Image Recognition Method Based on Robust Joint Sparse Feature Extraction

Abstract

The invention provides an image recognition method based on robust joint sparse feature extraction, which is based on the LPP algorithm and introduces L 2,1 Norm designs a more robust feature extraction method to enhance the robustness of classical algorithms. After that, an effective iterative algorithm is designed to obtain the optimal solution of the model, and the convergence of the iterative algorithm is proved, and the theoretical connection between the new method and the classical method is established. The effectiveness of the present invention is verified on the simulation data sets (COIL20, ORL).

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to an image recognition method based on robust joint sparse feature extraction. Background technique [0002] Face recognition refers to using a computer as a recognition tool to detect the position of a face from a dynamic or static image, then extract useful feature information from the face image, and finally match it with the data in the database. As early as the 1960s, the semi-automatic face recognition model was proposed by Bledsoe, and then he and Chan published the earliest report on face recognition. From the 1970s to the 1980s, some researchers began to study face recognition technology, which was of great significance to the later research on face recognition algorithms. In the 1990s, face recognition has made great progress along with the development of computer technology, and many recognition algorithms have appeared, which played a role in promoting the de...

AI Technical Summary

Problems solved by technology

[0003] With the development of artificial intelligence technology, face recognition plays a pivotal role in the fields of identity verification, human-computer interaction, video surveillance, etc., but existing classic algorithms such as principal component analysis, local preserving projection and edge discrimination analysis The algorithm uses L 2 Or the Frobenius norm is used as a model measure, so these methods are sensitive to noise, and the robustness of these algorithms is relatively poor, and there is still room for improvement in the performance of the extracted feature recognition

[0004] The traditional learning model introduced above is not suitable for some face recognition problems. The reasons are as follows: ① These two algorithms do not consider the influence of other factors on face recognition, such as shadows, lighting, etc., so under these factors its The recognition rate will drop

②The "curse of dimensionality" is easy to appear in face recognition, so the calculation required for this algorithm is very high, and it will also occupy a large amount of storage space

③ Locally maintaining the projection requires converting the original two-dimensional data into a one-dimensional vector, which will cause the dimension of the training sample to increase or even be too high, resulting in a small sample problem, that is, the singularity of the characteristic equation

Images