Salient building detection in natural image using SVM

Abstract

This paper present a novel algorithm via support vector machine to detect the salient buildings whose height and many features make them stand out. Two-level Haar wavelet decomposition is implemented on the image to enhance the building candidates. And then the desired regions are separated from the background. A set of structure features is proposed to capture the generic statistic properties of the salient building using Sobel operator. The proposed approach has been tested on many real examples with good results.