Bibliography

(eng): We present a novel approach to description of a multidimensional image histogram insensitive with respect to an additive Gaussian noise in the image. The proposed quantities, although calculated from the histogram of the noisy image, represent the histogram of the original clear image. Noise estimation, image denoising and histogram deconvolution are avoided.We construct projection operators, that divide the histogram into non-Gaussian and Gaussian part, which is consequently removed to ensure the invariance. The descriptors are based on the moments of the histogram of the noisy image. The method can be used in a histogrambased image retrieval systems.