Content-based image classification using PSO-SVM in fuzzy topological space

In this paper, the concept of particle swarm optimization (PSO) with support vector machine (SVM) in fuzzy topological space is proposed for classifying a given image. SVMs have their roots in statistical learning theory and have gained prominence because they are more accurate and reliable even if a limited number of training samples are used. However, the task of parameter selection in SVMs is quite tedious, as they require the selection of the kernel function parameter. In order to overcome this problem, the support vector machine is trained using the particle swarm optimization technique. In the proposed approach, this classification scheme is applied to the content selected for the classification. Moreover, in order to deal with the mixed pixel problem the input image class in the spectral space is decomposed into three parts: the interior, the boundary and the exterior in fuzzy topological space (fts). This decomposition is done using a threshold value based on the optimal intercorrelation coefficient. The interior-class pixels are classified using the PSO-SVM. The boundary-class pixels that are the fuzzy ones are reclassified on the basis of the connectivity theory of the fuzzy topology. The results of the experiment conducted on the remotely sensed image indicate that the proposed approach is more accurate than the existing fuzzy topology integrated support vector machine (FTSVM). Thus, it provides a better classification scheme.