The Research On Image Quality Assessment

For humans, the image is an important source of information existing in people’s daily life.The images collected by different ways maybe not alike, some may be clearly visible, whileothers may be fuzzy, so the deviations of the human understanding for image information appear.During the process of image collection and transmission, the image may be polluted by differentkinds of noises and become fuzzy. Those noises information can be got timely if the imagequality evaluation module is added to the overall system. By this way, the system can beoptimized by parameters adjustment to get the better and more complete image information.Thus, it is necessary to study on the image quality assessment.Based on the theory human visual system (HVS), the paper mainly studies on how to usethese three objective assessment methods to evaluate digital images faster, better and moreeffective under different conditions. The main research contents and results are as follows ingeneral：1.The SSIM method based on regional weighted entropy(full-reference). this paperproposes a new image assessment algorithm which is based on the regional weighted entropycalled RESSIM. To solve the problems of the SSIM method in the noise and compressionartifacts in the image. Referring to the different interest rates of human eyes to the different partsof an image, I put the regional weighted entropy into image quality assessment which can betterreflect the image content of the location information. Experiment results show that the RESSIMconform the visual property of human eyes. Compared to the traditional objective imageassessment methods, RESSIM is better consistency with the subjective assessment.2.Image quality assessment based on human visual weight and singular valuedecomposition(reduced-reference). The traditional image quality assessment method was basedon pixels, which needed the complete reference image and neglected the image’s own structure,expanded the definition of image quality assessment. In order to solve these problems, the paperstudies on the ability of singular value vector to represent the image structure, fully analysis thecharacters of the human vision system such as light intensity, vein, fringe and so on, gives ancorresponding light intensity character computing model, then a new image assessment algorithm based on the human vision system combining with the singular valuedecomposition(BHSVD) is given.Compared with other subjective evaluation algorithms,thisalgorithm has better effect to evaluate the image quality.3.The mechanical parts quality examination based on no reference. It proves thatfull-reference or half-reference quality assessment method is infeasible because of the specificcharacteristics of the mechanical parts. In the first, this paper preprocesses the image of thedefective machinery parts such as the crack and gap ones and so on, then divides the images ofthose mechanical parts by the best threshold value, at last, the mechanical parts defective or notcan be determined according to the parts graphic attributes such as centrifugal rate. Thisalgorithm is simple and high accuracy.