Image Quality Assessment Based On Scene Classification

With the development of computer and multimedia, sceneunderstanding and image quality assessment have been the urgent problemto solve. The main work of this paper is divided into two main categories:scene classification and image quality assessment.Firstly, we extract traditional feature descriptors from the images, andencode these features with ScSPM methods, which is to reduce thereproducibility and improve the accuracy of classification. And then, wepropose a novel model called Ensemble Bayesian Networks which iscombination of Bayesian networks and random forests algorithm. Finally,with a large number of experimental verification, the Ensemble BayesianNetworks shows a good performance on classification. It has the advantagesof both Bayesian networks and random forests. The Ensemble BayesianNetworks can classify the database quickly and the relationships of differentvariables are visual. Also the method does not over fit while classification,and it can get the importance of different variables.Based on the scene classification, we present a novel method toevaluate the image quality. We can obtain the content of images by the sceneclassification. We use different target area extracting methods according to the content of the image. This paper uses different new global and localfeatures according to the different contents of images, to assess the qualityof images. The experiments show a high classification performance by ourfeatures.