Pattern classification for assessing the quality of MPEG surveillance video
Tamer Shanableh, Faisal Ishtiaq · 2012
In this paper we propose the use of no-reference objective quality assessment to classify the quality of compressed surveillance video. The paper proposes a Macro-Block (MB) level no-reference objective Peak Signal to Noise Ratio (PSNR) classification based on pattern classification techniques. In the proposed system, the feature vectors are extracted from both MPEG coded videos and reconstructed images. The proposed feature extraction scheme is based on both the prediction errors of coded MBs and their prediction sources. The features are modeled using reduced multivariate polynomial classifiers, support vector machines and Bayes classifiers. The paper reports classification accuracy rates up 94%.