Single-sample face recognition method based on sparse probability distribution and multi-stage category screening

Abstract

The invention discloses a single-sample face recognition method based on sparse probability distribution and multi-stage category screening, comprising the following steps: first, a face image is divided into blocks and the blocks are divided into sub blocks, wherein it is assumed that the different sub blocks in the same block are different samples belonging to the same category, and thus, the problem that many sub-space learning methods and sparse representation methods cannot work or the performance decline under the condition of a single sample is solved; then, each block of the face image is recognized, and the category probability distribution of the face is obtained by means of voting; and finally, irrelevant categories are excluded iteratively using a multi-stage category screening structure based on the idea of minimum entropy, and an ideal face recognition classification effect is achieved. The method has good robustness to expressions, illumination change and occlusion, and has high recognition accuracy. Thus, a simple and effective solution to the single-sample face recognition problem is provided.

Description

technical field [0001] The invention belongs to the field of face recognition, and relates to a new and effective single-sample face recognition solution, in particular to an automatic face recognition system in which each object to be recognized has only one training image. Background technique [0002] As a non-contact biometric identification technology, face recognition has always been one of the hottest research topics in the field of pattern recognition and computer vision. It is widely used in identity authentication, security monitoring, multimedia entertainment, intelligent human-computer interaction, etc. field. Generally speaking, face recognition mainly refers to computer technology for identifying human identity through human facial visual information in digital images or video images. Compared with other biometric identification technologies such as fingerprint recognition and palmprint recognition, face recognition has outstanding features such as more conven...

AI Technical Summary

Problems solved by technology

[0004] The above-mentioned face recognition methods based on sparse or collaborative representation or traditional face recognition methods based on subspace learning are heavily dependent on the number of training samples, and most of the methods face recognition performance with a single-sample face recognition problem. Powerful, even some methods can't work at all in the case of a single sample. Only when the number of training samples of each class is large can it show better robustness to noise and occlusion

Therefore, when there is usually only a single training sample in many real-world applications such as ID card recognition, customs passport verification, and security monitoring, the recognition performance of these methods will drop sharply or even be completely infeasible.

Images