2026-08-13T20:15:55+03:30 http://jsdp.rcisp.ac.ir/browse.php?mag_id=16&slc_lang=fa&sid=1

16-428 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Salman Maskanati Maskanati@gmail.com Ahmad Keshavarz a.keshavarz@pgu.ac.ir The Joint-up, cursive form of Persian words and immense variety of its scripts, also different figures of Persian letters depending on their sitting positions in the words, have turned the Persian handwritings recognition to an intense challenge. The major obstacle of the most often recognition ways, is their inattention to sentence contexture which causes utilizing of a word with correct appearance within an incorrect sentence, when an input word is misrecognized. Sketching a solution that provides suitable analysis of sentence contexture, requires huge linguistic resources to take place as a fine representative for the chosen language to be recognized. In this article, a new method for online recognition of Persian words is presented which tries to improve recognition process by using the term contexture. In this article, the vocabularies collection of Persian language is divided into two groups. The first category is the vocabulary with all of their sub-words being supported by the database of handwritten subclasses, while these vocabulary form 68.2% of the total vocabulary, and the assumptions being scored at the recognition stage, are members of these vocabularies. The second category is the vocabulary that is not supported by the database. Obviously, if the recognition system does not support this vocabulary, it cannot recognize more than 30 percentages of the language's words. At the recognition stage, the symptoms are detected and a symptom tag is produced. Also, at this stage, using the same label, the vocabulary is also selected as the sign with the input word. (These vocabularies are chosen from those were not supported at the recognition stage). Scoring for hypotheses was done by combining recognition scores and linguistic models. The certain fact in this section is that it is impossible to calculate recognition scores due to the absence of hypothetical subheadings. Therefore, the vocabulary score being recognized in the previous steps, is used. According to the studies, it was concluded that if the word is equivalent to a member's input from a supported vocabulary, even if the result of the recognition is incorrect, in most cases the correct term is in the first four hypotheses. Usually, scores of the first few hypotheses are close to each other, and the other assumptions are far from the correct hypothesis. Since the system operates online, unnecessary computations should be avoided. Therefore, if the number of hypotheses in the recognition section are more than four hypotheses, only the first four hypotheses are calculated for the language model. To calculate the recognition score for new hypotheses, if there are fewer than four hypotheses in the recognition section, the lowest hypothesis score and otherwise the hypothesis score are considered for the recognition score of the new hypotheses. Then, as with previous assumptions, for the new hypotheses, the linguistic score is calculated, and then the final score is obtained for each hypothesis. Finally, the assumption with the highest score is considered as the system output, and the rest of the assumptions are displayed in the output to the user. Experiments show that even in the event of a mistake, the correct word is often presented as a second hypothesis in most cases, and in some cases as a third hypothesis. Also, to reduce the limits and rules that gainers compel to submit. The method demonstrated in this article includes the symptoms and morphemes framework of input handwritten are segregated and the framework of each morpheme with its symptoms is specified at first, then the symptoms of morphemes are specified and based on them a collection of words is being considered as a hypothesis. Each hypothesis is given a score by measuring the similarity to input handwritten and according to taken scores, the likely hypotheses are indicated. Then, this procedure is led to achieve hypotheses more likely by lingual models. To totalize the scores of a hypothesis, for the differences in scale of taken scores, a method of score normalization is being offered. The results demonstrate that by utilizing of a language model with an online system of handwriting recognition, a significant reduction of words recognition error rate is being achieved. In addition to error rate reduction, by taking advantages of this language model, a technique is being offered that can handle the Persian vocabulary recognition entirely. By availing the offered manner, the recognition precision at initial stage of letters level up to 95.9% and so the language model recognition up to 99.3% improved. So, using huge linguistic resources for Persian language and utilizing a language model, can improve the accuracy of recognition. For further work, reinforcement learning algorithm is suggested to adapt the algorithm for users.   Online Recognition Persian Handwriting k-nearest Neighbor Language Model User Limitation 2017 9 01 3 24 http://jsdp.rcisp.ac.ir/article-1-428-en.pdf 10.18869/acadpub.jsdp.14.2.3

