Fusion of visible and thermal images using support vector machines

Both in military and civilian applications, an increasing interest is being shown in fusing infra-red and visible images. In this paper, we propose a novel pixel-based infra-red and visible image fusion algorithm exploiting Discrete Wavelet Frame Transform (DWFT), Kernel Principle Component Analysis (K-PCA) and Support Vector Machine (SVM). Strong characteristics of DWFT such as translation invariant signal representation and directional selectivity add additional support to fusion process. K-PCA exploits the low frequency features mainly attributed from infra-red image, while SVM, on the other hand, exploits detail regions. Evaluations of the proposed technique through an image database show that the proposed method gives promising results both objectively and visually.