Application of image intensity local variance measure for analysis of distorted images

Abstract

An image corresponds to a grid of pixels in digital image processing and the numerical value of each pixel represents image intensity. Characteristics and effects of image smoothing and other similar filtering operations, image enhancement, image distortion, etc. can be evaluated using local variance variation of image intensity. Image obtained after compression and blurring are examples of distorted images. Study on local variance variation due to JPEG and wavelet compression is compared in this paper. Ratio of expectation of local variance before and after compression is used to analyze the local variance variation due to compression. Expected value of local variance of DCT and wavelet coefficients is calculated by defining the respective distribution as Laplacian distribution. Expected value of local variance of compressed image is obtained based on uncompressed image details. Various levels of blurring is parameterized by the standard deviation of Gaussian filter used in blurring.