Interest in fusing multiple sensor data for both military and civil applications has been growing. Some of the important applications integrate image information from multiple sensors to aid in navigation guidance, object detection and recognition, medical diagnosis, data compression, etc. While, human beings may visually inspect various images and integrate information, it is of interest to develop algorithms that can fuse various input imagery to produce a composite image. Fusion of images from various sensor modalities is expected to produce an output that captures all the relevant information in the input. The standard multi-resolution-based edge fusion scheme has been reviewed in this paper. A theoretical framework is given for this edge fusion method by showing how edge fusion can be framed as information maximisation. However, the presence of noise complicates the situation. The framework developed is used to show that for noisy images, all edges no longer correspond to information. In this paper, various techniques have been presented for fusion of noisy multi-sensor images. These techniques are developed for a single resolution as well as using multi-resolution decomposition. Some of the techniques are based on modifying edge maps by filtering images, while others depend on alternate definition of information maps. Both these approaches can also be combined. Experiments show that the proposed algorithms work well for various kinds of noisy multisensor images.
Image fusion, edge maps, information maps, noise filtering, multi-resolution, wavelets, Laplacian pyramids, multi-sensor images
