No icon

SPSIM: A SuperPixel-based SIMilarity Index for Full-reference Image Quality Assessment

SPSIM: A SuperPixel-based SIMilarity Index for Full-reference Image Quality Assessment


Full-reference image quality assessment algorithms usually perform comparisons of features extracted from square patches. These patches do not have any visual meanings. On the contrary, a superpixel is a set of image pixels that share similar visual characteristics and is thus perceptually meaningful. Features from superpixels may improve the performance of image quality assessment. Inspired by this, we propose a new superpixel-based similarity index (SPSIM) by extracting perceptually meaningful features and revising similarity measures. The proposed method evaluates image quality on the basis of three measurements, namely, superpixel luminance similarity, superpixel chrominance similarity, and pixel gradient similarity. The first two measurements assess the overall visual impression on local images. The third measurement quantifies structural variations. The impact of superpixel-based regional gradient consistency on image quality is also analyzed. Distorted images showing high regional gradient consistency with the corresponding reference images are visually appreciated. Therefore, the three measurements are further revised by incorporating regional gradient consistency into their computations. A weighting function that indicates superpixelbased texture complexity is utilized in the pooling stage to obtain the final quality score. Experiments on several benchmark  databases demonstrate that the proposed method is competitive with state-of-the-art metrics.

Existing System:

Early FR IQA methods, such as peak signal to noise ratio (PSNR) and mean squared error (MSE), evaluate image quality based on intensity differences between reference and distorted images. In these two methods, only a numerical comparison is performed while the visual mechanism of humans is ignored.




Proposed System:

Namely, superpixel-based similarity index (SPSIM), to accurately predict image quality. In this method, images are segmented into visually meaningful regions, namely, superpixels. Then, the mean values of the intensity and chrominance components are extracted within each superpixel and compared to describe local characteristics precisely. This procedure is proposed to address the first problem. In addition to the two similarity measures above, gradient similarity is employed to improve the performance on structural variations. Furthermore, in each superpixel, the regional consistency of gradient magnitudes between reference and distorted images is measured. This measure focuses on the overall changes of all gradients in one superpixel and is used to improve the accuracy of the three similarities. This process aims to solve the second and third problems. Finally, texture complexity is utilized as local weights to pool the pixel-wise similarity map into a single score. The main contributions of our work can be summarized briefly as follows: 1) we use superpixels, which is perceptually more meaningful and accurate, to extract features and reflect image characteristics; 2) the regional overall variations in features are considered and utilized to revise feature similarity.

Comment As:

Comment (0)