Multimodal continuous ant colony optimization for multisensor remote sensing image registration with local search
Introduction
Remote sensing image registration is one of the fundamental tasks in remote sensing image processing [3], [47]. Image registration aims to align two or more images into same coordinate system, and overlapping area is usually the area of interest [2], [34]. It is an important step for many remote sensing image processing procedures, such as image fusion [19], object recognition [26], change detection [13], etc. The performance of image registration has large influence on the performance of the follow-up procedure [4]. So the accurate result of remote sensing image registration is necessary [41].
In the literature, remote sensing image registration methods can be coarsely partitioned into two categories: feature-based and intensity-based [47]. Feature-based methods extract corresponding features (point, line and region), and obtain the geometric transformation through the matched features [45]. In [14], edge information is used to increase multitemporal image registration accuracy. And region feature can also be used for fine registration of very high resolution multitemporal images [15]. The most famous feature for image registration is scale-invariant feature transform algorithm (SIFT) [27], and a lot of corresponding remote sensing versions have been proposed, such as SAR-SIFT [6], PSO-SIFT [28], etc. These features play an important role in feature-based methods [20]. On the other hand, the intensity of the images is used as similarity measure in intensity-based methods [12]. The intensity-based methods usually contain two procedures: similarity measure and optimization algorithm [18]. The similarity measure is the key procedure in the intensity-based methods, because appropriate similarity measure directly influences the registration result. A lot of similarity measures have been proposed, such as sum-of-squared-differences (SSD), correlation coefficient (CC) [17], mutual information (MI) [29], and differential total variation (DTV) [22], [21]. MI is one of the most famous similarity measures, and it is widely used in image registration [23]. DTV is one of the best similarity measures, which is based on the differential total variation in the gradient domain [21]. In practice, due to the large differences of the imaging sensors, registration of multisensor remote sensing image is difficult [44]. Although some methods have been proposed, multisensor remote sensing image registration is still a challenging work [30]. As shown in Fig. 1, two multisensor images are registered through MI, and we can see that it is a multimodal problem [42]. There are a lot of local optima, and it is difficult to find the correct result. General optimization methods are usually designed for unimodal problem, which may be invalid for some multimodal problems. Multimodal optimization method is designed for multimodal problem, so it will be efficient and robust for multimodal problem.
Ant colony optimization (ACO), which is inspired by the foraging behavior of ant in the nature, is a novel heuristic approach in evolutionary computation [9]. Ants are able to find the shortest path from the food source to their nest over time by communicating with each other instead of visual cues [8]. Ants can deposit chemical information according to the distance of the path [36], and we call this chemical information pheromones [7]. Different from other evolutionary algorithms, ACO algorithm is a reactive search optimization method adopting the principle of “learning while optimizing” [43]. ACO is originally proposed for discrete problems, and has solved a lot of discrete problems effectively [32], [37]. Due to its high performance [1], [16], ACO algorithm has been extended to a continuous version () to solve continuous problems [38]. In , solutions are constructed through Gaussian kernel function. This solution construction strategy can maintain high diversity, and it is efficient for multimodal optimization [46]. Then, an adaptive multimodal continuous ACO algorithm (AM-ACO) is proposed, and it performs well for multimodal problems [46].
In this paper, multimodal continuous ant colony optimization is introduced for multisensor remote sensing image registration. ACO can preserve high diversity and is more robust for the problem which has many local optima [40]. ACO also can keep learning during the optimization procedure [25]. Meanwhile, multimodal continuous ACO can deal with difficult multimodal problems effectively [39]. These reasons all motivate us to propose a multisensor remote sensing image registration method based on multimodal continuous ACO. DTV [22], [21], which is one of the state-of-the-art registration methods, is used as similarity measure because of its robustness of intensity distortion. Meanwhile, we introduce the general DTV optimization method as an efficient local search operation according to the problem. Our method is evaluated on many multisensor images (optical images and SAR images). Experimental results show that our approach achieves the promising performance.
Section snippets
Image registration
In this paper, we focus on the parametric intensity-based registration methods [42]. Let R be the reference image, and S be the source image to be registered. Then, image registration problem can be formulated as:where C measures the dissimilarity between the reference image R and the source image S, T represents the parameter vector of the transformation model, and S(T) is the transformation image respect to T. The best transformation parameter vector is obtained when the
The proposed algorithm
In this study, the framework of adaptive multimodal continuous ACO (AM-ACO) is used as the global optimization method. AM-ACO is proposed to deal with multimodal optimization. It takes the advantage of ACOR in preserving high diversity and avoiding premature convergence. Then, the original optimization strategy of DTV is introduced as the local search operation.
