Noise robustness of a new algorithm used for classifying pixels in color images by hierarchical analysis of bi-dimensional histograms

This paper evaluates the noise robustness of a new unsupervised algorithm used for classifying pixels in color images, by means of a hierarchical analysis of bi-dimensional histograms. Two types of additional noise are considered : gaussian noise and uniform noise, with varying correlation between the three colorimetric planes. Quantitative results show the influence of the noise level on the classification.