Recognizing Static Signs from the Brazilian Sign Language: Comparing Large-Margin Decision Directed Acyclic Graphs, Voting Support Vector Machines and Artificial Neural Networks

TL;DR

This study compares the effectiveness of large-margin decision DAGs, voting SVMs, and neural networks in recognizing static signs from Brazilian Sign Language, analyzing their performance, efficiency, and training heuristics.

Contribution

It provides a detailed comparison of classifiers for LIBRAS sign recognition, highlighting how different heuristics and decision schemes impact performance.

Findings

Significant differences in accuracy and efficiency among classifiers.

Hyperparameter surface maps reveal optimal settings for each method.

Neural networks trained with Resilient Backpropagation show competitive results.

Abstract

In this paper, we explore and detail our experiments in a high-dimensionality, multi-class image classification problem often found in the automatic recognition of Sign Languages. Here, our efforts are directed towards comparing the characteristics, advantages and drawbacks of creating and training Support Vector Machines disposed in a Directed Acyclic Graph and Artificial Neural Networks to classify signs from the Brazilian Sign Language (LIBRAS). We explore how the different heuristics, hyperparameters and multi-class decision schemes affect the performance, efficiency and ease of use for each classifier. We provide hyperparameter surface maps capturing accuracy and efficiency, comparisons between DDAGs and 1-vs-1 SVMs, and effects of heuristics when training ANNs with Resilient Backpropagation. We report statistically significant results using Cohen's Kappa statistic for contingency tables.