Circuit Design Based On The Illumination Normalization Algorithm Of Face Image

With the rapid development of social economy and technology,face recognition is widely used in various life scenarios.However,face recognition still faces various problems,lighting change is one of the key problems.At present,the illumination normalization algorithm for face images relies more on CPU implementation,and the processing speed is slow,which limits the application of real-time processing of embedded terminals.Therefore,the circuit design of the illumination normalization algorithm for face images has important significance.The main research contents of this thesis includes three parts:illumination normalization algorithm design,circuit design and FPGA verification based on face image.In the first part,An illumination normalization algorithm based on discrete Fourier transform is designed in this thesis to improve the recognition accuracy.Firstly,the discrete Fourier transform is used to transform the image from time domain to frequency domain,then the feature enhancement is performed in the high frequency part,and the histogram equalization operation is performed in the low frequency part,finally,the high and low frequency inverse transformed image is reconstructed.In the second part,the 64 parallel degree FFT accelerator is designed based on the fast Fourier transform algorithm principle,including complex multiplication module,data storage module,address generator module and 4-point FFT operation module.Based on the SMIC 28nm process library for functional simulation,synthesis and integrated simulation,the function of the accelerator circuit is verified.In the third part,the Zynq XC7Z020 series development board is used to build the verification system to verify the board level and test the image processing speed.The test results show that the recognition rate of the AR face standard recognition library after the illumination normalization algorithm designed in this thesis reaches 98.3%,which is greater than 6%compared with 91.8%.The circuit synthesis results show that the total power consumption of the circuit is 1.5874mW and the area is 0.245mm~2.At the clock frequency of 100MHz,the image processing speed for 64×64 reaches 100fps,which satisfies the requirement of processing speed greater than 20fps.The illumination normalization algorithm implemented in this thesis has a good processing effect and fast processing speed,which lays a foundation for the high-accuracy identification system in the future highways,gas stations and other scenarios.