Application Of ICA And BT-SVM In Stereo Image Quality Assessment System

Now,3D technology has strong application prospects in many fields, such ascommercial, scientific, medical and so on. It becomes one of the hottest issues ininformation processing field. But, with the application of3D technology, theshortcomings and deficiencies of it gradually rendered in front of people. Forexample, after long-time viewing3D movies people feel uncomfortable such as soreeyes, dizziness and nausea. It is this discomfort and fatigue hampers the developmentand application of3D technology. Therefore, criterions assessing the every aspects of3D technology are urgently needed. Taking that into account, this article mainlystudies on developing a stereo image quality evaluation system that has adaptivecapability. This research provides a reasonable technique method for the developmentand application of3D technology.After introducing the details of significance, development status, human visualcharacteristics and3D imaging technology, it proposes a method for judging the gradeof stereo image, which uses support vector machine(SVM) based on statisticallearning theories and aims at the problems existing in quality assessment of stereoimage. Because that the data amount of stereo image is three times the data amount ofplane image, this article adopts principle component analysis(PCA) and independentcomponent analysis(ICA) to do reprocessing. It can deduce the data amount andextract a feature space. Then, map the data onto the high dimensional space, do gradeassessing using binary tree support vector machine (BT-SVM). In order to increasesystem identification accuracy, the separability method is adopted for determining thestructure of BT-SVM.There are100training samples and200testing samples in the experiment.BT-SVM method based on sample separability achieved the direct rate of92.5%. Toverify the performance of this method, the other two methods are given, adoptBT-SVM straightly and error correcting cods SVM(ECC-SVM), whose direct rateare lower than method proposed by this article. The result showed that method in thisarticle consists with human’s feelings, and has better generalization ability and lowertime and space complexity compared to the other two methods. Therefore, this articleprovides a better technical reference for stereo image quality assessment.