Image search using multiresolution matching with a mutual information model

To search for a pattern or a template in a reference image is an essential step in a number of image processing applications. We often focus on searching for some parts of the original large image throughout the whole reference one for accurate position and orientation. In this paper, a multiscale matching approach based on the mutual information model is proposed. We also discuss the possible means to decrease the amount of computation in the search technique by limiting the entire set of possible locations in a small subset ones. The performance and results of the algorithm are demonstrated. The explicit application is automatic terrain navigation. Other applications involve image database querying, partial-content based indexing, and image positioning etc.