Feature extraction for high-resolution imageries based on the human visual perception

Abstract

The wide applications of high spatial resolution remotely sensed images are calling for more and more accurately classified imageries. However, feature extraction, as a significant processing in classification procedures, fails to fully extract the spatial features from high-resolution imageries, and that causes inaccuracy in various applications. On the basis of investigating and modeling the mechanism of human visual perception to take advantage of the excellent ability of understanding images, we propose a novel feature extracting approach in this paper based on the shape adaptive neighborhood (SAN), and present scientific analysis towards the approach. Firstly, we summarized the previous research on the spatial feature extraction for high-resolution images, as well as on the human visual perception. Then the concept of SAN was proposed to model the visual perception and was applied to extract spatial features from high-resolution imageries. Finally, experiments on a SPOT-5 imagery using the proposed approach will be conducted, to do the classification for the Land Use / Land Cover (LULC) application. Additionally, quantitative assessment and analysis were also given on the overall precision and the Kappa coefficient of the classification results. Experimental results show that the SAN-based feature extraction approach is of good help for improving the accuracy of classification by using both supervised and unsupervised methods. Especially, classification with unsupervised procedure is noticeably improved, which will greatly forward its application in specific cases.