Face Image Illumination And Sharpness No-reference Assessment And Application

Face images always have different levels of degradation during the process of acquisition, compression and transmission. By face image quality assessment, it can effectively limit the low quality image input into the face recognition algorithm or system and improve the performance of face recognition. Obviously, no-reference face image quality assessment research is difficult, so it is also a hot research problem in the field of image quality assessment.The distortion of illumination and sharpness in face image has an influence on the performance of face recognition. The existing face image illumination quality assessment methods lack the overall consideration of light distortion such as light symmetry and light intensity. Furthermore, in terms of face image clarity quality assessment, evaluation methods based on specialty of human visual system are more conform to the human evaluation mechanism, and become a characteristic research direction. According to analyses above, this paper makes research on no-reference illumination assessment and sharpness assessment of face images.The main research works in this thesis are as follows:①Based on the survey of face image quality assessment development status, the thesis makes a research and analysis of the image subjective and objective quality Assessment. After introducing the face image standards which the International Organization for Standardization has passed and published, from two angles of illumination and clarity quality makes research on no-reference objective assessment of face image.②This paper research a no-reference illumination assessment method of face images. Firstly,the thesis analyses the illumination distortion of face images. Then, improving the universal image quality index(UQI) based on the facial symmetry, this paper proposed a pertinently weighted lighting symmetric assessment method. However, due to Illumination symmetry assessment can’t reflect the character of light intensity. So, through analyzing the gray histogram of illumination changing images, this thesis proposed a method to calculate the global light intensity of face images. Lastly, this paper proposes to fusion lighting symmetry with global illumination intensity by the form of product, realized face image no-reference illumination assessment as a whole. Experiments on some related face database verifies that the results of the proposed method have a consistency with the results of subjective assessment.③ This paper research a no-reference assessment approach for face image sharpness. After we clearly classify the definitions of face image sharpness and blur degree, and determining evaluation indexes of the performance of sharpness assessment method. This article research on a no-reference assessment approach for face image sharpness based on the specialties of human visual system(HVS) and edge detection. First, this measurement makes a luminance masking and space complexity masking to the face image according to the human visual system. Second, this method extracts the edge profile of face image under different resolutions by using wavelet transformation, and it uses the psychometric function and the Just Noticeable Blur(JNB) to extract the perceptual edge map at each resolution level. Amounting for the total number of perceptual edge pixels and the total number of perceptual blurred ones at each resolution level, the perception blur quality factor is defined as the proportion of them. Computing the weights which depending on the resolution level, we use the weighted average approach to compute sharpness quality score. Experiments on some related face database indicate that the results of the proposed method have a good performance in sharpness assessment of Gaussian blur and Defocus blur face image.④No-reference face image Illumination and sharpness assessment approach is used in face recognition. First, we use the no-reference illumination and sharpness image quality assessment which we have proposed to compute the illumination quality and sharpness degree, respectively. By setting quality score threshold according to the experiment, we expurgate the face image of poor quality. Then, we use the good quality face image as the input image of face recognition algorithms or systems. The face recognition experiments based on illumination and sharpness quality evaluation are conducted on the extended Yale B face database, CMU PIE face database and ***PSB face database. The experiment results show that face recognition import the proposed face image no-reference quality assessment methods can effectively improve the performance of recognition.