A blind system to identify and filter degradations affecting an image

For a high quality analysis and therefore a good pattern recognition, it is necessary to process degraded images. An image can be underexposed and degraded by noise or blur or a combination of them, or bright and degraded by noise or blur or a combination of them. The existing processing methods suppose that the nature of the degradations and some of the statistics are known. In this paper we propose a blind automatic procedure to (i) identify the nature of the degradation affecting an image and (ii) enhance and/or filter the images selected as dark and/or altered by a preponderant noise. The identification procedure is made of a sequence of three classifications, run by the non-parametric CHAVL algorithm, on different sets of statistics computed from the observed image. Considering the filtering aspect we propose a new method to estimate the standard deviation of the noise from histograms computed on homogeneous regions of any shape