外文翻譯--基于優(yōu)化pso直方圖加權(quán)模糊c均值的一種快速水下光學(xué)圖像分割算法（英文）

1、J. Marine Sci. Appl. (2011) 10: 70-75 DOI: 10.1007/s11804-011-1043-8 A Fast Underwater Optical Image Segmentation Algorithm Based on a Histogram Weighted Fuzzy C-means Improved by PSO Shilong Wang*, Yuru Xu and Yongj

2、ie Pang National Key Laboratory of Science and Technology on Autonomous Underwater Vehicle, Harbin Engineering University, Harbin 150001, China Abstract: The S/N of an underwater image is low and has a fuzzy edge. If usi

3、ng traditional methods to process it directly, the result is not satisfying. Though the traditional fuzzy C-means algorithm can sometimes divide the image into object and background, its time-consuming computation is o

4、ften an obstacle. The mission of the vision system of an autonomous underwater vehicle (AUV) is to rapidly and exactly deal with the information about the object in a complex environment for the AUV to use the obtained

5、 result to execute the next task. So, by using the statistical characteristics of the gray image histogram, a fast and effective fuzzy C-means underwater image segmentation algorithm was presented. With the weighted hi

6、stogram modifying the fuzzy membership, the above algorithm can not only cut down on a large amount of data processing and storage during the computation process compared with the traditional algorithm, so as to speed

7、up the efficiency of the segmentation, but also improve the quality of underwater image segmentation. Finally, particle swarm optimization (PSO) described by the sine function was introduced to the algorithm mentioned

8、above. It made up for the shortcomings that the FCM algorithm can not get the global optimal solution. Thus, on the one hand, it considers the global impact and achieves the local optimal solution, and on the other han

9、d, further greatly increases the computing speed. Experimental results indicate that the novel algorithm can reach a better segmentation quality and the processing time of each image is reduced. They enhance efficiency

10、 and satisfy the requirements of a highly effective, real-time AUV. Keywords: underwater image; image segmentation; autonomous underwater vehicle (AUV); gray-scale histogram; fuzzy C-means; real-time effectiveness; si

11、ne function; particle swarm optimization (PSO) Article ID: 1671-9433(2011)01-0070-06 1 Introduction1 The ocean is rich in mineral resources, biological resources and energy. In the 21st century, human beings are facing

12、 the challenges of the three hot issues about the population, the resources and environment. As the space and resource on the land are limited, the ocean will gradually become the important national strategic objecti

13、ves (Xu and Xiao, 2007). A large number of coastal countries, especially western industrial developed countries and China are developing autonomous underwater vehicle (AUV) for exploring in ocean area and survey of s

14、ea-bed service. Underwater target detection, search and recognition in the three-dimensional space are the key to realize intelligent operation (Yuan et al., 1997). Therefore, the computer vision system is particularl

15、y important, and the image information processing capacity is the key to underwater vehicle dynamic sensing environment, fast locating and tracking object is also the fundamental mission of completing underwater surv

16、eys and AUV operations. Received date: 2010-08-07. Foundation item: Supported by the National Natural Science Foundation of China under Grant No.50909025/E091002 and the Open Research Foundation of SKLab AUV, HEU un

17、der Grant No.2008003. *Corresponding author Email: wangshilong@hrbeu.edu.cn © Harbin Engineering University and Springer-Verlag Berlin Heidelberg 2011 Underwater images are very sensitive to various noises and oth

18、er interferences, e.g. poor lighting conditions under the water will cause false details of underwater images, such as self-shadow, false contour, etc. As a focal light source, the searchlight makes the light intensit

19、y show a larger difference. The illumination is the strongest in the center and is gradually weakening along the radius, which leads to the uneven image background gray. Under the water, the visibility is low, the tr

20、ansparency is only one thousandth of the air and the water body itself is absorbent and scatters light, which will result in low S/N and fuzzy details. Meanwhile a variety of suspended particles scattering and absorpt

21、ion of the light wave (electromagnetic wave) in water will lead to serious gray effect on the captured underwater image. Moreover, the impact of water and shaking of the camera lens and other factors results in some

22、image distortion. What’s more, considering the image formation process, the image acquisition is a mapping of two-dimensional image from a three-dimensional image. Above all, it can be said that the image itself has

23、a strong ambiguity. The image segmentation is one of the classic research topics in computer vision research fields. Underwater image segmentation is the premise to image analysis, understanding and visual recognition

