Optimization of Multiresolution Segmentation for Object-Oriented Road Detection from High-Resolution Images. Issue 2 (3rd March 2016)

• Record Type: Journal Article
• Authors: Saba, Fatemeh; Valadan Zoej, Mohammad Javad; Mokhtarzade, Mehdi
• Is Part Of: Canadian journal of remote sensing. Volume 42:Issue 2(2016)
• Page Start: 75
• Page End: 84
• Publication Date: 2016-03-03
• Subjects: Remote sensing -- Periodicals; 621.367805
• DOI: 10.1080/07038992.2016.1160770
• Languages: English
• ISSNs: 0703-8992

Abstract

Abstract : Abstract. This study aims to detect roads in high-resolution satellite images via an automatic object-oriented analysis. In this regard, an automatic image segmentation method is proposed to generate appropriate image objects. To achieve this goal, a genetic algorithm optimization with a new cost function is designed to set proper segmentation parameters, including the scale factor and the weights of the shape and compactness heterogeneities. The obtained image segments are then classified as roads or background features, using a fuzzy nearest neighborhood method. The proposed method is implemented on 2 pan-sharpened IKONOS images covering urban areas in the cities of Shiraz and Yazd, Iran. A comparison of these segments with those obtained using traditional manual methods proves the efficiency of this method. Moreover, the entire road detection system is compared with a support vector machine pixel-based classification; the proposed method improved the accuracy by 5%.