A Fuzzy Condition-Sensitive Hierarchical Algorithm for Approximate Template Matching in Dynamic Image Sequence

Abstract

Given a template of \(m \times n\) and an image of \(M \times N\) pixels, the latter being partitioned into blocks of \(m \times n\) pixels with interleaving, template matching aims at determining the best matched target block in the image with respect to the template. This chapter develops a hierarchical algorithm of template matching using decision trees. Nodes in the tree, here, represent the features used for matching, while the arcs denote the conditions on the features to separate relatively better candidate solutions from the rest. The proposed hierarchical matching scheme tests the feasibility of each block by checking the satisfiability of the conditions labeled along the arcs. The block that satisfies the condition at one level is transferred to the next level, and discarded from the system otherwise. Thus blocks that traverse the largest depth are better candidate solutions. Among these solutions, the one with the smallest Euclidean distance with the template is declared as the winner. The work differs with respect to classical hierarchical template matching by two counts. First, the conditions here are induced with fuzzy measurements of the features. Fuzzy encoding eliminates small changes in imaging features due to variations in lighting conditions and head movement. Second, information gain is used to determine the order of the features to be examined by the tree for decision making. The time-complexity of the proposed algorithm is of the order of \(MN/mn\). The algorithm has successfully been implemented for template matching of human eyes in facial images carrying different emotions, and the classification accuracy is as high as 94 %.