A Robust Outlier Elimination Approach for Multimodal Retina Image Registration
This paper presents a robust outlier elimination approach for multimodal retina image registration application. Our proposed scheme is based on the Scale-Invariant Feature Transform (SIFT) feature extraction and Partial Intensity Invariant Feature Descrip
- PDF / 499,576 Bytes
- 9 Pages / 439.363 x 666.131 pts Page_size
- 112 Downloads / 218 Views
Institute for Infocomm Research, Agency for Science, Technology and Research, Singapore {epong,jalee,jcheng,benghai,xug,wkwong,jliu}@i2r.a-star.edu.sg 2 National Healthcare Group (NHG), Eye Institute, Tan Tock Seng Hospital, Singapore {laude_augustinus,stephen_teoh,tock_han_lim}@ttsh.com.sg
Abstract. This paper presents a robust outlier elimination approach for multimodal retina image registration application. Our proposed scheme is based on the Scale-Invariant Feature Transform (SIFT) feature extraction and Partial Intensity Invariant Feature Descriptors (PIIFD), and we combined with a novel outlier elimination approach to robustly eliminate incorrect putative matches to achieve better registration results. Our proposed approach, which we will henceforth refer to as the residual-scaled-weighted Least Trimmed Squares (RSW-LTS) method, has been designed to enforce an affine transformation geometric constraint to solve the problem of image registration when there is very high percentage of incorrect matches in putatively matched feature points. Our experiments on registration of fundus-fluorescein angiographic image pairs show that our proposed scheme significantly outperforms the Harris-PIIFD scheme. We also show that our proposed RSW-LTS approach outperforms other outlier elimination approaches such as RANSAC (RANdom SAmple Consensus) and MSAC (M-estimator SAmple and Consensus).
1
Introduction
Image registration is the process of aligning an image to another image taken from the same scene/object but in different situations [7, 10, 15]. Image registration is an important prior processing step before other processes such as image mosaicking [2], image fusion [7], and retina verification etc. can be carried out. Image registration approaches can be classified into two different types: unimodal and multimodal [7, 15]. Multimodal image registration approaches are designed for registering pairs of images from different imaging modalities, such as attempting to register a retina color fundus image with a retina fluorescein angiographic (FA) image. The purpose of such multimodal retina image registration is to assist eye doctors to automatically register fundus and FA images captured from the retina of the same person to enable the doctor to more easily diagnose eye problems and diseases. The registered fundus-FA image pair and their fusion can aid the doctor in planning treatment and surgery for their patients. © Springer International Publishing Switzerland 2015 N. Navab et al. (Eds.): MICCAI 2015, Part II, LNCS 9350, pp. 329–337, 2015. DOI: 10.1007/978-3-319-24571-3_40
330
E.P. Ong et al.
On the other hand, unimodal image registration approaches are designed for registering pairs of images from the same imaging modality. Generally, unimodal image registration approaches do not work on multimodal image registration. One of the better known multimodal image registration approach in recent year is the Harris-partial intensity invariant feature descriptor (Harris-PIIFD) approach proposed by Chen et al. [3]. In [3],
Data Loading...
