Scholarlink

<p>Face detection remains a core challenge in computer vision, particularly for biometric and surveillance systems operating under limited computational resources. While the Viola–Jones (V–J) framework enabled early real-time detection, its robustness is insufficient for modern, unconstrained environments. This paper presents a hybrid, task-oriented extension of the V–J pipeline that integrates hierarchical CNN verification and an explicit Quality Assessment Module (QAM) to enforce biometric readiness. The proposed system combines multi-scale detection, quality-gated filtering, and multiplicative fusion of detection confidence and visual quality, while preserving the reject-fast philosophy of classical methods. Evaluation on WIDER FACE, FDDB, and AFW demonstrates a Conditional Detection Rate of 97.2%, a False-Positive Rate of 4.3%, and a throughput of 23 FPS. Results indicate statistical parity with contemporary deep learning detectors. Rather than proposing a new detector, this work reframes face detection as a quality-aware, computationally efficient and deployment-aware front end for biometric systems.</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

A quality-gated hybrid Viola–Jones pipeline for efficient face detection under computational constraints

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • R. Sathish,
  • S. Daniel Madan Raja,
  • E. Arul

摘要

Face detection remains a core challenge in computer vision, particularly for biometric and surveillance systems operating under limited computational resources. While the Viola–Jones (V–J) framework enabled early real-time detection, its robustness is insufficient for modern, unconstrained environments. This paper presents a hybrid, task-oriented extension of the V–J pipeline that integrates hierarchical CNN verification and an explicit Quality Assessment Module (QAM) to enforce biometric readiness. The proposed system combines multi-scale detection, quality-gated filtering, and multiplicative fusion of detection confidence and visual quality, while preserving the reject-fast philosophy of classical methods. Evaluation on WIDER FACE, FDDB, and AFW demonstrates a Conditional Detection Rate of 97.2%, a False-Positive Rate of 4.3%, and a throughput of 23 FPS. Results indicate statistical parity with contemporary deep learning detectors. Rather than proposing a new detector, this work reframes face detection as a quality-aware, computationally efficient and deployment-aware front end for biometric systems.