Implementation of Segmentation Techniques for Multispectral Satellite Images

Abstract

This paper reviews the wide range of Image segmentation techniques that are used to segment satellite images. The process of partitioning a digital image into multiple segments i.e. set of pixels is called segmentation. The pixels in a region can be similar due to some homogeneity criteria such as color, intensity or texture. A satellite image is a multispectral image that captures image data at specific frequencies across the electromagnetic spectrum. Spectral imaging can allow extraction of additional information the human eye fails to capture with its receptors for red, green and blue. Multispectral images are the main type of images acquired by remote sensing (RS).Since we use the LISS III the multispectral images lie in the band ranging from 0.62 to 1.78 micro meters. From the pool of segmentation methods available we consider the Kmeans, Fuzzy C means, marker controlled watershed, Wavelet, Level set, Quick Shift, Ant Colony Optimization for Thresholding techniques for segmenting satellite images. Several parameters such as mean, standard deviation, variance etc are considered in order to be used for comparison of the techniques.