Study On Image Quality Assessment Based On Perceptual Features

Human eyes are the final recipient of images. With the technical improvement of both software and hardware, we are now have increasingly high requirement on image quality. However, images are inevitably subject to different types of distortions in processing. Therefore studying the relationship between image distortion and quality, then judging the effect of the distortion on the visual perception, has important applications in practice.In this thesis, we study image quality assessment using perceptual features. Three effective algorithms are proposed for full-reference, no-reference image blur and general-purpose no-reference image quality assessment. The main contributions of this thesis are as follows.1. An image quality assessment algorithm based on multi-scale representation of structure is proposed. It is based on the finding that difference-of-Gaussian(DoG) can capture the structures of an image with flexibility and it is sensitive to image degradations. The DoG signals, computed based on scale space, are used to characterize the local structure changes in images. Image quality is estimated by evaluating the similarities of DoG signals at different octaves and scales. Information content weighting is also employed to adapt to the characteristics of the human visual system(HVS), based on which an overall quality score is obtained. Experimental results demonstrate that the proposed metric can evaluate the image quality accurately and generate quality scores highly consistent with human perception.2. It has been proven that sparse representation is consistent with the mechanism of human brain in processing visual information. In image sparse decomposition, the magnitudes of sparse representation coefficients are directly related to the sharpness of an image. Based on this observation, we propose presents a SPArse Representation based Image SHarpness metric(SPARISH). Natural images are first used to train an over-complete dictionary. For a test image, it is first divided into blocks, and the gradient and variance are computed. Then the blocks are decomposed using the over-complete dictionary, producing the sparse coefficients. Image energy is calculated based on the sparse coefficients. Finally, variances are employed to normalize the energy, producing the sharpness score. The experimental results demonstrate that SPARISH can generate sharpness scores highly consistent with human perception. It also outperforms the existing no-reference image sharpness metrics.3. Image distortions are typically characterized by degradations of structures. Dictionaries learned from natural images can capture the underlying structures in images, which are important for image quality assessment(IQA). Based on this observation, we propose presents a general-purpose no-reference image quality metric using a GRadient-Induced Dictionary(GRID). A dictionary is first constructed based on gradients of natural images using K-means clustering. Then image features are extracted using the dictionary based on Euclidean-norm coding and max-pooling. A distortion classification model and several distortion-specific quality regression models are trained using the support vector machine(SVM) by combining image features with distortion types and subjective scores, respectively. To evaluate the quality of a test image, the distortion classification model is used to determine the probabilities that the image belongs to different kinds of distortions, while the regression models are used to predict the corresponding distortion-specific quality scores. Finally, an overall quality score is computed as the probability-weighted distortion-specific quality scores. The proposed metric can evaluate image quality accurately and efficiently using a small dictionary. Experimental results demonstrate that the proposed metric can evaluate the image quality accurately and generate quality scores highly consistent with human perception.