Study On No-reference Quality Assessment Method For Gray Image

Due to some facts such as the relative motion and out of focus, digital images are subject to various kinds of distortions during acquisition, compression, processing, transmission, and reproduction. Measurement of image quality is of fundamental importance to numerous image and video processing applications. In many of the assess methods, the subjective evaluation has been regarded for many years as the most reliable form of quality measurement. However, it is too inconvenient, slow and expensive for most embedded and real time applications, furthermore, the mathematic models can not be applied to it. The full-reference and reduced-reference image quality assessment assume the knowledge of a reference image. This assumption limits their application domain. No-reference image quality assessment can evaluate the image quality without any information of the original image. Therefore, the research on the no-reference image quality assessment is of important. The research targets in the paper are gray images.In view of the fact that the no-reference image quality assessment can not use the information of the original image, a design principle has been proposed in the paper. I.e., on the premises of satisfying the assessment accuracy, monotony and consistency, the image quality assessment must follow the qualified quality law, the HVS-based model must be introduced and all the measurement metric must be synthesized to give an overall result. In this paper, in accordance with this design principle, four kinds of no-reference quality assessments for gray image were designed and implemented from the three aspects (image contrast, blur and signal to noise ratio). In the image contrast metric, an image contrast model based on HVS was established. Based on the fact that people are sensitive to the region of interest, the weight factors for region of interest and un-interest were proposed. In the image blur metric, the brightness mask model based on HVS and spatial complexity mask model were introduced. A human eye sensitivity model for gray image was established due to the different sensitivity for human visual to different gray level. In the signal to noise ratio metric, a spatial frequency model of image and a matrix of the contrast sensitivity were proposed. An image noise detection model was given and two no-reference quality assessment methods based on HVS were proposed. According to the experiment, the assessment result obtained from the above four methods were proved to have a positive correlation with the objective assessment results and in consistent with the human visual characteristics. This work has laid a good foundation for further investigation on the no-reference image quality assessment.