Abstract:
Image quality assessment means estimating the quality of an image. Image quality is a
characteristic of an image that measures the perceived image degradation. The quality of image
gets affected due to the noise or distortion occurred during the acquisition, transmission, storage
and compression. Broadly, quality can be measured in two ways - subjective and objective. In
subjective methods humans are asked to rate the video on different scales according to the
perceived quality. Objective methods eliminate human involvement by determining the quality
of an input image automatically using some algorithm or mathematical model. With the
advancement of digital technology, assessing quality automatically becomes more important.
We propose a simple yet efficient objective quality assessment method, structural similarity
based on high order moments. SSIM is based on the assumption that human eye is capable of
extracting structural information by viewing the image and this structural information is a good
measure of quality. Loss of structural information is considered as loss of quality. We attempt to
extend the SSIM by incorporating shape parameters of distributions. Quality of image is loss if
shape of objects is not preserved. High order central and joint moments are used as shape
descriptors in our new approach. We show that a high order moment adds useful extra
information to SSIM, which is relevant in quantification of local structures. We also show that
this additional information improves the correspondence of SSIM with human perception.
Results are taken on various types of distorted images of a standard dataset and new SSIM is
validated against SSIM index, mean square error and subjective ratings.
