Bangla Compound Character Recognition using Support Vector Machine (SVM) on Advanced Feature Sets

A systematic approach for handwritten character detection and recognition using SVM has many significant applications. The proposed technique extracts features to classify the complex structural variations of Bangla compound characters. The three features used for classification are - Longest Run Feature (LRF) based on CG based Partitioning, Histogram of Oriented Gradients (HOG) Feature and Diagonal Feature. This enhanced, powerful combination of features result in a 180-length variable feature vector for each character, which is found to be adequate enough to uniquely represent and identify each character. Prior, Bangla handwritten character recognition problem has not been addressed with the proposed feature extraction techniques. The extracted feature vectors are used during the training phase for building a SVM classifier, testing phase for classifying characters which are then compared with their respective labels during evaluation. The results obtained show higher efficiency regarding classifier accuracy as compared to other existing approaches.