Dehazed Image Quality Assesssment Based On Natural Scene Statistics

With the rapid development of multimedia and network technology, the distance communication among people is becoming more and more frequent. As an important way to obtain information, image is widely used in our work and life. People are also getting higher and higher requirements for image quality. So image quality assessment is following, including subjective quality assessment and objective quality assessment. Subjective quality assessment is operated by the observation. In order to ensure the accuracy of the subjective quality assessment, the assessment conditions are strict ruled, and the participation of the assessment staff should be as more as possible, which resulted in high cost and low efficiency of subjective quality assessment, more over the result of subjective quality assessment is influenced by the subjective factors largely. So the objective quality assessment has become the focus of the study. According to the degree of dependence on the reference image, the objective quality assessment is divided into three kinds of assessment methods, full reference, reduced reference and no reference.However, in the process of image acquisition, transmission, reading and storage, the degradation of image quality is exit. Especially, due to the impact of the haze, the degradation occurs from the process of acquisition in recent years. Therefore, technology of the image dehazing has attracted the attention of scholars, and has been significantly developed. The typical technology of the image dehazing is divided into two kinds, one based on image enhancement and another based on image restoration. There is no uniform and generally accepted assessment method for the effect of the image dehazing. Usually the objective quality assessment method of the image dehazing is based on a priori knowledge of the contrast of fog image is poor. For the phenomenon of color distortion in image dehazing, it can be used to evaluate the quality of the image by considering the two aspects of the combination of contrast and color.In this thesis, we refer to the BRISQUE which is a kind of no-reference quality assessment method based on natural scene statistics. In the BRISQUE method, the mean subtracted contrast normalized(MSCN) is extracted, and the distribution of the coefficient is fitted by the generalized Gaussian distribution. In addition, the distribution of the product of the adjacent MSCN, is fitted by the asymmetric generalized Gaussian distribution. According to the sensitivity of the feature parameters to various image degradation, the image quality is evaluated using support vector machine to classify.There will be a corresponding change for MSCN coefficient in the images before and after dehazing. However, it is proved that we can not obtain a satisfied result by using BRISQUE method to evaluate the quality of the image which is dehazed. In this thesis, we extract the parameters which are closely related to the image dehazing, as the characteristic parameters of the image dehazing quality evaluation by analyzing the parameters of BRISQUE. By quantifying the difference between the characteristic parameters of the image before and after dehazing, the effect of the image is evaluated. Through testing a large number of images, a rule is concluded whick can be used in the assessment of image dehazing.