Optimized Filtering With Binary Descriptor for Blind Image Quality Assessment (original) (raw)
Related papers
Unsupervised Blind Image Quality Assessment based on Multi-Feature Fusion
Image quality affects the visual experience of observers. How to accurately evaluate image quality has been widely studied by researchers. Unsupervised blind image quality assessment (BIQA) requires less prior knowledge than supervised ones. Besides, there is a trade-off between accuracy and complexity in most existing BIQA methods. In this paper, we propose an unsupervised BIQA framework that aims for both high accuracy and low complexity. To represent the image structure information, we employ Phase Congruency (PC) and gradient. After that, we calculate the mean subtracted and contrast normalized (MSCN) coefficient and the Karhunen-Loéve transform (KLT) coefficient to represent the naturalness of the images. Finally, features extracted from both the pristine and the distorted images are adopted to calculate the image quality with Multivariate Gaussian (MVG) model. Experiments conducted on six IQA databases demonstrate that the proposed method achieves better performance than the state-of-the-art BIQA methods.
BLIND IMAGE QUALITY ASSESSMENT WITH LOCAL CONTRAST FEATURES
The aim of this research is to create a tool to evaluate distortion in images without the information about original image. Work is to extract the statistical information of the edges and boundaries in the image and to study the correlation between the extracted features. Change in the structural information like shape and amount of edges of the image derives quality prediction of the image. Local contrast features are effectively detected from the responses of Gradient Magnitude (G) and Laplacian of Gaussian (L) operations. Using the joint adaptive normalisation, G and L are normalised. Normalised values are quantized into M and N levels respectively. For these quantised M levels of G and N levels of L, Probability (P) and conditional probability(C) are calculated. Four sets of values namely marginal distributions of gradient magnitude Pg, marginal distributions of Laplacian of Gaussian Pl, conditional probability of gradient magnitude Cg and probability of Laplacian of Gaussian Cl are formed. These four segments or models are Pg, Pl, Cg and Cl. The assumption is that the dependencies between features of gradient magnitude and Laplacian of Gaussian can formulate the level of distortion in the image. To find out them, Spearman and Pearson correlations between Pg, Pl and Cg, Cl are calculated. Four different correlation values of each image are the area of interest. Results are also compared with classical tool Structural Similarity Index Measure (SSIM)
Blind picture quality evaluation (BIQA) plans to assess the perceptual nature of a twisted picture without data with respect to its reference picture. Existing BIQA models normally focus the picture quality by breaking down the picture insights in some changed area, e.g., in the discrete cosine change area or wavelet space. In spite of the fact that incredible advancement has been made as of late, BIQA is still a troublesome assignment because of the absence of a reference picture. Considering that picture neighborhood contrast highlights pass on essential basic data that is firmly identified with picture perceptual quality, we propose a novel BIQA model that uses the joint bits of knowledge of two sorts of generally used elements: 1) the Gradient Magnitude (GM) and 2) the Laplacian of Gaussian (LOG) reaction. We utilize a versatile methodology to together standardize the GM and LOG components, and demonstrate that the joint measurements of standardized GM and LOG highlights have attractive properties for the BIQA undertaking. The proposed model is broadly assessed on three huge scale benchmark databases, and appeared to convey very aggressive execution with cutting edge BIQA models, and with some understood full reference picture quality appraisal models Keywords:Blind Image Quality Assessment, No Reference (NR), Gradient Magnitude (GM), Laplacian of Gaussian (LOG), Jointly Adaptive Normalization (JAN).