Illumination-robust facial image local texture enhancement method

Abstract

The present invention relates to an illumination-robust facial image local texture enhancement method includes the following steps that: logarithmic transformation is performed on the gray value of an inputted original face image I, so that a logarithmic transformation result image I' can be obtained; Gaussian differential filtering and bilateral differential filtering are performed on the logarithmic transformation result image I', so that differential filtering result images IDoG and IDoB are obtained, image information fusion is performed on IDoG and IDoB, so that a fusion result image I" is obtained; and the fusion result image I" is divided into sub-image blocks, grayscale-equalization processing is performed on each sub-image block by means of a mean normalization method, and the sub-image blocks are spliced according to division positions, and then the pixel gray value range of a spliced image is compressed by means of a hyperbolic tangent function, and an image is outputted. With the method of the invention adopted, facial images imaged under different illumination conditions can be processed; illumination influence can be eliminated; face local texture information can be enhanced; and recognition accuracy in face recognition application can be improved. The method has the advantages of low algorithm complexity and high light robustness.

Description

technical field [0001] The invention relates to the technical field of automatic recognition of human face images, in particular to an illumination-robust local texture enhancement method of human face images. Background technique [0002] Face recognition technology has developed rapidly in the past few years, and some excellent face recognition algorithms have been successfully transformed into commodities and applied in actual production and life. However, the use of such software has certain limitations, and most of them require indoor or controlled environments. Under uncontrollable conditions, there are still many challenging problems, such as changes in facial expressions, age, scenes, lighting, and scale. . Among them, the change of lighting conditions is the most frequent, and it is also one of the factors that most affect the stability of face recognition. It has been proved in relevant literature that "the image difference of the same face under different lightin...

AI Technical Summary

Problems solved by technology

According to the principle of optical imaging, the Retinex algorithm and its improved algorithms, such as multi-scale Retinex (Mutli Scale Retinx, MSR), adaptive single-scale Retinex (Adaptive Single Scale Retinex, ASR), etc., use the reflectivity of objects to represent imaging objects, and decompose them into Components that have nothing to do with lighting can reduce the impact of lighting changes on imaging, but this type of method does not work well in the case of drastic lighting changes or side light sources

For example, the Retinex algorithm is prone to "halo phenomenon", and algorithms such as MESR and ASR do not consider whether the details in the image are lost

Discrete Cosine Transformation (DCT) is also often used in face image lighting preprocessing, which can better preserve category information that is not sensitive to lighting, expression, and posture, but this method is mainly used to eliminate the influence of one-sided lighting , when reconstructing an image based on DCT coefficients, a few low-frequency components are retained, and most high-frequency components are discarded, so that the resulting image loses some effective detail information compared with the original image

Xiaoyang Tan and William Triggs proposed a local texture enhancement algorithm (Tan and Triggs, TT) to eliminate the influence of illumination on face images, mainly including "grayscale gamma correction", "Gaussian difference filtering", "contrast equalization " and other steps, this method can effectively remove the influence of overexposure and shadows on the face image, and at the same time retain the basic elements such as the illumination change and detail features of the face, but in the case of side light sources, the light of the face image The boundary is also prone to "halo phenomenon". At the same time, since only the Gaussian difference method is used for filtering processing, the Gaussian difference filter is equivalent to a band-pass filter, which will cause some useful high-frequency information to be lost at the boundary of the face image contour. This will affect the accuracy of face recognition

Images