Abstract
This paper presents a novel approach to generate bag-of-words model based trajectory descriptions for handwritten strokes. We demonstrate how multiple distinct representations can be generated for the same stroke to accommodate writing variations and capture local features at stroke-segment level. The proposed descriptions can be utilized in template matching for handwriting recognition/correction, writer identification, signature verification, etc. The suitability of the proposed shape representations is experimented in a number of settings and used to build a language independent pen-based system for handwriting learning with feedback.