Superpixel-Based Two-View Deterministic Fitting for Multiple-Structure Data
This paper proposes a two-view deterministic geometric model fitting method, termed Superpixel-based Deterministic Fitting (SDF), for multiple-structure data. SDF starts from superpixel segmentation, which effectively captures prior information of feature
- PDF / 4,714,325 Bytes
- 17 Pages / 439.37 x 666.142 pts Page_size
- 3 Downloads / 189 Views
2
Fujian Key Laboratory of Sensing and Computing for Smart City, School of Information Science and Engineering, Xiamen University, Xiamen, China [email protected], [email protected], [email protected] School of Computer Science, The University of Adelaide, Adelaide, Australia [email protected]
Abstract. This paper proposes a two-view deterministic geometric model fitting method, termed Superpixel-based Deterministic Fitting (SDF), for multiple-structure data. SDF starts from superpixel segmentation, which effectively captures prior information of feature appearances. The feature appearances are beneficial to reduce the computational complexity for deterministic fitting methods. SDF also includes two original elements, i.e., a deterministic sampling algorithm and a novel model selection algorithm. The two algorithms are tightly coupled to boost the performance of SDF in both speed and accuracy. The key characteristic of SDF is that it can efficiently and deterministically estimate the parameters of model instances in multi-structure data. Experimental results demonstrate that the proposed SDF shows superiority over several state-of-the-art fitting methods for real images with singlestructure and multiple-structure data.
Keywords: Deterministic algorithm Feature appearances
1
·
Superpixel
·
Model fitting
·
Introduction
Geometric model fitting is a challenging problem in computer vision. A major problem in model fitting is how to tolerate numerous outliers, which are ubiquitous in the real-world. RANSAC [1] is one of the most popular fitting methods due to its robustness to outliers. Using the same random sampling technique as RANSAC, many robust fitting methods (e.g., gpbM [2], SCAMS [3], RCG [4] and PEARL [5,6]) have been proposed to improve RANSAC. There are also many robust fitting methods (e.g., SWIFT [7] and T-linkage [8]), developed based on different sampling techniques, during the past few decades. However, these fitting methods cannot guarantee the consistency in their solutions due to their randomized nature. As a consequence, the fitting results may vary if these methods do not sample a sufficient number of subsets. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part VI, LNCS 9910, pp. 517–533, 2016. DOI: 10.1007/978-3-319-46466-4 31
518
G. Xiao et al. Partition data points Generate hypotheses Select model instances
(a)
(b)
(c)
(d)
Fig. 1. Overview of the proposed method for homography estimation. (a) An image pair with keypoint correspondences. (b) Superpixel generation (each segment with the same color denotes a superpixel). (c) The procedure of the proposed method. (d) The fitting result according to the estimated model instances (the keypoint correspondences with the same color belong to the inliers of the same model instances). (Color figure online)
Recently, some deterministic methods (e.g., [9–13]) have received much attention for model fitting. In contrast to the unpredictability of non-deterministic fitting methods, these fitting methods can deterministically
Data Loading...
