A new fitting criterion based on calculation of Pearson correlation between ordered data sample and ordered idealized data sample for normal distribution is proposed. It is shown that for image processing tasks where analyzing of data is carrying out in space of spectral coefficients of discrete cosine domain the proposed criterion considerably outperforms the fitting criterion χ2 that is most often used in practice. An advantage of the proposed criterion is invariance to value of noise variance in data sample.
fitting criterion, image processing, discrete cosine transform
“Korreliatsionnyi kriterii soglasiia dlia otsenki gaussovosti vyborok dannykh v zadachakh obrabotki izobrazhenii”,
Information Processing Systems,