Spatial Masking in No Reference Mode

Abstract

Compression of images and video introduces artifacts like blockiness and blurriness. However blockiness is the dominant and common distortion which is resulted due to the discontinuities in the luminance level at DCT block boundaries. This abrupt luminance change is periodic and hence generates harmonics in frequency domain. The No Reference image quality meter is designed to determine the objective quality using the frequency domain approach. The blockiness is measured in frequency domain in which the ratio of harmonics to other AC components is calculated and used to estimate the blockiness artifact. The important part of the work is to mask the distortion in no reference mode, where the reference image is not available. The concept of self-masking is introduced to find the masking function from the coded image itself. At low distortion levels the masking obtained from reference and coded images will roughly be the same. So the masking from coded image is used to perform spatial masking in no reference mode. However, if the image is heavily distorted, the masking in full reference meter doesn’t play an important role. After the masking of coded image according to its spatial activity, the blockiness is calculated locally and accumulated in the end for a single quality metric. The results in the end show that this approach is very simple to implement and has easy calculations. The work is tested on images from LIVE image database and the Pearson’s Correlation Coefficient of 90% is obtained even without training on any image.