Enhanced Sketch Recognition via Ensemble Matching with Structured Feature Representation

Abstract

This research explores the use of advanced ensemble techniques including boosting, stacking, hierarchical ensembles, and Bayesian model averaging to improve design recognition performance. Using structural features and combining multiple classifiers, the proposed approach captures complex patterns and hierarchical relationships in design data. Experimental evaluations showed significant improvements against baseline methods, achieving classification accuracy of 92.5%, precision of 91.8%, recall of 90.6%, and F1 score of 91.2%. Boosting and stacking proved to be very effective in capturing complex data features, while hierarchical ensembles effectively handled layer dependencies. The averaging Bayesian model improved reliability by providing a robust uncertainty estimate. Challenges such as computational complexity and dataset balance issues were identified, along with recommendations to improve dataset ordering. These results highlight the potential of combined techniques in advancing contour recognition and provide valuable insights for future research and practical applications in computer vision and pattern recognition.