SAR image segmentation based on quantum-inspired multiobjective evolutionary clustering algorithm
Introduction
Image segmentation is an important issue in computer vision and pattern recognition. An image is partitioned into multiple regions, where pixels in the same region are similar to each other in reference to some nature or computed properties, such as color, intensity, and texture, and neighboring regions are evidently different in reference to the same nature. With more and more remotely sensed image data, unsupervised classification technique will be very helpful and crucial for image understanding. Motivated by these, this letter suggests a clustering method for SAR images.
Among the existing clustering methods, K-means algorithm is the most popular and simplest one. However, the K-means [1] algorithm (KM) may get stuck at suboptimal solutions, depending on the choice of the initial cluster centers. Neural networks and evolutionary algorithms belong to neural processing technique which has been successfully applied to many areas such as the control design [2], [3], [4], machine learning [5], [6], pattern recognition [7], etc. As global optimization techniques, Evolutionary Algorithms (EAs) are likely to be good tools for clustering task. U. Manlik et al. [8] proposed a Genetic Algorithm-based Clustering technique (GAC) for the clustering problem. It has been demonstrated that the GAC provides a performance being superior to that of the K-means algorithm. Many clustering-based image segmentation methods have been proposed and studied so far. For example, Mohamad et al. [9] introduced a nonparametric unsupervised artificial neural network and hybrid genetic algorithm for satellite images segmentation. Xia et al. [10] proposed a clustering method for synthetic aperture radar (SAR) images based on a Markov random field by means of graph cuts to search the optimal clusters. Saha and Bandyopadhyay [11] studied a genetic clustering technique for automatically segmenting remote sensing satellite images. In the paper by Hasanzadeh and Kasaei [12], a size-weighted fuzzy clustering algorithm based on membership-connectedness and watershed transform to partition multispectral image was introduced. Most of the above unsupervised classification techniques used a single-objective function to access the goodness of the quality of partitions, which is very appropriate for particular type of data but has bad results for other types of data. Hence, it is necessary to simultaneously optimize several validity measures that can capture the different data characteristics.
Handl and Knowles [13] proposed an evolutionary multiobjective clustering approach with automatic k-determination (MOCK), discussing the conceptual and practical advantages of multiobjective clustering in detail. However, the applications of multiobjective clustering to image segmentation are seldom reported [14]. This may be due to the fact that it is difficult to apply current multiobjective clustering technology to image segmentation, owing to an extremely large amount of data needed to be handled.
In this letter, the segmentation problem using partition clustering is viewed as a combinatorial optimization problem. But the existing optimization methods, e.g., conventional genetic algorithms (CGAs), are often time-consuming, and their convergence speed is slow and easy to trap in local optimal value. Quantum computing is known for its principles of quantum mechanics such as uncertainty, superposition, interference and implicit parallelism [15]. These properties make it have better diversity and better trade-off between the exploration and the exploitation than common evolutionary algorithms. Some researchers have introduced the quantum principles into evolutionary computing. Han and Kim [16], proposed a quantum-inspired evolutionary algorithm for combinatorial optimization based on the concept and principles of quantum computing. Jiao and Li et al. [15] designed a quantum-inspired immune clonal algorithm for global optimization.
In order to overcome the defects of the single-objective clustering algorithms and solve the image segmentation problem more efficiently, we proposed a quantum-inspired multiobjective evolutionary clustering algorithm (QMEC). QMEC is based on the quantum computing and multiobjective optimization algorithm framework, which aims to search for the optimum clustering centre effectively, and locate edge position accurately to improve the performance of image segmentation. The novelty of QMEC lies in the following issues: (1) an effective, quantum-inspired, multiobjective, remote sensing image segmentation method is proposed, which has the strategies of quantum bit representation and better global optimizing ability by quantum rotation gate updating; (2) in order to obtain sufficient information for remote sensing image feature representation and discrimination, we constructed a fused feature set both by undecimated wavelet decomposition [17] and gray-level co-occurrence matrix (GLCM) [18].
Section snippets
Pre-processing works
First, some preprocessing will be done on the original images, including feature extraction based on undecimated wavelet decomposition and GLCM, and watershed raw segmentation. Evolutionary computation with population iteration on the level of pixels is usually very time-consuming because the number of pixels in a remotely sensed image is very large. Therefore, we use watershed algorithm to segment the original image into non-overlapping patches. Additionally, a proper features extraction
Quantum-inspired MultiObjective Evolutionary Clustering algorithm (QMEC)
In this section, we will give the detailed description about the proposed QMEC algorithm. As given in Fig. 1, the proposed QMEC is illustrated in brief.
The major steps of QMEC are described in the following subsections.
Experimental setup
We use one simulated SAR image and two remote sensing images with size to validate the proposed algorithm, and compare the results with MOCK [13], GAC [8], and KM [1]. All the algorithms use the same preprocessing, and 30 independent runs on each test image are performed. Experimental results are evaluated by two external indexes, the CCR and the index I [21]. For both of them, the larger the index value is, the better the performance is. Fig. 2, Fig. 3, Fig. 4 give the typical
Conclusion
In this letter, we proposed a novel image segmentation approach based on quantum-inspired multiobjective evolutionary clustering algorithm (QMEC). Principles of quantum computing are combined with multiobjective evolutionary algorithms to solve images segmentation problems. The watershed segmentation is employed to segment images into non-overlap small regions which can cut down the computation. To validate the performance of QMEC, we make the experimental study on one simulated SAR image and
Acknowledgements
This work was supported by the Program for New Century Excellent Talents in University (No. NCET-12-0920), the National Natural Science Foundation of China (Nos. 61272279, 61272282, 61001202 and 61203303), the Fundamental Research Funds for the Central Universities (Nos. K5051302049, K5051302023, K50511020011, K5051302002 and K5051302028) and the Fund for Foreign Scholars in University Research and Teaching Programs (the 111 Project) (No. B07048).
References (23)
- et al.
Genetic algorithm-based clustering technique
Pattern Recognit.
(2000) - et al.
Artificial immune multi-objective SAR image segmentation with fused complementary features
Inf. Sci.
(2011) - et al.
A k-means clustering algorithm
Appl. Stat.
(1979) - et al.
Decentralized output-feedback neural control for systems with unknown interconnections
IEEE Trans. Syst. Man Cybern., Part B, Cybern.
(2008) Adaptive backstepping dynamic surface control for systems with periodic disturbances using neural networks
IET Control Theory Appl.
(2009)- et al.
Globally decentralized adaptive backstepping neural network tracking control for unknown nonlinear interconnected systems
Asian J. Control
(2010) - et al.
Hybrid multiobjective evolutionary design for artificial neural networks
IEEE Trans. Neural Netw.
(2008) - et al.
Pareto-based multiobjective machine learning: an overview and case studies
IEEE Trans. Syst. Man Cybern., Part C, Appl. Rev.
(2008) - et al.
Adaptive evolutionary artificial neural networks for pattern classification
IEEE Trans. Neural Netw.
(2011) - et al.
Multicomponent image segmentation using a genetic algorithm and artificial natural network
IEEE Trans. Geosci. Remote Sens. Lett.
(2007)