Scholarlink

<p>In recent years, periocular region, the area of the face around the eyes, has received more attention in the development of soft biometrics. Although it does not uniquely identify an individual, it aids in reducing the search space and identifying distinct patterns, thereby providing additional information and enhancing recognition performance. Despite the fact that periocular patterns are unique, the use of contact lenses or eye diseases that alter eye color can complicate authentication process by obscuring feature patterns and altering inter-class and intra-class distributions. Therefore, this study proposes a method to mask out the ocular region (sclera and iris area) from periocular images, i.e., it automatically converts a well-known UBIPr.v2 dataset to a cross-eyed dataset. Experiments are conducted by training deep models with samples from the original dataset, as well as the generated dataset. A significant improvement in the accuracy of 99.67% is obtained with the proposed method while it is 99.30% with the actual dataset samples. Other statistical parameters are also measured and compared that proves the superiority of the proposed approach.</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

Cross-eyed dataset generation, simulation and evaluation using attention based residual module for gender identification

  • 期刊论文

  • 发表日期：

    2025年7月23日

查看全文
  • Gautam Kumar,
  • Ashish Ranjan,
  • Md. Shah Fahad,
  • Sambit Bakshi

摘要

In recent years, periocular region, the area of the face around the eyes, has received more attention in the development of soft biometrics. Although it does not uniquely identify an individual, it aids in reducing the search space and identifying distinct patterns, thereby providing additional information and enhancing recognition performance. Despite the fact that periocular patterns are unique, the use of contact lenses or eye diseases that alter eye color can complicate authentication process by obscuring feature patterns and altering inter-class and intra-class distributions. Therefore, this study proposes a method to mask out the ocular region (sclera and iris area) from periocular images, i.e., it automatically converts a well-known UBIPr.v2 dataset to a cross-eyed dataset. Experiments are conducted by training deep models with samples from the original dataset, as well as the generated dataset. A significant improvement in the accuracy of 99.67% is obtained with the proposed method while it is 99.30% with the actual dataset samples. Other statistical parameters are also measured and compared that proves the superiority of the proposed approach.