This document summarizes a research paper that proposes recognizing handwritten Odia (an Indian language) numerals using a single layer perceptron neural network. It first extracts gradient and curvature features from images of handwritten digits. These features are used to train a single layer perceptron classifier. The system achieves 85% accuracy on a dataset of 100 handwritten digit patterns written by 100 people. It aims to provide an efficient way to recognize Odia numerals using a non-linear classifier with reduced complexity compared to other methods.