MLP model uses fixed edge detection for handwritten character recognition

Researchers have developed a baseline model for handwritten character recognition using a simple multilayer perceptron (MLP) combined with a fixed Sobel-Feldman operator. This approach separates the edge extraction process from the learned classification, transforming images into horizontal and vertical derivative maps before feeding them into the MLP. The model achieved high accuracy, with 98.54% on the MNIST dataset and 92.50% on the EMNIST Letters dataset, demonstrating the effectiveness of first-order gradients in preserving discriminative structure while also highlighting specific ambiguities that arise from relying solely on edge geometry. AI

IMPACT Demonstrates the potential of simplified feature extraction for recognition tasks, offering a baseline for future research.

RANK_REASON Research paper detailing a new baseline model for character recognition. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Azam Nourisaeed
• MNIST database
• multilayer perceptron
• Sobel-Feldman operator

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →