Offline Tamil Handwritten Character Recognition Using Statistical Based Quad Tree
M. Antony Robert Raj, S. Abirami · SSRN Electronic Journal · 2016
This work presents a novelty with the combination of statistical and structural feature extraction algorithms which are represented in a hierarchical form by PM-Quad tree. In offline Tamil hand written character recognition, the feature extraction phase plays crucial role due to the nature of character structure and dependencies of writers. Tamil language contains naturally tough character structure and its complexity levels are enormous which includes more curves and loops, where the writers creating more difficulties when they are writing. This paper is concentrated on the locational analysis to locate place of the character structures. The originality of this works lies on the statistical analysis used for analyzing the pixel points available in the character image, this pixel values are hierarchically represented by the PM-Quad tree algorithm. This structure further simplified as vector values which suitable for classification algorithm. This features are evaluated with the classification algorithm Suppor Vector Machine (SVM) to predict the correct character. The statistical analysis, pixel densities and its locational factors are handled differently in tree format, where the feature caps are minimized by reducing the Quad tree length. This combination is bringing good recognition result when comparing our previous statistical algorithms.