Text Independent Writer Identification for Telugu Script Using Directional Filter Based Features
Chris Andrew Gadde, Santhoshini Reddy, Viswanath Pulabaigari, Umapada Pal · 2017
In this paper, a novel approach for writer identification from handwritten Telugu script is introduced. A new dataset of Telugu handwritten documents of 150 writers called the IIITS-THDB is created under this work for doing the experimental analysis and this dataset is made available publicly. Descriptive convolution based features are extracted from given handwritten text using directional filters. Feature reduction is performed to find the optimal linear subspace for the extracted features and identification is done using these reduced features. Smaller blocks of text with uniform textures are created using the available data that show an improved identification accuracy during experimentation. We also study the effects of noise and compare the results of our features with various other features over our dataset.