Experimental study
In this section, we evaluate the performance of the proposed method on three sets of multisensor images (optical images and SAR images). We compare our algorithm with CLPSO (comprehensive learning particle swarm optimization) [24], DE [5] and ACO [23]. CLPSO is a famous version of PSO for multimodal problems. Our method is called LMACO, which is short for multimodal continuous ACO with local search.
Conclusion
In this paper, we introduce multimodal ACO algorithm for multisensor remote sensing image registration, and efficient local search operation is added. Due to the large differences of different imaging sensors, multisensor remote sensing image registration is a challenging work. General methods usually obtain local optimum, since it is a multimodal problem and sensitive to the initial position. Multimodal ACO algorithm can preserve high diversity and has the global search ability for multimodal
References (47)
- et al.
Performance analysis of the multi-objective ant colony optimization algorithms for the traveling salesman problem
Swarm Evolut. Comput.
(2015) - et al.
Differential evolution as a viable tool for satellite image registration
Appl. Soft Comput.
(2008) - et al.
Ant colony optimization theory: a survey
Theor. Comput. Sci.
(2005) - et al.
Multisensor image fusion using the wavelet transform
Graph. Models Image Process.
(1995) - et al.
Classification with cluster-based bayesian multi-nets using ant colony optimisation
Swarm Evolut. Comput.
(2014) - et al.
Ant colony optimization for continuous domains
Eur. J. Oper. Res.
(2008) - et al.
A hierarchical heterogeneous ant colony optimization based approach for efficient action rule mining
Swarm Evolut. Comput.
(2016) - et al.
Adapted ant colony optimization for efficient reconfiguration of balanced and unbalanced distribution systems for loss minimization
Swarm Evolut. Comput.
(2011) - et al.
Image registration methods: a survey
Image Vision. Comput.
(2003) - et al.
An automatic image registration for applications in remote sensing
IEEE Trans. Geosci. Remote Sens.
(2005)
A survey of image registration techniques
ACM Comput. Surv.
The effects of image misregistration on the accuracy of remotely sensed change detection
IEEE Trans. Geosci. Remote Sens.
Sar-sift: a sift-like algorithm for sar images
IEEE Trans. Geosci. Remote Sens.
Ant colony system: a cooperative learning approach to the traveling salesman problem
IEEE Trans. Evolut. Comput.
Ant system: optimization by a colony of cooperating agents
IEEE Trans. Syst. Man Cybern. Part B: Cybern.
A bare-bones ant colony optimization algorithm that performs competitively on the sequential ordering problem
Memet. Comput.
A cluster-based differential evolution with self-adaptive strategy for multimodal optimization
IEEE Trans. Cybern.
A novel coarse-to-fine scheme for automatic image registration based on sift and mutual information
IEEE Trans. Geosci. Remote Sens.
Change detection in synthetic aperture radar images based on image fusion and fuzzy clustering
IEEE Trans. Image Process.
Edge-based registration-noise estimation in vhr multitemporal and multisensor images
IEEE Geosci. Remote Sens. Lett.
Segmentation-based fine registration of very high resolution multitemporal images
IEEE Trans. Geosci. Remote Sens.
Intensity-based image registration using robust correlation coefficients
IEEE Trans. Med. Imaging
Cited by (57)
An interactive ACO enriched with an eclectic multi-criteria ordinal classifier to address many-objective optimisation problems
2023, Expert Systems with ApplicationsRobust registration for infrared and visible images based on salient gradient mutual information and local search
2023, Infrared Physics and TechnologyVLSG-SANet: A feature matching algorithm for remote sensing image registration
2022, Knowledge-Based SystemsCitation Excerpt :MI algorithm can measure the co-occurrence relationship between feature items and categories and is widely used in image registration. DTV is one of the best similarity measures, which is the total difference change based on gradient domain [16,17]. However, in practical applications, there are many local optima and it is difficult to find the correct result, so optimization algorithms are needed to solve these problems.
Designing a bi-level emergency medical centers' chain to increase the resilience of EMS's supply-chain using ACO-QAP: A case study of Tabriz
2022, International Journal of Disaster Risk ReductionParameter adaptation-based ant colony optimization with dynamic hybrid mechanism
2022, Engineering Applications of Artificial IntelligenceCitation Excerpt :Kumar et al. (2018) presented a hybridized regression-adaptive ACO for the path planning. Wu et al. (2019) presented a multimodal continuous ACO. Zhao et al. (2020) presented an adaptive reference point mechanism to modify decomposition-based many objective ACO to enhance the optimization ability.