Local texture descriptors and projection histogram based handwritten Meitei Mayek character recognition

Abstract

In this paper, we use the local texture descriptor and projection histogram feature for handwritten Meitei Mayek (Manipuri script) recognition. The different variations of the local binary pattern (LBP) namely, uniform LBP (ULBP), improved LBP (ILBP), center-symmetric LBP (CS-LBP) account for local texture descriptors. These features along with the projection histogram, separately and combined are presented to machine learning algorithms, k nearest neighbor (KNN), support vector machine (SVM) and Random Forest (RF) for classification of characters. The experiments by these feature descriptors with the classifiers have been evaluated on self-collected handwritten Meitei Mayek character dataset having 9800 samples. High accuracy is achieved even with the simple KNN classifier. Furthermore, classification with SVM and RF are explored, and the results are compared with the pixel-based methods which use the intensities value directly and a classic CNN model for recognition. The comparative results show that local texture descriptors and projection histogram strongly outperform pixel-based methods. The overall superior accuracy is achieved when the feature descriptors are combined with KNN classifier and performed even better than the CNN model.