Frequency Domain Digital Image Segmentation based on a Modified kMeans

Abstract

The segmentation of image is the basic thing for  understanding the images whether it is a color image or gray  scale image. It is used in the various image processing  applications, computer vision, etc. In this thesis work we have  used multiple clustering approaches to segment the image in  our initial step like Normalized cut, kMeans, and Mean shift. The main aim was to obtain feature extraction, to reduce  convergence, to reduce computation time, and to overcome  the over segmentation caused by the noise, also incorrect  spread of intensity. Hence the optimal solution has been  derived through the Modified kMeans through which the  feature extraction and the separation of overlapping objects  were evaluated by making use of wavelet transform and  computation time was reduced by considering approximation  band coefficients of DWT contribution in an image through  which overall performance was improved. Proposed work has  been implemented in MATLAB environment.