An MLP Baseline for Handwriting Recognition Using Planar Curvature and Gradient Orientation

TL;DR

The paper addresses handwriting recognition by testing whether second-order geometric cues alone—curvature magnitude $|\kappa|$, curvature sign $\\operatorname{sign}\kappa$, and local gradient orientation $\\theta$—can drive an interpretable, lightweight MLP classifier. It introduces a practical pipeline to compute curvature and orientation maps from raster glyphs, vectorises them into a fixed 2352-D input, and trains a compact MLP that achieves 97% accuracy on MNIST and 89% on EMNIST Letters, rivaling a small CNN baseline. The work highlights the discriminative power of curvature-based representations for handwritten characters and argues that deep learning benefits can be realized with transparent, hand-engineered features. It also discusses limitations related to noise amplification and orientation wrap, and proposes concrete directions for improving robustness, representation, and scalability.

Abstract

This study investigates whether second-order geometric cues - planar curvature magnitude, curvature sign, and gradient orientation - are sufficient on their own to drive a multilayer perceptron (MLP) classifier for handwritten character recognition (HCR), offering an alternative to convolutional neural networks (CNNs). Using these three handcrafted feature maps as inputs, our curvature-orientation MLP achieves 97 percent accuracy on MNIST digits and 89 percent on EMNIST letters. These results underscore the discriminative power of curvature-based representations for handwritten character images and demonstrate that the advantages of deep learning can be realized even with interpretable, hand-engineered features.