Perceptual quality assessment is important for testing, optimizing and iecting relevant multimedia systems, services and algorithms. Current objective scores are not close enough to subjective ones. The purpose of this research is to use models based on the human visual system to promote objective quality assessment systems. In this regard, first, a binocular combination model for combining stereo pair images is proposed. Then, two no-reference and general-purpose objective quality assessment models are presented based on the feature extraction from these combined images. The first model uses the spatial domain features of combined images to train a stacked generalizer. The second model is based on sparse representation of combined images using pristine image dictionaries. These methods have increased the best correlations with subjective scores by up to 1% and 5% respectively. In the next step, two full-reference methods for quality evaluation of retargeted images are provided. The first method provides three groups of features to cover different types of retargeting distortions. The second method introduces a pyramidal integration model for different types of features, unlike most of the existing methods that just use global pooling. The proposed quality assessment methods provide rank correlation of up to 5% higher than the other methods. Despite the same efficiency, the second method is much faster than the first one and is fully applicable for real applications. Key words subjective quality assessment, Objective quality assessment, Perceptual quality, Stereo images, Retargeted images, Binocular combination, Spatial natural scene statistics, Sparse representation.
