Research Of Full Reference Video Quality Assessment Based On Visual Characteristics

The methods of objective video quality assessment aims to use the various visual characteristics existing in HVS to construct an objective mathematical model evaluating the video quality,so as to achieve the maximum linear fitting effect with the subjective quality assessment results of the video.The objective full reference video quality evaluation method based on visual perception characteristics proposed in this paper is based on the visual threshold model and the characteristics of visual attention modeling and solving.The main research results are as follows:(1)Just Noticeable Distortion(JND)plays an irreplaceable role in Human Visual System(HVS).By analyzing and studying just noticeable distortion characteristics,this article proposed a new approach for calculating the temporal-spatial just noticeable distortion based on pixel domain calculation.The proposed method is based on singular value decomposition(SVD),which separates the useful information and noise in the images to calculate the just noticeable distortion value.Brightness contrast and texture masking are used to calculate the spatial just noticeable distortion value of the useful perceptible part;the just noticeable distortion model of the time domain of the perceptible information area is derived from the motion information.Combining the spatial domain just noticeable distortion and the time domain just noticeable distortion,the useful information partial model is obtained;the noise part of the image has simple structure and less information,and the noise part JND value is obtained by the weight parameter.The value of final overall just noticeable distortion is obtained by blending the JND value of noise and of the perceptible region parts.The experimental results confirm the performance of the proposed temporal-spatial just noticeable distortion model.Compared with other just noticeable distortion extraction algorithms,the noise-tolerance capability is improved by about 2 dB under the same conditions.Can be applied to many video processing areas such as video quality assessment or video coding.(2)A full reference video quality evaluation model that combines visual attention characteristics in the visual system with a threshold model is proposed.The method of solving the threshold of video JND proposed by(1)is used to judge the distortion in the video frame.The visual attention characteristic is used to dynamically adjust the accuracy of the video quality evaluation result by using the model of global just noticeable distortion,and it is also more suitable for the visual perception process of the HVS system.The visual attention characteristic is embodied as solving the video saliency,and the saliency value is regarded as a weight parameter,which is applied to the video quality assessment solution after JND judgment,aiming at local and dynamic adjustment of the solution accuracy and more conforming to subjective Evaluation process.The experimental results of linear fitting of objective evaluation results and subjective quality evaluation results show that the proposed video quality evaluation method has a significant improvement in performance the compare with the peak signal to noise ratio(PSNR)and other predecessor algorithms which can basically satisfy the needs of the video quality assessment.