A hierarchical algorithm for fuzzy template matching in emotional facial images

The paper aims at developing a hierarchical algorithm for matching a given template of m × non an image of M × N pixels partitioned into equal sized blocks of m × n pixels. The algorithm employs a fuzzy metric to measure the dispersion of individual feature of a block with respect to that of the template. A fuzzy threshold, preset by the user, is employed to restrict less likely blocks from participation in the matching. A decision tree is used to test the feasibility of a block for matching with the template. The tree at each link examines the condition for fuzzy thresholding for one feature of the image. If the block satisfies the condition, it is passed on to the next level in the tree for testing its feasibility of matching with respect to the next feature. If it fails, the block is discarded from the search space, and the next block from the partitioned image is passed on for examination. The process goes on until all the blocks pass through the decision tree. If a suitable block satisfies all the test conditions in the decision tree, the block is declared as the solution for the matching problem. The ordering of features to be examined by the tree is performed here by an entropy measure as used in classical decision tree. The time-complexity of the algorithm is of the order of MN/mn, and the elegance of the algorithm lies in its power of approximate matching using fuzzy conditions. 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 96%.