An Improved Handwritten Digits Recognition Using Histogram & ML Techniques

Abstract

This project presents an improved technique called Handwritten Digit Recognition System (HDRS) based on supervised pattern recognition. The proposed system aims to identify and recognize digits within handwritten numbers in grayscale images. The architecture consists of two principal phases: Training (Phase-I) and Digit Recognition/Testing (Phase-II). Phase-I contains three stages: pre-processing (improving quality using standard arithmetic operations), feature extraction (histogram technique over image blocks), and classification into 10 distinct classes. Phase-II introduces additional stages including segmentation (splitting images into non-overlapping blocks containing individual digits), mapping using a threshold method, pattern matching, and probability-based validation. Experimental results show that the proposed system achieves high accuracy, making it highly suitable for practical handwritten digit recognition tasks.