Intuitive Visualization and Objective Metric Evaluation for Software Programming Based Advanced Image Segmentation

This study is to investigate image segmentation approaches using software programming from diverse perspectives. The outcomes will be evaluated with intuitive visualization qualitatively and objective metrics quantitatively. Watershed segmentation is applied for pattern recognition. The watershed transform locates watershed ridges and catchment basins in an image, assuming that bright pixels are high and dark pixels are low on the surface. Segmentation can be improved using markers for foreground objects and background locations. Another approach to separate groups of objects is clustering. K-means segmentation assigns each object a space location via content-based clustering. Through K-means clustering using distance metric criteria, aerial images can be processed and analyzed. Matlab programming is a useful software tool for image processing while visualization effects of two approaches can be observed directly from visual appeals. To evaluate image segmentation objectively, quantitative metrics should be taken into account. Thus, the notions of the discrete entropy, gray level energy, relative entropy and mutual information are used to measure effects of image segmentation approaches.