Rotation Invariance in Transform Features for Handwritten Devanagari Character Recognition

摘要

Features extracted in any pattern recognition task, need to be invariant to translation, scaling and rotation. Handling translation and scaling in character recognition application is much easy compared to handling rotation in characters. The rotated characters affect the recognition rate. Therefore, it becomes crucial to extract features which are invariant to rotation in order to improve the recognition rate. This paper focuses on transform based features for extracting rotation-invariant features from handwritten Devanagari characters. The results indicate that Convolved Wavelet coefficient features are best at extracting rotation invariant features as compared to other transforms, considered in this work.