Edge and region segmentation in high-resolution aerial images using improved kernel density estimation: A hybrid approach

Abstract

Edge detection and segmentation are the two main approaches being used since last three decades for successful image analysis in remote sensing domain. Although many intensive studies were undertaken, they all were limited to high-resolution aerial images and none addressed this problem exhaustively. The purpose of this study was to investigate both edge detection and segmentation by employing a novel hybrid method combining probability density function and partial differential equation to obtain accurate estimations. The newly proposed method is implemented in two phases: the first phase deals with smoothening that include improved kernel density estimation (KDE) with anisotropic diffusion coefficient function kernel with both adaptive bandwidth and constant threshold selection using Shannon entropy, in addition to a weighting parameter of 3 × 3 window for lower probability of the whole image in diffusion function; whereas in the second phase, edge detection and segmentation are dealt with by incorporating two prominent techniques, namely diffusion coefficient equation and six-sigma control limit. We carried out a cross-sectional analysis using different datasets such as SIPI database and ground truth images for smoothing, edging and segmentation. Afterward, the results were compared with the other state-of-the-art techniques. Finally, the performance measures of the implemented technique were evaluated by means of entropy, fractal dimension, and an equivalent number of looks for smoothened images, by the Pratt metric for edge detection, and in the case of segmentation, misclassification error was considered. The experimental results demonstrated that the proposed scheme outperforms its counterparts in all aspects. Hence, the proposed hybrid scheme is better and robust, and results in accurate estimation for the given datasets.