An Analysis of Subspace Methods for Large South Indian Datasets

Optical Character Recognition (OCR) is one of the important fields in image processing and pattern recognition domain. Handwritten Character Recognition has always been a challenging task. The complexity of accurate recognition of Multi Lingual South Indian Scripts makes its recognition a challenging task for the researchers. Multi Lingual characters are a challenging task because of the high degree of similarity between the characters. This paper presents an analysis of subspace methods for recognition of handwritten isolated Multi Lingual South Indian Scripts for the Kannada, Tamil, Malayalam languages. The study was carried out with a huge dataset containing 33,640 handwritten samples. The proposed method preprocesses the 841 different classes of characters obtained from scanned documents of the Multi Lingual South Indian Scripts for the Kannada, Tamil, Malayalam languages. Both Principal Component Analysis (PCA) & Fisher Linear Discriminant Analysis (FLDA) approaches are used to extract the features of characters. For classification Probabilistic Neural Network (PNN) approach is used with the combination of both PCA & FLDA feature extraction method. Based on classification of character the computed results performance of both PCA & FLDA based PNN classification was analyzed & discussed here.