A Research About Stereo Images Quality Objective Evaluation Method Based On Statistical Learning

When compared with2D imaging technology,3D imaging technology alwaysprovides the end users with shocking and realistic spot experience, therefore it hasacquired a great concern in the field of research and application, and is widely used in3DTV, remote education, health care etc. In recent decades, researchers have investedlots of studies about3D imaging technology, these techniques greatly accelerate thepace of application of3D imaging technology in life. However some uncomfortablesymptoms such as dizziness, nausea often happen, so people urgently need a standardto evaluate the3D imaging technology. Due to stereo image content, terminal displayand viewer himself are all likely to cause the above symptoms, so in this paper wefocus on the stereo image content and design a stereo images objective qualityevaluation method based on statistical theory.After a full introduction of background and development status of stereo imageobjective quality evaluation system, human visual characteristics and other relatedknowledge, in this paper we proposed to respectively combine Support VectorMachine with Principal Component Analysis and Independent Component Analysis inorder to achieve an objective quality evaluation approach for stereo images. In thismethod, PCA and ICA are used to extract the second and higher order statisticalfeatures, removing redundancy and reducing dimension at the same time, then SVMbased on statistical theory, as a classifier, is used for judging the extracted features togive the quality of stereo images.In our experiment,371images are used including130images of different gradesas a training set, and the remaining241images of different grades as a testing set. Inaddition, we suggest to analysis system’s performance under different PCA and ICAfeature dimension. Experimental results show that when the feature dimension is12,the method of ICA combined with SVM has a highest recognition rate of93.36%,while when the feature dimension is18, that of the method of PCA combined withSVM is93.78%. By comparing the two methods, we finally find a more suitable onefor stereo images quality evaluation, and at the same time, a new thought about theapplication for PCA and ICA in3D imaging technology is provided.