Automatic Road Feature Extraction using MRF from LANDSAT-8 OLI Images

Abstract

Road detection from the remote sensing images is essential in automating map updation and town planning applications. In this paper, we propose an index based road feature extraction from OLI images using the Markov Random Fields (MRF) model. The proposed index is named as Normalized Difference Asphalt Road Index (NDARI) with a combination of two indices. These two indices are derived based on the reflectance differences of asphalt in the NIR, SWIR1 and Blue bands of OLI Images. They produce the low values for the asphalt features compared to other features in the image and results in roads that are darker than other features. To highlight the road features in the NDARI images, Mathematical Morphology (MM) i.e., Bot-Hat transform is used. Finally, image segmentation is done by using the MRF model. The parameters (mean and variance) are estimated by the Maximum-Likelihood Estimation (MLE) and Expectation-Maximization (EM) for each class label. The methodology is performed on the LANDSAT-8 OLI images and the results are presented, it is observed that the proposed method is producing the satisfactory results.