Scholarlink

<p>Sign language is a language used by deaf people to communicate with others. This language is not widely known by ordinary people; thus, it is essential to develop an approach that can recognize sign language to facilitate communication between deaf people and others. Developing a robust sign language recognition system is challenging due to the presence of many visually similar signs and the limited availability of datasets. Recent approaches to sign language recognition have faced difficulties, such as high computational complexity and low recognition accuracy. In this paper, we propose a novel method, dubbed ASLR-DPC, based on ConvMixer to recognize Arabic sign language. We fine-tune ConvMixer to create a simple yet effective approach for sign language recognition. This proposed ASLR-DPC uses patches of a sign language image and then applies depthwise convolutions, followed by pointwise convolutions, to mix spatial and channel dimensions. We examine the proposed ASLR-DPC using a public sign language dataset (ArSL2018), which consists of 54,049 images representing 32 different classes. The experimental results demonstrate the effectiveness of our approach, achieving an outstanding recognition accuracy of 98.69%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ASLR-DPC: Arabic Sign Language Recognition Based on Depthwise and Pointwise Convolutions

  • 期刊论文

  • 发表日期：

    2025年5月14日

查看全文
  • Mohammed Aloraini

摘要

Sign language is a language used by deaf people to communicate with others. This language is not widely known by ordinary people; thus, it is essential to develop an approach that can recognize sign language to facilitate communication between deaf people and others. Developing a robust sign language recognition system is challenging due to the presence of many visually similar signs and the limited availability of datasets. Recent approaches to sign language recognition have faced difficulties, such as high computational complexity and low recognition accuracy. In this paper, we propose a novel method, dubbed ASLR-DPC, based on ConvMixer to recognize Arabic sign language. We fine-tune ConvMixer to create a simple yet effective approach for sign language recognition. This proposed ASLR-DPC uses patches of a sign language image and then applies depthwise convolutions, followed by pointwise convolutions, to mix spatial and channel dimensions. We examine the proposed ASLR-DPC using a public sign language dataset (ArSL2018), which consists of 54,049 images representing 32 different classes. The experimental results demonstrate the effectiveness of our approach, achieving an outstanding recognition accuracy of 98.69%.