Research On Illumination Compensation Methods Of Face Recognition And Its Realization In FPGA

People mainly rely on the visual system to obtain the outside information,once the visual pathway gets damaged or lesions can cause the decline of the human vision,and even can lead to blindness.As a new kind of auxiliary equipment, the visual prosthesis can help patients generate phosphene and even come to see the light again by stimulate the visual nervous system.The blind can complete the task of recognition well in daily life under the assistance of the visual prosthesis,but with the limit of the changing illumination condition, the face recognition rate declines hardly,so how to eliminate the effects of illumination in face recognition has great significance.In order to overcome the problem of halo artifact on Retinex theory,a novel Retinex method based on bilateral filtering was proposed,then its hardware structure is presented ,which estimated the illumination by using the bilateral filtering and enhances the reflection image adaptively in the logarithmic domain.Compared with the classical SSR(Single-scale Retinex) algorithm, the MSR(Multi-scale Retinex) algorithm and the normal SSR based on the bilateral filtering on MATLAB simulation results show that the new method can eliminate the illumination effect on face more effectively under various illumination condition and can improve the robustness of illumination.The proposed algorithm has 26% higher than that of the classical MSR algorithm in average brightness,the contrast is improved by 23% and the average face recognition rate in face database is improved by 13%.It can have a better applicability in face recognition system.This paper completed the hardware design of the improved algorithm based on FPGA,and simulated the function of modules one by one.Then a platform was established on the board to verify the function of the algorithm.In order to verify the effect of different image processing results,this paper loaded different image data into the algorithm processing module through serial port,and displayed the processed image through VGA.At the same time ,compared the extracted image data captured by Signal Tap with the results of MATLAB software.The final result shows that the hardware implementation of the algorithm had a similar treatment with the MATLAB simulation results, it can eliminate the effect of illumination and enhance the details in dark area of the face image.