Comparison and research of threshold segmentation method in infrared target image

Comparison and research of threshold segmentation method in infrared target image

Abstract: This paper mainly takes infrared image target detection technology as the background, in the image threshold segmentation, based on maximum entropy criterion and genetic algorithm, the one-dimensional maximum entropy method (Ksw method) and three different threshold segmentation methods based on genetic algorithm KSW entropy method, such as single threshold value and double threshold value, are studied. The performance and the difference of them are verified by experiment simulation. The experimental results show that the KSW entropy method based on genetic algorithm can be used in infrared target detection system, and obtain good results, in order to verify whether it is universal, in other scientific fields such as: Coastal Wharf and other directions have also done the corresponding simulation experiments, the results are more satisfactory.

keywords : infrared target detection; threshold segmentation; Maximum entropy method; genetic algorithm

Study and compare the method of infrared target in the image threshold segmentation

abstract:  This article mainly takes the infrared image targets detection Techniques as the background, and based on the maximum entropy criterion and genetic algorithm in the image threshold Segm Entation, to study three different threshold segmentation method of the one-dimensional maximum entropy method (KSW method ) and KSW entropy method based on genetic algorithm of single Threshold, double threshold. And their performance and differences is verified by the experimental simulation. The experimental results show that KSW entropy method based on genetic algorithm of double threshold segmentation method, Can is used in infrared target detection system, and achieved good results. In order to verify it generality in and scientific fields, such as:the coastal wharf direction, we also make Correspon Ding simulation experiment, the result is satisfactory.

Keywords: infrared target detection; Threshold segmentation; The maximum entropy method; Genetic algorithm

1 Introduction (Introduction)

Image Segmentation is a basic problem in the field of image processing, and it is also a key technique in automatic target recognition (ATR), which is the basis of target feature extraction, recognition and tracking.

At present, in many image segmentation methods, the threshold method has been widely used for its simple, fast, practical and effective characteristics. How to choose a reasonable threshold to effectively separate the target from the background is the key problem to be solved by the threshold method, if the threshold value is too high, too many target points are mistakenly attributed to the background, and the threshold value is too low, the opposite situation will occur. At present, a great deal of research has been done on it. However, it has not been possible to find a threshold selection method to effectively segment all images.

Based on the infrared target image, the contrast between the target and the background is low, the edge is blurred, and in many cases the relative area of the target in the infrared image is small, so the traditional threshold method is difficult to get good results, and the target may contain more background information. In this paper, one-dimensional maximum entropy method (Ksw method) and three different threshold segmentation methods based on genetic algorithm of KSW entropy method, such as single threshold value and double threshold value, are studied, the performance of these three methods is verified by experiments, their merits and demerits are analyzed, and their advantages and disadvantages are tried to be applied to other scientific fields. The KSW entropy method based on genetic algorithm is used to show the satisfactory effect of two-threshold segmentation.

2 Introduction to Principles and algorithms (the Principle and algorithm introduction)   

2.1 Basic Threshold Segmentation principle

Set the original gray-scale image as a certain criteria in a gray value t as a threshold, the image is divided into two parts, then the segmented two value image is:

(1)

When it is generalized to a dual-threshold method, the segmented image is:

(2)

wherein, for the image of the two different thresholds of the split point, a segmented image of a grayscale value, the experiment set it to 180. Based on this, it can also be extended to three threshold or even multi-threshold segmentation, this paper only need to do double threshold, it is not described in detail.

2.2 The principle of maximum entropy value algorithm

The best entropy threshold method (Ksw entropy method), which is proposed by Kapur, does not require prior knowledge, and the image of non-ideal Shuangfeng histogram can be segmented. When the concept of Shannon entropy in information theory is used in image segmentation, the entropy of gray histogram is measured, and the optimal threshold is found, and the starting point is to make the information of the object and background distribution of the image the most.

Firstly, the single threshold segmentation method is introduced, according to the concept of Shannon entropy, the entropy measurement of the gray scale image histogram is as follows:

(3)

This is the probability that the first grayscale appears. Set the threshold value to divide the image into two parts, the target and the background, so

(4)

(5)

After the image is divided into two parts by the threshold, the probability distribution is as follows:

The entropy associated with each distribution is the following formula:

(6)

(7)

The total entropy of the image is:

(8)

The optimal threshold is to maximize the total entropy, i.e.:

(9)

It can be generalized to multi-threshold segmentation method, when there are multiple objects on the same background image or need to divide different levels of targets, multi-threshold is used. The optimal Entropy automatic threshold method is suitable for multi-threshold (set as threshold) segmentation. At this time

