Flexible Object Recognition: A New Approach toward Increasing Noise Tolerance in Contour Pattern Matching

Pattern recognition with computer systems presents a great challenge to researchers in computer science. In some cases patterns can be differentiated by their silhouette and recognized through the contour around its shape. In this study, shape is deliberately examined in terms of its flexibility, a measure defined to be the extent to which curves could be deformed. The proposed flexible object recognition (FOR) method widely employs flexibility to find corners, to form segments, to match segment pairs, and eventually to calculate the dissimilarity of contour pairs. Further analysis on classification with such dissimilarities shows rather high accuracy rates across various application domains, which may provide an evidence of the robustness for the proposed FOR method.