A bag of words description scheme based on SSIM for image quality assessment

Abstract

This paper addresses the need to use the knowledge about the human perceived quality, adding machine learning models to the objective quality estimation. A new technique is proposed based on the division of images into several cells where the mean of the SSIM metric is computed. A sliding window over a grid of cells that divide the image will define a set of image descriptors that are aggregated using a bag of words. This model is able to improve the typical values provided by SSIM and defines a new path for the application of machine learning to image quality evaluation.