A Hybrid Approach for Handwritten Character Recognition Using Stroke Width Variation and Transformer-based Feature Encoding

This project proposes a novel approach that combines stroke width-based seg- mentation and deep learning-based recognition for handwritten documents. The preprocessing employs a Feature-Driven Stroke Width Variation (SWV) technique for robust text segmentation. For recognition, a ResNet-based CNN extracts hierarchical features followed by a Transformer encoder to model sequential spatial dependencies. Experimental results demonstrate the effectiveness of this hybrid model in recognizing complex handwritten characters with varying stroke widths and distortions