PCSCSM: Promise Correlation based Super Pixel Clustering with Similarity Measurement for Image Quality Assessment

Abstract

The comparison of extracted features from square patches are performed usually in the assessment of Full-reference image quality assessment. Visual meanings are not contained by this patches. A set of images pixels called super pixels, which is a meaningful as well as share common visual characteristics.  In image regions only, image pixels are meaningful. They indicates the requirement of regional quality assessment. Poor local quality is exhibited by large difference of local features in traditional Full-Reference (FR) method.

For commonly used features, this is not true. This work proposes a Promise Correlation based Super Pixel Clustering with Similarity Measurement (PCSCSM) method to resolve these issues. Quality of an image can be predicted accurately by this method. PCSCSM are based on designing a similarity measure (or distance) over the image pixels, such that records that are highly similar according to that measure are likely to be duplicates and image pixels that measure as significantly dissimilar are likely to represent different entities.

The superpixels are formed by segmenting image into a visual regions with meaning. Within every superpixel, mean value of chrominance and intensity are extracted. Local characteristics are described by comparing these mean values. In pooling stage, used a weighting function which indicates superpixelbased texture complexity to get a quality score. A single is obtained by mapping pixel wise similarity using texture complexity as local weights.