Session:

Image Processing and Computer Vision
Tuesday, March 02, 2004, 11.50 – 12.10

Session Chair:

D. Linkens

Paper Title:

Mutual Information Restoration of Multispectral Images Using A Generalized Neighborhood Operation

Author(s):

H. Z. Rafi, University of Tehran, Iran
H. Soltanian-Zadeh, Medical Image Analysis Lab. at Radiology Department of Henry Ford Health systems Detroit, USA

Abstract:

Information theory based techniques for signal and image processing are now considered as a viable alternative to other popular techniques. This paper presents a new multispectral filter based on mutual information maximization to mutually restore multispectral images. For the sake of simplicity we consider only two multispectral images, but the idea can be generalized to more images. Since multispectral images contain analogous information about a scene, as a rule their mutual information is assumed to be maximal; but noise and other independent artifacts decrease their mutual information. A generalized neighborhood operation based on an alternative mutual information measure is used to increase the mutual information between the two neighborhood windows, sliding simultaneously on both images. The main feature of this generalized neighborhood operation is that it updates all pixels inside the neighborhood window. This filter does not assume any specific relation among the gray level intensities of images, and uses both inter-frame and intra-frame information to suppress noise. Application of the proposed method to simulated images shows the outperformance of this method compared with Perona-Malik method which has received much attention in recent years because of its capability in both noise reduction and edge enhancement.