Research On Illumination Problems In Face Recognition

Recently,face recognition technology has made enormous progress.However,the performance of the technology is affected by illumination condition,which is one of the main factors impairing the rate of face recognition.It is of great theoretical and practical significance to research the illumination problem in the face recognition technology.In this paper,the face illumination problem is discussed from two aspects,which are the algorithm of face illumination invariant feature extraction and the algorithm of preprocessing of face image illumination.(1)The algorithm of face illumination invariant feature extraction based on edge detection operator is proposedThe application of Sobel operator,Prewitt operator,Kirsch operator and Laplacian operator in edge detection operator are discussed,respectively.Two kinds of algorithm of face illumination invariant feature extraction are proposed The one is the algorithm of face illumination invariant feature extraction based on edge detection operator and divsion operation,and the other is the algorithm of face illumination invariant feature extraction based on logarithmic transformation and edge detection operator.The basic principle of the algorithm of face illumination invariant feature extraction based on edge detection operator and division operation is that the pixel of face image sharped with edge detection operator was divided by the corresponding pixel of original image.Thus,the lighting invariant feature is obtained and the slowly variation in the light part of face image is eliminated.The basic theory of the algorithm of face illumination invariant feature extraction based on logarithm transformation and edge detection operator is that the face image is sharped by edge detection operator in logarithm domain.Consequently,the details of facial feature are obtained and light components in face images are removed.The results of experiment on Extended Yale B database and CMU PIE database demonstrate that the algorithm can address the problem of illumination in face recognition and improve the recognition rate of face.(2)The algorithm of face illumination invariant feature extraction based on smoothing filter is presentedThe mean filtering method and the Gaussian smoothing filtering method are discussed in this paper,respectively.Two kinds of face illumination invariant feature extraction algorithms are proposed.The one is the algorithm of face illumination invariant feature extraction based on smoothing filter and division operation,and the other is the algorithm of face light illumination invariant feature extraction based on logarithmic transformation and smoothing filter.The illumination invariant feature is acquired by the method that the pixel of face image smoothed by utilizing smoothing filter are divided by the corresponding pixel of original face image in the algorithm of face illumination invariant feature extraction based on a smoothing filter and division operation.The illumination invariant feature is achieved by the method that the pixel of face image that processed through smoothing filter in logarithm domain are subtracted the pixel of the face image that smoothed via smoothing filter in the algorithm of face illumination invariant feature extraction based on logarithmic transformation and smoothing filter.The consequence of experiment on Extended Yale B database and CMU PIE database show that the algorithm can effectively eliminate the influence of non-uniform illumination on face image and ameliorate the rate of face recognition.(3)The algorithm of face illumination preprocessing based on image transformation is proposedIn this paper,the image transform techniques such as discrete cosine transform and wavelet transform are discussed.The image transformation technique is that discrete cosine transform,wavelet transform were combined with Wallis algorithm and gradientfaces algorithm to study the face illumination preprocessing.Four kinds of face illumination preprocessing algorithms are proposed,which are face illumination preprocessing algorithm based on discrete cosine transform and Wallis algorithm,face illumination preprocessing algorithm based on discrete cosine transform and gradientfaces algorithm,face illumination preprocessing algorithm based on wavelet transform and gradientfaces algorithm and face illumination preprocessing algorithm based on empirical mode decomposition and gradientfaces algorithm.The results of experimental indicate that the algorithm has robustness to face with different illumination condition and can effectively improve the rate of face recognition on the condition with complicated illumination condition.