首页    期刊浏览 2024年12月13日 星期五
登录注册

文章基本信息

  • 标题:Registro automático, preciso e robusto de imagens com RANSAC (Random Sample Consensus) adaptado para o descritor SIFT (Scale Invariant Feature Transform)
  • 本地全文:下载
  • 作者:Rubens Antonio Leite Benevides ; Kalima Pitombeira ; Jorge Centeno
  • 期刊名称:Research, Society and Development
  • 电子版ISSN:2525-3409
  • 出版年度:2022
  • 卷号:11
  • 期号:14
  • 页码:1-26
  • DOI:10.33448/rsd-v11i14.36631
  • 语种:English
  • 出版社:Grupo de Pesquisa Metodologias em Ensino e Aprendizagem em Ciências
  • 摘要:Image registration is a common problem in computer vision with several applications which consists of finding the correct transformation between pairs of overlapping images. This work aims to present an automatic and accurate model for image registration using the SIFT descriptor and the adapted RANSAC estimation method. The registration occurs through the estimation of the homography between the pairs of images, which use the point correspondences given by the SIFT descriptor. Matches are classified using a dynamically estimated error threshold. The analysis considers the dispersion of the residual over various error thresholds and adopts the one that minimizes the dispersion and the magnitude of the error. The model is tested with 8 heterogeneous pairs of images divided into two groups: 4 pairs obtained with a professional camera and 4 pairs obtained with a common camera. Due to the high quality of the images in the first group, few iterations of the model are necessary for a good estimate of the correct homography. In the second group, the model showed that it is capable of building mosaics between pairs of images with an overlap of less than 20%, finding exact correspondences between pairs of images regardless of the acquisition method. Furthermore, it was able to handle up to 65% corruption between matches, with a total execution time of a few seconds.
  • 关键词:Image registration;SIFT;Homography;RANSAC.
国家哲学社会科学文献中心版权所有