A study on multispectral image and video fusion with some applications

Abstract

Abstract In this thesis, we address different challenges in multispectral image and video fusion. Our first problem is on multispectral multifocus image fusion. A novel focus measure, based on steerable local frequency,is proposed.The proposed measure is shown to perform uniformly well across different spectra like visible, near infra-red and thermal. We further enhance the quality of the multispectral multifocus fusion by guided filtering and a graph based saliency model. The second problem is on causal multispectral video fusion. In this connection, we first solve the problem of causal video segmentation. An efficient causal video segmentation method is proposed using superseeds and graph matching. We then design a novel superpixel based causal multispectral video fusion method suitable for real-time surveillance applications. As a part of this solution, superpixel based spatio-temporal saliency model as well as superpixel based multiple fusion rules are developed. For the third problem, we consider an application of fusion in the domain of multi-biometric recognition. Here, we perform a signal level fusion of 2D and 3Dpalmprint data and apply a graph based recognition strategy. The proposed solution is shown to yield high recognition accuracy.