A model based on local graphs for colour images and its application for Gaussian noise smoothing

Abstract

[EN] In this paper, a new model for processing colour images is presented. A graph is built for each image pixel taking into account some constraints on links. Each pixel is characterized depending on the features of its related graph, which allows to process it appropriately. As an example, we provide a characterization of each pixel based on the link cardinality of its connected component. This feature enables us to properly distinguish flat image regions respect to edge and detail regions. According to this, we have designed a hybrid filter for colour image smoothing. It combines a filter able to properly process flat image regions with another one that is more appropriate for details and texture. Experimental results show that our model performs appropriately. We also see that our proposed filter is competitive with respect to state-of-the-art methods. It is close closer to the corresponding optimal switching filter respect to other analogous hybrid method.