| Resumen |
This paper proposes a novel algorithm for removing noise in graylevels and color images. The core concept on which is based the image filtering algorithm is a measure of spatial correlation between colors: we say that two colors are spatially correlated if they appear nearby more often in the image, with respect to other pairs of colors. In the discrete case, this measure of spatial
correlation is represented by an Adjacency matrix, which can be calculated efficiently. Firstly it is presented an iterative filter with a simple scheme of 4 neighbors and local weighted averages. Subsequently, we present a non-iterative autocalibrable filter and a strategy to estimate their parameters. Given its similarity in its implementation with the bilateral filter (FB), comparisons are made with this kind of filters using a current technique which estimates the parameters on the FB. It can be seen that the proposed filter generally produces better results than the FB. |
