Scholarlink

<p>Face recognition has become a solid solution for uses like automated access control, surveillance, and identity verification without any human interaction. Although current methods offer great accuracy under controlled environments with frontal faces, their performance tends to degrade when confronted with partially visible, out-of-focus, or occluded facial images. In order to overcome these constraints, this research introduces a hybrid face recognition framework consisting of optimization-based improvement, generative restoration, and selective feature extraction. The pre-processing phase uses Particle Swarm Optimization (PSO) to refine the input image quality, followed by restoration from Generative Facial Prior Generative Adversarial Network (GFPGAN) for enhanced clarity of the degraded areas. In comparison with the conventional schemes that use the entire facial area, the suggested method focuses on the eyes, nose, and mouth as the primary region of interest (ROI), thereby supporting difficulties arising from occlusions due to masks, glasses, or weather conditions. Feature extraction is performed with Discrete Wavelet Transform (DWT), and the resulting feature set that has been reduced is analyzed using Convolutional Neural Network (CNN) for classification. Experimental results show that the proposed framework realizes a range of 90% to 95% recognition accuracy, better than traditional approaches but with significantly minimized computational overhead by feature set minimization.</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

PSO-GFPGAN-DWT-CNN: an optimized hybrid face recognition, restoration and classification framework

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • Reena Sharma,
  • Vijay Kumar Sharma,
  • Arjun Singh,
  • Vijay Shankar Sharma,
  • Bhanu Pratap Soni

摘要

Face recognition has become a solid solution for uses like automated access control, surveillance, and identity verification without any human interaction. Although current methods offer great accuracy under controlled environments with frontal faces, their performance tends to degrade when confronted with partially visible, out-of-focus, or occluded facial images. In order to overcome these constraints, this research introduces a hybrid face recognition framework consisting of optimization-based improvement, generative restoration, and selective feature extraction. The pre-processing phase uses Particle Swarm Optimization (PSO) to refine the input image quality, followed by restoration from Generative Facial Prior Generative Adversarial Network (GFPGAN) for enhanced clarity of the degraded areas. In comparison with the conventional schemes that use the entire facial area, the suggested method focuses on the eyes, nose, and mouth as the primary region of interest (ROI), thereby supporting difficulties arising from occlusions due to masks, glasses, or weather conditions. Feature extraction is performed with Discrete Wavelet Transform (DWT), and the resulting feature set that has been reduced is analyzed using Convolutional Neural Network (CNN) for classification. Experimental results show that the proposed framework realizes a range of 90% to 95% recognition accuracy, better than traditional approaches but with significantly minimized computational overhead by feature set minimization.