Research On The Infrared Image Quality Assessment Method

Infrared image will be degraded inevitably in the process of acquiring, storing and transmitting, which causes serious impact on infrared imaging guidance, airborne optical pod and video surveillance. Therefore, image quality need to be evaluated to obtain a sharp image in military and civil fields. In this paper, the establishment of infrared image database, infrared blurred image quality assessment, blur classification and parameter identification and hybrid quality metric for multiply distorted infrared images are described as follows.Firstly, in view of the lack of public infrared image quality assessment database, the infrared blurred image database and multiply distorted infrared image database are established. Then single-stimulus method is adopted to obtain the subjective scores of infrared images, which are used to testify the performance of infrared image quality metrics.Secondly, we do research on image sharpness assessment after analyzing the distortion types of infrared image. First, the reference image is generated by degrading the input image using Gaussian blur. Then, saliency regions of infrared image are obtained by combining local variance and local mean. Next, the singular value deviation of saliency regions of input image and reference image is calculated and a nonlinear mapping is adopted to compute parameter indexes. Finally, the performance of image quality metric is tested on visible and infrared image quality database and experiment results show that our method is highly competitive with state-of-the-art methods.Thirdly, in order to be convenient for the subsequent image restoration, blur classification method which utilizes the information in cepstrum is adopted to determine the existence of motion blur and out-of-focus blur. Then identification method of mixed blur is integrated to obtain blur parameters. Finally, the proposed method and existing algorithms are tested on image database and experiments show that our method has a higher accuracy.Finally, in order to accurately describe the distortion type of actual infrared image, we do research on synthetic quality assessment of infrared image which contains Gaussian blur and white noise. Then, the parameter correction method is proposed to improve the accuracy of noise estimation owing to the fact that existing algorithms ignore the influence of blur on noise estimation. Next, comprehensive quality score is obtained by normalizing the results of image blur and noise to the same interval. Finally, multiply distorted infrared image database is used to test the performance of different metrics.