Surveillance Image/Video Quality Assessment

With the rapid technology development and cost reduction of theremote surveillance system, more and more surveillance systems are used tomeet the requirement of security in all aspects of human life. A number ofpotential problems emerge with the promotion of the surveillance system,among which the surveillance video quality assessment is the mostimportant one. In surveillance system, out-of-focus blur and blockingeffects obviously affect the image quality. If the problems of image qualitycan’t be detected, the performance of surveillance systemis affected. Today,the scale of surveillance system is very large. Normally, surveillance systemincludes hundreds of surveillance videos. It is impractical to hire hugeamount of people to assess the quality of surveillance videos. Therefore,Objective evaluation of the quality of the surveillance video has become anew research direction of surveillance system.Objective image quality measurement can be classified into threecategories,full-reference、reduce-reference、no-reference,bythedegreeofthe availability of an reference image. Based on the actual situation of theapplication, we choose no-reference objective assessment algorithm toevaluate the image quality. In this paper, we propose two kinds ofno-reference objective image quality assessment algorithms to evaluate thequality of the blockiness image. The first one is based on the feature thatnatural image contains a lot of redundant information. The redundantinformation can be picked out in distorted image for reconstructing thestructure information of the reference image. The quality of distortedimages is calculated by comparing the structure similarity between thestructure information of distorted images and reconstructed structure information. The second one is based on using the sample to optimize theexisting frequency quantization noise model. Then the spatial noisestatistics is calculated by the linear relationship between the optimizedfrequency quantization noise model and the spatial noise model. Finally, thestructure information of the reference image is reconstructed by the spatialnoise statistics and the distorted image statistics. The calculation of qualityof the distorted image is the same as the first one. At the same time, thispaper also proposes an algorithm about assessing the quality of the blurredimage. The basis of this algorithm is that the slope of the natural imageamplitude spectrum can describe the blur degree of the natural image. Inorder to overcome the shortcoming of the slope of the natural imageamplitude spectrum algorithm, a contrast perception model is added to theslope of the natural image amplitude spectrum algorithm.At end of this paper, the main information of an achieved real-timesurveillance video quality assessment system is de scribed. An accurateevaluation of the surveillance video is given by this system. So themanagement staff’s workload and the maintenance cost of surveillancesystem are reduced. In order to evaluate the performance of the imagequality assessment algorithm, the criteria of Video Quality Experts Groupand the world-renowned subjective image library are chosen. Theexperimental results demonstrate that the results of our algorithms have ahigh degree of consistency with the subject result. When compared with theother objective image quality assessment algorithms, our algorithms show acertain degreeof excellence. When evaluating the actual surveillance videos,our real-time surveillance video quality assessment system also shows ahigh degree of consistency with subjective evaluation.