Local Region-Based Hand Segmentation Using Deep Learning

  • Jeon, Junhyun; 
  • Kim, Taehun; 
  • Jeong, Yoosoo; 
  • Park, Kilhoum
Citations

SCOPUS

0

초록

Segmenting objects through the Digital Image Processing makes it difficult to draw high accuracy in a complex background. In this paper, we propose a method of using deep-learning segmentation to segment objects in complex environments, providing ROI(Region of Interest) of images as a pre-processing, and Thresholding in the post-processing. The proposed method is to detect the position of the hand in a complex image using a deep learning-based object recognition algorithm employing the YOLOv4 model; to expand the ROI so that the deep learning-based segmentation techniques using the U-Net model can be applied locally, not on the entire image; and to process by Thresholding through Otsu's Binarization method. We applied the proposed algorithm to hand images with complex backgrounds and verified the effectiveness of the algorithm by measuring the IoU values of the masks of correct answers and results. © 2023, Korean Institute of Communications and Information Sciences. All rights reserved.

키워드

Deep Learning; Hand; Object Detection; Object Segmentation
제목
Local Region-Based Hand Segmentation Using Deep Learning
저자
Jeon, Junhyun; Kim, Taehun; Jeong, Yoosoo; Park, Kilhoum
DOI
10.7840/kics.2023.48.9.1135
발행일
2023
유형
Article
저널명
한국통신학회논문지
권
48
호
9
페이지
1135 ~ 1143