Fusion and classification of multi-source images by SVM with selected features in a kernel space

The objective of this study concerns the classification of a scene observed by different types of images, which generates large amounts of data to be processed. We have therefore chosen to use the classification SVM (Support Vector Machines) who is known for treating high-dimensional data. Although different sources of information can provide additional information to address the ambiguities, they introduce, at the same time, some redundant information. Our idea for the fusion of these data is to extract the useful information from all data to obtain an effective classification. The selection of the most discriminating features is carried out in the SVM kernel space, because the selection can be done linearly in this space. This selection also helps to reduce the size of data to be classified. The selection criteria are based on class separability. We propose a system based on SVM classification with the selection of characteristics to classify a brain tumor using three types of 3D MRI images. Our system can follow-up the evolution of a tumor along a therapeutic treatment.