The advent of Internet and mobile technology and increasing the interest in multimedia services on wired and wireless networks lead to extensive use of encrypted and compressed images and videos. Since the final receptors of these Medias are humans, viewers’ comments about the video quality are very important. So the image and video quality assessment has become a big challenge. Video quality assessment is the key issues to improve the monitoring and control systems, improve methods for image and video processing, error concealment and many other applications. Subjective video quality assessment is a challenging problem since it is costly and time consuming. That is why in the recent years the development of the objective video quality assessment method is more considered. In this thesis a new method of video quality assessment is proposed in reduced reference scope which utilizes spatial and temporal entropic differences and comparison of statistical parameters of natural scenes. In this method a combination of statistical models and perceptual characteristics is used for designing the video quality assessment algorithm. Gaussian Scale Mixture (GSM) and Generalized Gaussian Distribution (GGD) are used to model wavelet coefficients of each frame and frame differences in order to measure different spatial and temporal information between original and distorted videos. Then the difference between spatial and temporal information is sent into a multi-layer neural network. This network receives spatial and temporal information difference as an input vector. Target vector includes subjective scores which is available from live (Laboratory for Image Video Engineering) video quality assessment dataset. Then the network weights are adjusted and trained. There are various methods to evaluate the efficiency of image and video quality assessment algorithms, including the correlation between subjective and objective data. The Spearman Rank Order Correlation Coefficient (SROCC) of predicted scores by the neural network and subjective score is equal to 0.8276 that in comparison with the STRRED and MOVIE algorithms improved considerably. In addition, feature vector is sent to a RBF network to cluster videos according to their quality using k -means algorithm. For justify; LINE-HEIGHT: normal; MARGIN: 0cm 2.45pt 0pt 0cm; unicode-bidi: embed; DIRECTION: ltr" Keywords: objective video quality assessment, Reduce Reference methods, Entropy, Gaussian Scale Mixture (GSM), natural scenes statistic, Generalized Gaussian Distribution (GGD), Neural Network.
