Method for obtaining human face illumination invariant images based on multiscale anisotropic diffusion

Abstract

The invention relates to a method for obtaining human face illumination invariant images based on multiscale anisotropic diffusion, belonging to the field of image processing technologies. The invention is based on a Lambertian convex surface model for decomposing the human face image to a small-scale feature image and a large-scale feature image. The small-scale feature image can be regarded as the ideal human face illumination invariant feature image. The core is characterized in that new descriptors with inconsistent intervals are introduced for strengthening edge retention capability of an anisotropic diffusion algorithm to low frequency domain images so as to greatly weaken image halo effect of the algorithm; meanwhile, a new transfer coefficient is provided, and noised caused by edge sharpening is reduced; and an anisotropic diffusion constraint is introduced, and the method is more suitable for treating the illumination problem of the human body image. Experiments show that the invention can obtain good treatment effect even in extremely poor lighting conditions and can effectively improve robustness of face recognition or face certification to changes in lighting conditions.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a method for obtaining a face illumination-invariant image, which can be directly applied to a real-time face recognition or face authentication system under the condition of illumination change. Background technique [0002] Due to the important position of face recognition or face authentication in the fields of national security, military security and public safety, the research on face recognition or face authentication technology has developed rapidly. But so far, the impact of face recognition on ambient lighting has insurmountable defects, which are mainly due to the change of the face image caused by the influence of the illumination change, which is even greater than the change caused by the individual differences of the face image. In addition, the ambient lighting during face authentication is different from that during registration, and the recogn...

AI Technical Summary

Problems solved by technology

This type of method has a very ideal effect, but the disadvantage is that the complexity of the algorithm is quite large, and there is a high requirement for the number of face image samples, and it is usually difficult to obtain practical applications.

The fourth category is the method of extracting illumination invariant features, which can be divided into edge-based image feature methods and feature image-based methods. Research shows that the edge-based image feature method is more effective in face recognition algorithms when the illumination changes are more complex. The performance of the method decreases sharply with the increase of the illumination angle change.

Images