A New Algorithm for Detection of Vegetation and Shadow Regions in High Resolution Aerial/Satellite Images Based on Principal Component Analysis

Evaluation of vegetation cover by using the remote sensing data can provide enhanced results with less time and expense. In this paper, we propose a new automatic algorithm for detection of vegetation and shadow regions in high-resolution satellite/aerial images. It uses only color channels of the image and involves two modeling and evaluation phases. In the modeling phase, after extracting color and texture features for the pixels within a few manually-indicated vegetation areas, two distinct models are obtained based on the principle component analysis. In the evaluation phase, the same color and texture features are primarily computed for every pixel of the given image. Then, an error value is computed for each pixel which determines the model mismatch score. Finally, by thresholding the error image which includes all the error values, the vegetation and shadow regions are distinguished from the background. Experimental results demonstrated outstanding solution quality of the proposed algorithm compared to a number of counterpart algorithms.