BROWSE
Detail
Conference Paper
Impact of Image Pre-processing on Feature Correspondences
Cited - time in 
Share 
Share
- Authors
- Muhammad Umer Kakli, Yongju Cho, Jeongil Seo
- Issue Date
- 2019-08
- Citation
- International Conference on Big Data Applications and Services (BigDAS) 2019, pp.1-4
- Language
- English
- Type
- Conference Paper
- Abstract
- Features matching is an essential and challenging step in a wide range of computer vision applications. The task becomes even more difficult if the matching image pairs suffer from distortions such as blurriness or uneven illumination. The performance of feature matching algorithms decreased drastically in such scenarios. Therefore, in this paper, we propose to pre-process the input images such that the maximum number of matching correspondences can be obtained. We first modify the image intensities by applying the Retinex theory based algorithm Low-light IMage Enhancement via Illumination Map Estimation (LIME). Next, scaling is applied to intensity modified images to further improve the number of matched correspondences. The proposed preprocessing scheme is tested with various images of different resolution, brightness, lightening, viewpoint changes, and structural information. The results demonstrate that the number of matched features and epipolar correspondences are significantly increased. On average, the number of matched features and epipolar correspondences are increased by ~144% and ~108%, respectively.
- KSP Keywords
- Computer Vision(CV), Feature correspondences, Feature matching, Features Matching, Image Pairs, Low-light image enhancement, MAP estimation, Retinex Theory, Structural information, Uneven Illumination, Wide range