IMAGE SEGMENTATION USING REMOTE SENSING

Abstract

Remote sensing is used for collection different information with different properties. In this paper, a methodology for remote sensing image segmentation, that utilizes each spectra land texture info. Linear filters square measure wont to offer increased spacial patterns. For every element location, we have a tendency to cypher combined spectral and texture options victimization native spectral histograms, that concatenate native histograms of all input bands. It has tendency to regard every feature as a linear combinationof many representative options, each of that corresponds to a phase. Segmentation is given by estimating combination weights, that indicate phase possession of pixels. It has a tendency to gift segmentation solutions wherever representative options square measure either noted or unknown. It has a tendency to conjointly shows that feature dimensions are often greatly reduced via mathematical space projection. The size issue is investigated, associated greedan rule is conferred to mechanically choose correct scales, that doesn't need segmentation at multiple scale levels.