Color image gradients for morphological segmentation: the weighted gradient improved by automatic imposition of weights

In a previous paper some metrics were proposed to compute gradient from color images by exploiting the intuitive notion of dissimilarity among colors. The weighted gradient, one of the six methods introduced, provided the best segmentation results (following a subjective visual criterion) when applied in conjunction with the watershed from markers technique. Despite the excellent results achieved by the linear combination of the gradients from each band of the original image under the HSI color space model, the weighted gradient lacked a systematic method to give weights to each gradient. The goal of this work is to propose such a systematic method, by computing the similarity between the image to compute the gradient and an "ideal image", whose histogram has a uniform distribution. Some segmentation experiments were done and the automatic weighted gradient provides results as good as the manual one.