Alphabet Classification of Sign System Using Convolutional Neural Network with Contrast Limited Adaptive Histogram Equalization and Canny Edge Detection

Abstract

Purpose: There are deaf people who have problems in communicating orally because they do not have the ability to speak and hear. The sign system is used as a solution to this problem, but not everyone understands the use and meaning of the sign system, even in terms of the alphabet. Therefore, it is necessary to classify a sign system in the form of American Sign Language (ASL) using Artificial Intelligence technology to get good results.Methods: This research focuses on improving the accuracy of ASL alphabet classification using the VGG-19 and ResNet50 architecture of the Convolutional Neural Network (CNN) method combined with Contrast Limited Adaptive Histogram Equalization (CLAHE) to improve the detail quality of images and Canny Edge Detection to produce images that focus on the objects in it. The focused result is the accuracy value. This study uses the ASL alphabet dataset from Kaggle.Result: Based on the test results, there are three best accuracy results. The first is using the ResNet50 architecture, CLAHE, and an image size of 128 x 128 pixels with an accuracy of 99.9%, followed by the ResNet50 architecture, CLAHE + Canny Edge Detection, and an image size of 128 x 128 pixels with an accuracy of 99.82 %, and in third place are the VGG-19 architecture, CLAHE, and an image size of 128 x 128 pixels with an accuracy of 98.93%.Novelty: The novelty of this study is the increase in the accuracy value of ASL image classification from previous studies.