SEGMENTATION is the first step in analyzing or interpreting an image automatically. It divides an image into its multiple segments which provides us a more meaningful understanding of the image. The level of subdivision needed for an image is purely depends on the complexity of the problem to be solved. Image segmentation leads to a set of all regions which cover the whole image. All the pixels of a particular region are related to each other with respect to color, intensity or texture and the regions adjacent to it are different with respect to the same. Image segmentation has different applications like face recognition and image retrieval, locating tumors etc.. It is very difficult to get a reliable and optimal segmentation. The algorithms of image  segmentation are mainly divided into two categories according to the intensity value of the pixels: similarity based approach, discontinuity based approach. The algorithms which follows discontinuity approach partition an image according to the sudden change in intensity level.  The algorithms which follows similarity approach partition an image according to the regions that are having pixels with some similar characteristics. Thresholding technique follows similarity approach. The simplest approach for image segmentation is thresholding. It can either be bi-level or multilevel category. Bi-level thresholding separates the objects from the background of an image by a single threshold value. Multilevel thresholding segments a grey level image into different distinct segments. For this, it determines more than one threshold value to create output images having multiple segments.

The thresholding technique uses artificial bee colony which is an optimization technique proposed by D. Karaboga[1]. It is influenced by the foraging nature of honey bees. The most important advantage of ABC is that it is very simple to implement and there are very few control parameters. A survey is conducted to elaborate the benefit by using ABC to get threshold value for image segmentation.

Artificial Bee Colony

Swarm intelligence is based on both the decentralized and self-organized behavior of swarms like fish schools, bird flocks and the colony of social insects such as termites, ants and bees (Ma et al. 2011). ABC is modeled to simulate the foraging behavior of honey bees. The swarms of real honey bees work in a smart way collectively for searching the food. Honey bees have some special qualities like they can store information in memory and share them with others and they also can take decisions depending upon that. Due to this intelligent behavior, researchers are attracted to give attention to the foraging activity of the honeybees.

In ABC, the community of bees is mainly categorized into 3 different groups: employed bees, onlooker bees, and scouts. In ABC the required solutions are represented as the food sources of honey bees. A bee searches for a specific food source when needed and then exploit it to get the total profitability facts. After getting the information, the employed bee returns to their hive, unloads the honey, and shares the information with the unemployed bees with a certain probability. Onlooker bees compiled the information got by observing the dance of employed bees and select the origin of food for herself. Scout bees looking for a new food source randomly around the hive, when an existing food source got exhausted. On average, in the colony of honey bees, 50% of bees are employed and 50% are onlooker bees and only 5% to 10% of the total is scouts.

ABC-Thresholding Approaches in Image Segmentation

Zhang and Wu [2] compared the Shannon entropy and Tsalli’s entropy using an ABC algorithm for finding the optimal multi- level thresholding in 2011. The experiments demonstrated that the Tsallis entropy is superior to traditional Shannon entropy thresholding, and the use of artificial bee colony gives faster result than either genetic algorithm or particle swarm optimization.

A maximum entropy-based artificial bee colony thresholding (MEABCT) technique was proposed by Horng in 2011 [3]. He compared the proposed method with another four techniques. Those are fast Otsu’s method, PSO, HBMO, and hybrid cooperative–comprehensive learning-based PSO algorithm (HCOCLPSO).

In 2011, Ma et al. [4] proposed a SAR image segmentation that uses ABC. The proposed algorithm finds an appropriate threshold value in a continuous grays scale interval. The algorithm is compared with Genetic Algorithm (GA) based and Artificial Fish Swarm (AFS) based segmentation methods. The result shows it is superior to other methods in terms of segmentation accuracy and segmentation time.

Kumar et al. [5] proposed a bi-level thresholding approach which uses the strength of MRLDE, PSO, and ABC aided with Otsu’s method in 2013. Modified random localization differential evolution (MRLDE) is a new technique that follows differential evolution. The author studied the working of MRLDE and proposed a new technique termed as Otsu+MRLDE. It is simulated and the results are compared with Otsu method.

ABC algorithm is once again used by Cuevas et al. [6] in 2012 to select the threshold value for segmenting an image. This method, approximate a one- dimensional image histogram using a Gaussian mixture mode. The parameters required for this are determined by the ABC algorithm. The complex computations of gradient-based methods which are very time taking improves a lot. The proposed method shows faster convergence and lower sensitivity for initial conditions as compared with Expectation Maximization algorithm. It shows improvement in complex time-consuming computations.

In 2013, B. Akay [7] used both PSO and ABC for getting the multilevel threshold. PSO and ABC are the two mostly used swarm intelligence-based global optimization algorithms. These are used with both Kapur’s method and between class variance to get the optimal threshold. Evaluating the results author concluded that ABC algorithm can be successfully used together with entropy criterion and between class variance for getting multilevel threshold values. Both PSO and ABC algorithm are able to replace Otsu’s method if the threshold levels is kept upto two. But if the threshold value increase ABC performs better than the other two methods. Simulated results also shows that both the algorithms are scalable and the time required for execution increases linearly with the problem size.

In 2015, Bhandari et al. [8] uses a new modified artificial bee colony (MABC) algorithm for segmenting satellite image. MABC is used to generate multilevel threshold values which overcomes the computational cost. Proposed algorithm uses three methods as objective functions: Kapur’s entropy, Otsu’s method, and Tsalli’s entropy. This algorithm gives more accurate optimal value as compared to other three algorithms: ABC, PSO and GA. MABC algorithm uses an improved solution search equation which is based upon the theory that the bees always search near the best solution of preceding iteration to get better exploitation. It has been concluded that the new algorithm can search the multilevel thresholds more efficiently than others.

References

[1]. Karaboga, Dervis. An idea based on honey bee swarm for numerical optimization. Vol.

200. Technical report-tr06, Erciyes university, engineering faculty, computer engineering department, 2005.

[2]. Zhang Y, Wu L (2011) Optimal multi-level thresholding based on maximum Tsallis entropy via an artificial bee colony approach. Entropy 13(4):841–859

[3]. Horng MH (2011) Multilevel thresholding selection based on the artificial bee colony algorithm for image segmentation. Expert Syst Appl 38(11):13785–13791L.

[4]. Ma M, Liang J, Guo M, Fan Y, Yin Y. SAR image segmentation based on artificial bee colony algorithm. Applied Soft Computing. 2011 Dec 1;11(8):5205-14.

[5]. Kumar S, Kumar P, Sharma TK, Pant M (2013) Bi-level thresholding using PSO, artificial bee colony and MRLDE embedded withOtsu method. Memet Comput 5(4):323– 334

[6]. Cuevas E, Osuna-Enciso V, Zaldivar D, Pe´rez-Cisneros M, Sossa H (2012a) Multithreshold segmentation based on artificial immune systems. Math Probl Eng. https://doi.org/10.1155/2012/874761

[7]. Akay B (2013) A study on particle swarm optimization and artificial bee colony algorithms for multilevel thresholding. Appl Soft Comput 13(6):3066–3091

[8]. Bhandari AK, Kumar A, Singh GK (2015a) Modified artificial bee colony based computationally efficient multilevel thresholding for satellite image segmentation using Kapur’s, Otsu and Tsallis functions. Expert Syst Appl 42(3):1573–1601