Research On Image Quality Assessment Of Video Surveillance

In recent years,with the high development of multimedia,video surveillance system has been developed at high speed.Many types of distortion are introduced in the processes of collection,compression and transportation,which can lead to a series of security issues.The quality of terminal image directly affects the accuracy of information.Accordingly,the evaluation of surveillance image quality is very important to our life.In practical application,it is hard to get original surveillance image.Thence the quality assessment of surveillance image often chooses no-reference algorithm.No-reference algorithm based on machine learning has been greatly developed in recently years,especially the development of deep learning makes the performance of related algorithms greatly improved.The research on related methods is divided into two directions: feature optimization and machine learning model optimization.In this article,we focus on the part of feature optimization to improve the performance.In this article,firstly,we define a new feature extraction method and introduce a new parameter.Then,we introduce the bag of words model to improve the diversity of features on different distortion types and multiple distortions.Finally,support vector machine is applied to model the relationship between image features and quality.According to the high correlation between text processing and image processing,we further introduce the topic model to solve the problem caused by multiple distortions.In reference to human visual system and the subjective evaluation standard of video surveillance image,we introduce visual attention mechanism and information entropy to improve the algorithm.We determine the type of concerned object relying on the monitoring scene and separate it by high-precision classifier.Referring to the differences of information entropy between the important areas and other regions,a new latent semantic analysis method is proposed.Finally,we use the support vector machine to build a high-performance effectiveness evaluation model.