Handwritten Oriya Digit Recognition Using Maximum Common Subgraph Based Similarity Measures

摘要

Optical Character Recognition have attracted the attention of lots of researchers lately. In the current work we propose a graph based approach to perform a recognition task for handwritten Oriya digits. Our proposal includes a procedure to convert handwritten digits into graphs followed by computation of the maximum common subgraph. Finally similarity measures between graphs were used to design a feature vector. Classification was performed using the K-nearest neighbor algorithm. After training the system on 5000 images an accuracy of 97.64 % was achieved on a test set of 2200 images. The result obtained shows the robustness of our approach.