Analysis of dct spectrum statistics for image processing applications
Tutkimustuotos › › vertaisarvioitu
|Julkaisu||Telecommunications and Radio Engineering|
|DOI - pysyväislinkit|
|Tila||Julkaistu - 2015|
Statistical characteristics of coefficients of discrete cosine transform (DCT) in 8x8 pixel blocks of grayscale images are analyzed. Robust estimators of distribution scale and tail heaviness are used in analysis since it is shown that the considered distributions are essentially non-Gaussian and they are better described by generalized Gaussian distribution (GGD) with variable parameters. Parameters of these distributions are determined for noise-free images and the same images corrupted by additive white Gaussian noise. It is demonstrated that the noise presence considerably changes distribution that approaches to Gaussian for high frequency DCT coefficients. Parameters of GGD model depend on both image properties and noise variance. The usefulness of the obtained results for two operations of image processing (variance estimation and filtering) is shown.