(10)

is the segmentation threshold, and there are, for the total entropy to take the maximum value, namely:

(11)

In particular, for double-threshold cases, that is, there are:

(12)

The optimal threshold is to maximize the total entropy, i.e.:

(13)

2.3 Ksw Entropy method based on genetic algorithm

Genetic algorithm (genetic algorithm) is a kind of stochastic search method which has evolved from the evolutionary law of the organism (survival of the fittest, the fittest genetic mechanism). It was first proposed by Professor J.holland of the United States in 1975, and its main feature is to operate directly on the structure object, there is no qualification of derivation and function continuity, inherent implicit parallelism and better global optimization ability, using probabilistic optimization method, can automatically obtain and guide the optimized search space, adjust the search direction adaptively, do not need Determine the rules.

Genetic algorithm is a search algorithm with the iterative process of "Generation + detection", and its basic processing flow is as follows:

(1) parameter encoding . This element is to establish a spatial mapping, that is, the problem of the solution space and coding space mapping, each of the different candidate solutions with a separate string number.

(2) setting of the initial group . The population is initialized.

(3) The design of the fitness function . In the population, the chromosome of each individual is decoded into a functional form suitable for computational fitness.

(4) select . The greater the adaptability of the individual as a good individual to reproduce the next generation, the greater the likelihood of the choice to breed the next generation.

(5) cross . Cross-swaps the two selected locations of the same individual where the next generation is propagated.

(6) variation . Flip the genes in the string according to the probability of a mutation.

(7) Repeat steps (4) to step (6) so that the result satisfies the genetic condition that has been set.

By applying this genetic algorithm to the above KSW entropy method, the following designs are available:

Design of genetic algorithm for single threshold segmentation of 2.3.1 KSW

(a) encoding : Because the image gray value between 0~255, so the individual chromosomes are encoded as 8-bit binary code, which represents a segmentation threshold value. The value of the initial generation population is randomly generated, and its corresponding fitness values are high and low.

(ii) population model : If the population is too large, the calculation of the fitness value per generation is significant, so the population should be set up reasonably. Here, the population is set to 10 and the maximum reproduction algebra is 100.

(c) decoding : The binary chromosome array is decoded to the value between 0~255, in order to its fitness value.

(iv) fitness function : the formula (8) is used as the fitness value function. At the same time, the linear calibration of the fitness function is taken.

(v) selection : According to the convergence theorem of genetic algorithm, the betting round method (Monte Carlo method) is adopted first, then the elite strategy is used.

(vi) crossover : An important feature of cross-swapping is that it can produce a different body from the parent. The greater the crossover probability, the greater the likelihood of cross-operation; If the crossover rate is too low, the convergence rate may decrease. Single-threshold segmentation because there is only one parameter, so a dot crossover, where the crossover probability is set to 0.6.

(vii) variation : Mutation probability is 0.1.

(eight) termination criteria : Specifies that when the algorithm executes to the maximum algebra (termination condition) or 15 generations of evolution, the maximum fitness value of the population is still unchanged (stable condition), the algorithm stops running, the individual with the highest fitness value is the segmentation threshold value.

Design of genetic algorithm for 2.3.2 KSW dual threshold segmentation

First, the code : The single threshold segmentation in the 8-bit binary code string is changed to 16 bits, the first 8 bits represents a threshold value, and the last 8 bits represents another threshold value.

Second, the population model : Two-threshold segmentation is a multi-parameter genetic programming, where the population is set to 20, the reproductive algebra is 100.

decoding : Decoding the binary chromosome array to a number between two 0~255 as a double-threshold value.

Iv. Fitness function : the formula (12) is used as the fitness value function. At the same time, the linear calibration of the fitness function is taken.

V. Choice : According to the convergence theorem of genetic algorithm, the betting round method (Monte Carlo method) is used first, then the elite strategy is adopted.

Six, Cross : The use of two-point intersection, randomly generated two intersection points in the first 8 and the last 8. The crossover probability is 0.6.

Seven, mutation : mutation probability is 0.1.

Viii. Termination Criteria : In the double threshold segmentation, it is stipulated that the maximum fitness value of the 30-generation evolutionary population has not changed to stable condition.

3 Experimental results and performance analysis (experimental results and performance analyses)

3.1 Experimental results (experimental results)

All the experiments in this paper are done on the computer of PC CPU @2.60ghz,4gram,intel (R) Core (TM) i5-3230m graphics card, the experiment environment is MATLAB r2014a, in order to verify the difference of threshold segmentation effect of three algorithms in this paper, In this paper, three images are selected for simulation, the original image is 3.1 (a) infrared aircraft image, 3.1 (e) of the infrared Palm image, 3.1 (i) Coastal Wharf images, the simulation process, the image is first converted to 256-level grayscale image, and then respectively, with three kinds of algorithms to simulate it, Get the image after the threshold is split, see the annotations for each image for details.

