Discrete Entropy and Relative Entropy Study on Nonlinear Clustering of Underwater and Arial Images

For underwater and aerial images, the dispersing in atmosphere and the fluctuation in current flow are essential factors to consider. It is evitable that these types of images will be affected by uncertainties. As a result, image segmentation is especially useful for the processing of underwater and aerial images. Segmentation acts as a basic approach to clarify both feature ambiguity and information noise. It categorizes an image into separate parts which correlate with objects or areas involved. Image segmentation by clustering refers to grouping similar data points into different clusters. K-means clustering requires that the number of partitioning clusters be specified and its distance metric be defined to quantify the relative orientation of objects. Being a competitive learning method, winner-take-all (WTA) methodology has been selected to update one particular cluster centroid each time, which is an effective and optimal approach. K-means clustering is capable of both simplifying computation and accelerating convergence. To evaluate the role of image segmentation in image processing process, quantitative measures should be defined. The discrete entropy of the grayscale image is a statistical measure of randomness which can be used to characterize original and segmented images. The measure of the proximity between the probability density functions of the clustered and original images is described as relative entropy. Both measures are proposed to further study the influence of image segmentation via clustering. This study has the potential to apply on national defense and resource exploitation.