Using texture to tackle the problem of scale in land-cover classification

Abstract

Object-Based Image Analysis (OBIA) is a form of remote sensing which attempts to model the ability of the human visual system (HVS) to interpret aerial imagery. We argue that in many of its current implementations, OBIA is not an accurate model of this system. Drawing from current theories in cognitive psychology, we propose a new conceptual model which we believe more accurately represents how the HVS performs aerial image interpretation. The first step in this conceptual model is the generation of image segmentation where each area of uniform visual properties is represented correctly. The goal of this work is to implement this first step. To achieve this we extract a novel complementary set of intensity and texture gradients which offer increased discrimination strength over existing competing gradient sets. These gradients are then fused using a strategy which accounts for spatial uncertainty in boundary localization. Finally segmentation is performed using the watershed segmentation algorithm. Results achieved are very accurate and outperform the popular Canny gradient operator.