Figure 1 Infrared plane image threshold segmentation results

Fig. 1 Infrared plane image threshold segmentation results

3.1 (a) grayscale image of the original image 3.1 (b) One-dimensional maximum entropy thresholding Segmentation Image 3.1 (c) KSW entropy method based on genetic algorithm for single thresholding segmentation image 3.1 (d) dual threshold segmentation image based on genetic algorithm KSW entropy method

Figure 2 Infrared Palm image threshold segmentation Results

Fig. 2 Infrared Palm image threshold segmentation Results

3.1 (e) Grayscale image of the original image 3.1 (f) One-dimensional maximum entropy thresholding Segmentation image 3.1 (g) KSW entropy method based on genetic algorithm for single thresholding Segmentation image 3.1 (h) The KSW entropy method of dual thresholding segmentation image based on genetic algorithm

Fig. 3 Image threshold Segmentation results of coastal Wharf

Fig. 3 Coastal Terminal image threshold segmentation Results

3.1 (i) Grayscale image of the original image 3.1 (j) One-dimensional maximum entropy thresholding Segmentation Image 3.1 (k) Ksw entropy method based on genetic algorithm single thresholding Segmentation image 3.1 (L) Two-threshold segmentation image based on genetic algorithm KSW entropy method

3.2 Profiling (performance analysis)

From the above simulation image results can be seen: based on the genetic algorithm KSW entropy method double threshold segmentation algorithm in the image texture details compared to the first two segmentation methods show superior performance, through the simulation can be obtained above image segmentation threshold and search threshold value of the time spent, it is counted in the table 1 , the results are as follows:

Table 13 Performance Analysis Table of the algorithm

Table 1 Three kinds of algorithm performance analysis

	Figure 3.1 (a)			Figure 3.1 (E)			Figure 3.1 (i)
	Single/double threshold value		Search time (s)	Single/double threshold value		Search time (s)	Single/double threshold value		Search time (s)
Algorithm 1	40		0.173	126		0.165	149		0.656
Algorithm 2	76		0.020	115		0.027	149		0.103
Algorithm 3	41	165	0.070	76	172	0.154	127	210	0.168

(Note: Algorithm 1 is one-dimensional maximum entropy method, the algorithm 2 is based on the genetic algorithm of the KSW entropy method of single threshold segmentation algorithm, algorithm 3 is based on the genetic algorithm KSW entropy method double threshold segmentation algorithm. In addition, because of the intelligent simulation software, the same image in the same algorithm to obtain the data there is a subtle difference, so the above data is only relative)

As can be seen from table 1: As an optimization algorithm for image segmentation, genetic algorithm can greatly shorten the time to find the threshold, but the search time is longer because the double valve value is much larger than the single valve value in the search space.

4 Conclusion (conclusion)

In this paper, based on the maximum entropy criterion and genetic algorithm in the image threshold segmentation, the one-dimensional maximum entropy method (Ksw method) and the three different threshold segmentation methods based on the KSW entropy method, such as single threshold and double threshold, are studied, and their performances and differences are verified by experimental simulation. The experimental results show that the genetic algorithm, as an optimization algorithm, can greatly shorten the time of searching for the threshold when it is used in image segmentation, and it can be used in the infrared target detection system based on the KSW entropy method of the genetic algorithm, so as to verify whether it has universality, In other scientific fields, such as: Coastal Wharf and other directions have also done the corresponding simulation experiments, the results are more satisfactory.

5 References (Reference)

[1] Digital Image Processing, 3rd edition, Rafael C. Gonzalez and Richard E. Woods, Prentice Hall, 2010.

[2] Jiang Jianping, Peng Chunning. Two-threshold segmentation method for aircraft target infrared image sequence [J]. Ordnance, vol. 1.2004.

[3] IPOs, Wang Fangmei. Application of improved maximum entropy algorithm in image segmentation [J]. Computer simulation, 2011.

[4] species of Jinsong, Zhou Xiaoquan, Wang Hongqi. The best entropy thresholding image Segmentation method based on genetic algorithm [J]. Journal of Beihang University, 1999, (6): 747~.

[5] Liu Guohua, Baohong, Li Wenshu. Implementation of genetic algorithm program using MATLAB [J]. Computer Application Research , 2001.

Comparison and research of threshold segmentation method in infrared target image