Scholarlink

<p>In this research, a novel deep learning -based method called OCEAN (Otsu Combined Entity Aware Network) has been introduced to enhance face detection accuracy, particularly in the presence of occlusion. Initially, Adaptive Median Filtering is applied to the input occluded images to remove noise. The pre-processed images are then analyzed using the YOLOv7 network to detect occluded objects on the face. Next, the face entities on the occluded regions are segmented using the Otsu image segmentation algorithm. Angular Vector Projection Scaling is applied to these segments to correlate the entities using an Angular Vector Transformation Matrix based on the training dataset. This step helps identify variant features and non-variant hotspot variables from pixel variations in the occluded face objects. Finally, the face is detected using the YOLOv7 network based on the regenerated image and training samples. The proposed method enhances the accuracy by 5.18%, 4.21%, and 2.33% compared to YOLOv3, Tiny-YOLOv4, and YOLOv5, respectively.</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

Person Re-Identification Using Deep Learning Based YOLO Network with Partially Occluded Faces

  • 期刊论文

  • 发表日期：

    2025年2月10日

查看全文
  • G. Pandiya Rajan,
  • D. Manimegalai

摘要

In this research, a novel deep learning -based method called OCEAN (Otsu Combined Entity Aware Network) has been introduced to enhance face detection accuracy, particularly in the presence of occlusion. Initially, Adaptive Median Filtering is applied to the input occluded images to remove noise. The pre-processed images are then analyzed using the YOLOv7 network to detect occluded objects on the face. Next, the face entities on the occluded regions are segmented using the Otsu image segmentation algorithm. Angular Vector Projection Scaling is applied to these segments to correlate the entities using an Angular Vector Transformation Matrix based on the training dataset. This step helps identify variant features and non-variant hotspot variables from pixel variations in the occluded face objects. Finally, the face is detected using the YOLOv7 network based on the regenerated image and training samples. The proposed method enhances the accuracy by 5.18%, 4.21%, and 2.33% compared to YOLOv3, Tiny-YOLOv4, and YOLOv5, respectively.