THE IMPLEMENTATION OF MEMETIC ALGORITHM ON IMAGE: A SURVEY

The growth of information technology is equal to the use of the algorithm. One of the most well-known algorithms is Memetic Algorithm (MA). MA is a part of the evolutionary algorithm and has been implemented on the most complex computational challenges. MA could be implemented in any field of research such as optimization, scheduling, prediction, image processing, image recognition, and many more. However, this research concerns the survey on the implementation of MA in image classification, image processing, and image recognition to find how many works are conducted related to MA and image. In this work, we use the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analysis) method to survey another research about the implementation of MA, specifically on the image. The purpose of this survey is to determine the extent of the implementation of MA in image data. Finally, we found each 1 paper uses MA for image retrieval, 12 papers use MA for image processing and 18 papers use MA for image recognition and 2 papers use MA for image classification. 6873 THE IMPLEMENTATION OF MEMETIC ALGORITHM


INTRODUCTION
The Memetic algorithm represents one of the most distinctive and thriving research fields in evolutionary computation. The term memetic algorithm has been used widely to describe the enhancement of searching procedures based on individual or population approaches. Memetic algorithm refers to Baldwinian Evolutionary Algorithm, Lamarckian Evolutionary Algorithm, Cultural Algorithm, and Local Search Genetic Algorithm.
The memetic algorithm is a part of the evolutionary algorithm and has been implemented on the most complex computational challenge [1]. Memetic algorithm [2] [3] is an enhancement of the evolutionary algorithm with local search separation [4]. This is a simple algorithm with reliable performance [5][6], provides solutions to many real problems with high accuracy [7][8] [9].
Many things developed to address optimization problems caused by huge research fields, such as economic, biology, chemistry, physics, and many more. Many researchers use a memetic algorithm to address complex problems as graph partition, graph coloring, packing, and many other common problems. The newest applications implementation of memetic algorithm are artificial neural network [10], pattern recognition [11], robotic motion planning [12], beam orientation [13], circuit design [14], electric service restoration [15], medical expert systems [16], single scheduling machine [17], automatic scheduling [18], person scheduling [19], nurse rostering optimization [20], processors allocation [21], maintenance scheduling [22], VLSI design [23], clustering of gene expression profiles [24], feature/gene selection [25] [26], and multi-class, multiobjective feature selection [27] [28]. Generally, the MA flowchart is shown in figure 2 below. 6874 ASSIROJ, WARNARS, ABDURACHMAN, KISTIJANTORO, DOUCET The MA is an enhancement of the evolutionary algorithm with local search separation [2] [3]. This is a simple algorithm with reliable performance [29], provide solution of any real problems with high accuracy [30]. The MA is also an enhancement of the genetic algorithm (GA). GA runs a local search in parallel condition, therefore it will never be stuck in the local extreme. However, it has to verify the appropriate solution in every iteration so that it will increase processing time, which means this algorithm is slower than others. This weakness can be overcome by adding a local search feature hereinafter known as MA. The MA is a heuristic search method, a combination of genetic algorithm and separated local search method that can increase the quality of solution [31]. The local search feature in MA can be implemented before or after the selection process, crossover, and mutation. It is also useful to minimize search space. Below is the pseudocode of the memetic algorithm.

RESEARCH METHOD
We have surveyed several papers related to the implementation of the Memetic Algorithm (MA) then conducted a review using the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analysis) method [32]. This method has five stages, 1st is Defining Eligibility Criteria, 2nd is Defining Information Resources, 3rd is Literature Selection, 4th is Data Collection, and 5th is Data Item Selection.
To aim for the best selection result, those 5 stages must be done in sequence, make sure that every stage is well done before moving to the next stage. Repeat the previous stage if there are deficiencies.

Stage 1: Defining Eligibility Criteria
As defined in [36], the eligibility criteria are specified by Inclusion Criteria (IC). In this work, we defined the articles into three criteria, which are:

Stage 2: Defining Information Resource
1) The articles can be found in online academics repositories such as Google Scholar, IEEE Xplore, ScienceDirect, and Springer Link.
2) In those online academic repositories, we will select the articles that appropriate to this work.

Stage 3: Literature Selection
1) Determining keywords. The 1st keyword is "memetics algorithm", we use this keyword to know more about MA in general. The 2nd is "memetics algorithm and implementation", we use this keyword because we want to know about MA and its implementation. The 3rd keyword is "memetics algorithm and image", and we use it to know specifically about MA on the image. The 4th is "memetics algorithm and implementation and image", and 6877 THE IMPLEMENTATION OF MEMETIC ALGORITHM we use this keyword to get some information about the implementation of MA on image data. The 5th keyword is "memetics and image", this is almost the same as number 3, but we do not include the word "algorithm" to get information about memetics and images in general. The last keyword is "memetics and fingerprint", we use this keyword because in future work, specifically, we will focus on the implementation of MA on fingerprint image datasets.
2) Select articles that are related to criteria by the title, abstract, and article's keywords.
3) Articles that are not eliminated from the previous stage will be read, fully, or partially to define items' eligibility.
Only short-listed articles will be assessed to find the linkage to this work. These articles will be re-assessed by doing steps 3 to step 4 above.

Stage 4: Data Collection
We have created a data extraction form to collect data manually. From each keyword, this survey assesses 33,965 articles based on keywords "memetics algorithm"; 22,343 articles based on keywords "memetics algorithm and implementation"; 19,513 articles based on keyword "memetics algorithm and image"; 17,014 articles based on the keyword "memetics algorithm and implementation and image", 19,149 articles based on the keyword "memetics and image"; and 1,156 articles based on the keyword "memetics and fingerprint". The total of searched articles is 113,140.

Stage 5: Data Item Selection
Data are obtained from short-listed articles that explain the use or the implementation of the memetics algorithm. Finally, we decide and select 33 articles that are appropriate to the survey based on the titles, abstracts, keywords, and contents. The other papers are not included in the survey because they are not eligible in every selection stage (stage 1 to 4). Tables 1, 2, 3, and 4 show the data that have been collected from each source.

RESULT AND DISCUSSION
The research proposed to observe the implementation of Memetics Algorithms (MA) that have been done by other researchers. Based on this purpose, the research identifies many implementations of the MA shown in table 5. Table 5 shows research papers that are focus on the method and implementation of MA. Most of them are from the journal and only four papers from the conference. Based on the 33 papers in table 5, we divide the implementation of the MA into four categories, image processing, image recognition, image classification, and image retrieval. 12 papers use MA for image processing, 18 papers use MA for image recognition, and every 2 papers use MA for image classification, and 1 paper use MA for image retrieval.  According to table 6 above, the research analyzes the implementations of the memetic algorithm (MA) in various research fields. Most of the implementation of this algorithm is in image processing and image recognition, with various proposed methods. The amount is shown in Figure   3.

CONCLUSION
According to the survey that has been done, the MA implementation for image processing consists of 12 papers, 1 paper uses MA for retrieval, 2 papers use MA for image classification and the most is 18 papers use this algorithm for image recognition. In image processing, MA can distinguish colors, shapes, and image's texture. The result of this research could be a reference for future research about MA implementation that categorized into four, image processing, image classification, image retrieval, and image recognition. Some method has proposed for the novel of its implementation.