A robust non-local total-variation based image registration method under illumination changes in medical applications. (March 2019)

• Record Type: Journal Article
• Authors: Aghajani, Khadijeh; Yousefpour, Rohollah; Zohrehvandi, Mahbanou
• Is Part Of: Biomedical signal processing and control. Volume 49(2019)
• Page Start: 96
• Page End: 112
• Publication Date: 2019-03
• Subjects: Nonrigid image registration -- Spatially-varying intensity distortion -- Weighted total variation; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2018.11.001
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: In this paper, we propose a new efficient numerical method to improve geometric registration of image pairs when there exits locally varying intensity distortion. The proposed method does registration and intensity correction, simultaneously. We assume that the illumination changes in the images are smooth, so we use weighted total variation as a regularization term on the correction field in order to register the two images. Weighted total variation reduces the smoothness-effect on the coefficients across the edges. The proposed objective function contains l 1 norm as a data similarity term. Abstract: Independence of neighboring pixels and image stationarity are major concepts in conventional similarity metrics, used in image registration tasks. The accuracy of image registration decreases due to the presence of spatially varying intensity distortion in images. In this study, we hypothesized that changes in image illumination have limited total variation (TV). Accordingly, we introduced a similarity metric by reducing the weighted TV in the residual image. The primal dual method was then chosen to solve the proposed registration problem. The efficiency of the proposed method was compared to conventional methods, including the residual complexity (RC) method, the robust Huber similarity measure (RHSM), and the local linear reconstruction method (LLRM) which have been very successful in this field. The efficacy of the proposed method was confirmed by experimental findings on real-world and synthetic images.