Color-Based Segmentation of Geometrical Shapes Using the Modified PCNN

The pulse-coupled neural networks PCNN demonstrated the ability to perform a wide range of image-processing tasks, including segmentation. Image segmentation is crucial in a variety of applications. In this paper, the modified PCNN was selected, among the other PCNN models, to implement this process and segment geometric shapes. A set of graphical images were selected for testing, in which the selected shapes were isolated, partially overlapped, connection of similar shapes, connection of different shapes and orientation, and mosaic images. Also, real images, such as houses or vehicles, with geometric shapes, were tested. After testing wide ranges of the values of the PCNN parameters to find the most effective values, to keep them fixed at that values. The number of the iteration was the only parameter that was used to change the output, the maximum number of iteration for testing all the cases was less than 20 iterations. The results showed the potentials of using PCNN in segmenting geometric shapes with a specific color, even in images with very small details, and in few cases, more than one color were segmented. This method for segmentation is fast, with less complexity process, reliable for many types of images with geometric or symmetrical shapes, some of the tested images contain up to 10 different colors. This approach for segmentation can be useful if it is integrating into vision systems of industrial, or humanoids robots for proposes of focusing on a specific parts in scenes. Processing tasks, including segmentation. Image segmentation is crucial in a variety of applications.