However, for face recognition tasks, the training samples for each category are generally limited, which makes the assumption of sparse representation models no longer hold.

Finally, for incomplete training sample sets, algorithms based on sparse representation generally cannot achieve good recognition results, which largely limits the application of such algorithms in actual scenarios

Images