Abstract
In this paper, we propose an original contribution to improve the segmentation of multidimensional medical color images using the Hopfield Neural Network (HNN) classifier. In our previous work, the segmentation problem has been formulated as an energy function composed of two terms: the sum of squared errors and a noise term to help HNN in its minimization process of the segmentation problem's energy function to reach a local minimum close to the global minimum. Here, we demonstrate that considering the sum of a higher weighted error than the sum of squared errors leads the HNN classifier to go more deeply in the energy landscape. The proposed system is evaluated on 20 pathological liver color images and shown to be efficient and very effective in making crisp segmentation of the data set.