Related Experiment Video
Updated: Mar 30, 2026
05:58
Digital Handwriting Analysis of Characters in Chinese Patients with Mild Cognitive Impairment
Published on: March 11, 2021
5.2K
Feature Set Evaluation for Offline Handwriting Recognition Systems: Application to the Recurrent Neural Network Model
IEEE Transactions on Cybernetics
|November 13, 2015
Summary
This study introduces a novel framework for evaluating handwriting recognition features using a collaborative approach. It quantifies feature set importance through combination weights, outperforming existing methods.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Handwriting recognition system performance relies heavily on extracted features.
- Existing methods for feature selection lack a systematic evaluation framework beyond simple recognition rate comparison.
Purpose of the Study:
- To propose a novel framework for evaluating feature sets in handwriting recognition.
- To quantify the importance and complementarity of different feature sets.
Main Methods:
- Utilized a weighted vote combination of recurrent neural network (RNN) classifiers.
- Modeled the combination in a probabilistic framework as a mixture model with two weight estimation methods.
- Benchmarked feature sets using RNN classifiers on Arabic (IFN/ENIT) and Latin (RIMES) script databases.
Main Results:
- Developed a method to quantify feature set importance via combination weights.
- Demonstrated that the proposed framework provides a competitive combination model.
- Established the first feature set benchmark for RNN classifiers in handwriting recognition.
Conclusions:
- The proposed collaborative framework effectively evaluates and quantifies the importance of feature sets for handwriting recognition.
- The combination weights offer insights into feature strength and complementarity.
- The system achieves state-of-the-art performance, offering a robust benchmark for future research.
