| Popis: |
Word detection and recognition from natural scenes is very challenging and vastly popular research topic under the domain of computer vision. The problem comprises of various subproblems: character detection, optimal word formation from detected character, word recognition, etc. The paper focuses on optimal word formation from detected words only. Previously, pictorial structure model was used for optimal word formation where dynamic programming was applied. In this paper, chemical reaction-based meta-heuristics algorithm named as Chemical Reaction Optimization (CRO) has been used for word formation from detected character for larger instances. We have designed the solution generation, reaction operators and scoring function for the specific problem. Comparison with dynamic programming (DP) shows that DP generates optimal solution for small instances but for large instance it stuck due to memory limit exceed. CRO on the other hand has good optimality properties and better execution time than DP. |