24、 technology of AUV, and is also one of Shilong Wang, et al. A Fast Underwater Optical Image Segmentation Algorithm Based on a Histogram Weighted Fuzzy C-means Improved by PSO 72iterative process and accelerates the conve

25、rgence speed. For an image ( n L H = × ), ( , ) f x y is the image gray value in the position of ( , ) x y , {0,1,..., 1} f l ∈ ? , and l the number of all gray series of the image. Define the histogram of a gray

26、 image as ( ) h j , 1 10 0 (j)= ( )H Lxy x yh j δ? ?= = ∑∑(4) where ( ) 1, ( , ) ( ) 0,else xyf x y j j δ = ? = ? ? and 0,1,2,..., 1 j l = ? . After standardization, it follows that ( ) ( ) h j p j L H = × . Usin

27、g the fuzzy index m as the weight 1/ ( ) ( ) m j p j ω = .Thus, a weighted histogram revising membership is: 1 1 2 1( ) 2 1( )( )c m i ij j k kj v u j v ω??=? = × ? ∑ ?(5) Corresponding cluster centers: 11( )( )l

28、m ij j i l m ij ju jvu==×= ∑∑???(6) The new objective function: 21 1 ( , ) ( ) ( , )c n m ij i i jJ u d j v= = =∑∑ ? ? ? U V(7) where ( , ) i d j v ?is the distance from j-pixel to i-clustering center. 3.2 Further

29、improving the fuzzy membership The value of the membership ij u ? of the new algorithm mentioned above is in the range between 0 and 1. This indicates that each data will belong to a class by different degree of membe

30、rship. The value of membership affects the convergence rate of iterative progress. So we here give a formula that can enlarge the maximum degree in the membership and further reduce the other membership. Set j-sample

31、belongs to o-category by the largest membership, and thus the improved expression for the degree of membership: (2 ) oj oj oj u u u = ? ? ? ? ? (8) (1 ) ij ij ij u u u = ? ? ? ? ? , 1,2, , ; i c i o =

32、 ≠ …(9) By the formula above, we can find that if the maximum degree of membership value is small, then the coefficient in front of it will be larger and on the contrary, the truth is also true. So the formula can meet

33、 the requirements that enlarge the maximum value of the membership degree and reduce other membership. 4 The fast HWFCM algorithm 4.1 Theory introduction Fuzzy C-means clustering algorithm is a local search algorith

34、m, and it is very sensitive to the initial value. The improper selection of initial value is likely to lead to convergence of local minimum point. In the solution space, the particle swarm optimization (PSO) algorithm

35、 is not restricted to one point, but a group of points simultaneously that can avoid falling into local solutions (Yao and Xu, 2007). So PSO is a kind of dynamic population-based evolutionary computing technology, sh

36、owing strong convergence, easy to implement with no more parameter settings (Higasshi and lba, 2003). Each particle in PSO is a solution in the solution space. It adjusts its flight by itself and the companion’s flyin

37、g experience. The best place that each particle experienced during the flight is the optimal solution of the particle itself, which is called individual extremum ( best p ), replaced by the extreme value of each class

38、( best c ) in this paper. The best position experienced by the entire group, that is, the optimal solution the entire group got at present, is called the global best ( best g ). Particle swarm optimization algorithm wil

39、l update particles’ speed and location by the following Eqs.(10) and (11). 1 1 1 best 2 2 best ( ) ( ) i i i i V V C r c X C r g X ω ? = + ? ? ? + ? ? ?(10) 1 i i i X X V η + = +(11) where 1 C , 2 C are constants, c

40、alled the learning factor; 1 2 , r r are random numbers between 0 and 1, w is inertia weigh, i V is the flight speed in the i-generation, and i X represents the particle position in the i-generation. Eberhart and Sh

41、i (2000) discovered through the research that in Eq.(10), when ω is large, the algorithm will have strong global search capability and whenω is small, the algorithm tends to local search. It is obtained that we can g

42、et a better search result through the iterativeω value from the maximum to the minimum, which first focuses on the global search and then enhances the ability of the local search. So, particle swarm optimization (PSO)

43、 described by sine function is introduced to HWFCM algorithm for the underwater image segmentation, in order to overcome the shortcomings that fuzzy C-means is over-reliant on the initial value and is easy to fall int