16-468 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Jamal Ghasemi j.ghasemi@umz.ac.ir Somayeh Kord somayye.kord@yahoo.com Mohamad Gholami m.gholami@umz.ac.ir Cardiac arrhythmias are one of the most common heart diseases that may cause the death of the patient. Therefore, it is extremely important to detect cardiac arrhythmias.  3 categories of arrhythmia, namely, PAC, PVC, and normal are considered in this paper based on classifier fusion using evidence theory. In this study, at first a sample is carrying out the ECG signal with 250 point.  Moreover, in each of the sampling, the maximum values will be obtained. Then, the average of the calculated values would be considered as adaptive thresholding and the total signals are multiplied by the inverse adaptive thresholding. After fixing the adaptive thresholding at number one, total resulting signal is becoming the power of 2. In this situation, the amounts smaller than one, are weakened and the larger than one amounts are reinforced. The smaller amount is removed and other amounts are held. Then, the maximum in each of the sampling is considered. In sampling areas that there is no peak, some maximum can be identified with zero value that these points should be removed from the set maximum. To find the maximum point where the maximum is close to the borders of sampling, two peaks may be placed in one field. This problem leads to removing one peak and non-recognition of the smaller peak. Some peaks near the border of sampling, for example the previous or next point on the border may be identified as the peak which eliminates the major peak and identifies the unrealistic peak. To solve this problem, the 80-point sampling is performed around each detected peak and the maximum value is obtained at the sampling areas. In this way, the correct peaks are identified and the wrong one will be deleted. In some parts, the peak signal is not quite sharp, and maybe two or more points that are adjacent to each other with the same value, will be considered as a peak. In other words, a closed peak is detected several times, which leads to detection of extra and incorrect peaks. In these circumstances, according to an amount that only belongs to one peak, just one of them should be considered and the other should be removed. After these steps, an obtained signal which includes peaks R, is compared with the original signal. To achieve the correct answer, it changes the number of sampling points and each time the result is compared with the previous values and with the original signal, too, until finally the major peaks will be identified. Then, HRV signal be will calculated. Linear properties contain root mean square of successive differences between normal intervals (RMSSD) and standard deviation of normal to normal intervals in a row (SDNN) and also heart rate (HR Mean) are calculated. Around each peak, 81 points window is inserted. These points for each peak is in one row. So resulting matrix (X) has 81 columns and its rows are the number of R peaks. SVD of matrix(X) is calculated. The obtained Matrix S will include the individual values. These singular signal values are non-linear features. If all used values are single, they can eclipse the linear features which will lead to the lack of features’ effect. Because of this reason, it is used only from the largest single value as a non-linear feature. The combination of linear and non-linear characteristics as input is applied to MLP, Cascade Feed Forward and RBF neural networks and every (single) answer is studied. The answers for each class have a level of probability that any classifier can independently be taken to the classification of cardiac arrhythmias. A class that has the greatest probability is allocated to the data. These probabilities show the uncertainty of the answers. Each of the classifiers is considered as a witness. All the possibilities for different classes of each witness uncertainties function are modeled and crime function is defined. In other words, belief structure is formed for evidence. At this stage, by combined Demster law, the mass functions will combine together. In this situation, the level of uncertainty is much reduced and the class with the highest crime will be selected as the answer. According to the survey results, the combination of linear and non-linear characteristics for training and testing the neural networks classifiers has increased the accuracy of the answer. In other words, the extraction of more features leads to better training the neural networks and increases the accuracy of the classifiers. It can be noted that the using classifiers uncertainty principle and combining them by using the evidence theory has increased the accuracy of the final classification. The results of this study show that the proposed method was able to classify cardiac arrhythmias in the presence of noise and provided an acceptable answer for the intended issue. In sum, the proposed method has been able to classify 3 categories of cardiac arrhythmia such as PVC, PAC and NORMAL with high accuracy. This is performed in the best situation with sensitivity greater than 0/98.   ECG signal Classifier Neural Networks Evidence theory 2017 9 01 25 42 http://jsdp.rcisp.ac.ir/article-1-468-en.pdf 10.18869/acadpub.jsdp.14.2.25

16-344 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Azadeh Kianisarkaleh azade.kiyani@gmail.com Mohammad Hassan Ghassemian ghassemi@modares.ac.ir Feature extraction performs an important role in improving hyperspectral image classification. Compared with parametric methods, nonparametric feature extraction methods have better performance when classes have no normal distribution. Besides, these methods can extract more features than what parametric feature extraction methods do. Nonparametric feature extraction methods use nonparametric scatter matrices to compute transformation matrix. Nonparametric Discriminant Analysis (NDA) is one of the nonparametric feature extraction methods in which, to form nonparametric scatter matrices, local means of samples and weight function are used. Local mean is calculated by k nearest neighbors of each sample and weight function emphasizes on boundary samples in between class scatter matrix formation. In this paper, modified NDA (MNDA) is proposed to improve NDA. In MNDA, the number of neighboring samples, when measuring local mean, are determined considering position of each sample in feature space. MNDA uses new weight functions in scatter matrix formation. Suggested weight functions emphasizes on boundary samples in between class scatter matrix formation and focus on samples close to class mean in within class scatter matrix formation. Moreover, within class scatter matrix is regularized to avoid singularity. Experimental results on Indian Pines and Salinas images show that MNDA has better performance compared to other parametric, nonparametric feature extraction methods. For Indian Pines data set, the maximum average classification accuracy is 80.34%, which is obtained by 18 training samples, support vector machine (SVM) classifier and 10 extracted features achieved by MNDA method. For Salinas data set, the maximum average classification accuracy is 94.31%, which is obtained by 18 training samples, SVM classifier and 9 extracted features achieved by MNDA method. Experiments show that using suggested weight functions and regularized within class scatter matrix, the proposed method obtained better results in hyperspectral image classification with limited training samples.   Hyperspectral images Feature extraction Supervised classification Hughes Phenomenon Limited training samples 2017 9 01 43 58 http://jsdp.rcisp.ac.ir/article-1-344-en.pdf 10.18869/acadpub.jsdp.14.2.43

16-295 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Seyed mohamad bagher Sajadi mb.sajadi@qiau.ac.ir Hassan Rashidi Hrashi@atu.ac.ir Behrooz Minaei bidgoli B_minaei@iust.ac.ir In Natural Language Processing (NLP) studies, developing resources and tools makes a contribution to extension and effectiveness of researches in each language. In recent years, Arabic Named Entity Recognition (ANER) has been considered by NLP researchers due to a significant impact on improving other NLP tasks such as Machine translation, Information retrieval, question answering, query result clustering, etc. While most of these researches are based on Modern Standard Arabic (MSA), in this paper, we focus on Classical Arabic (CA) literature. We propose a corpus called NoorCorp with 130k labeled words for research purposes which is annotated by expert human resources manually. This corpus is based on a Historic-Islamic book of 1200 years ago including 1843 sentences and 127550 words. We also collected about 18k proper names from old Hadith books as a gazetteer which is called NoorGazet used as a future. In this paper, we propose a new approach to extract named entities (NEs) including person, location, organization and time. We use hybrid approach benefiting from advantages of Rule based approach and Machine learning approach. We divided the NoorCorp into two parts of training and test sets containing 80% and 20% of the data set respectively. Prediction model, based on Boosting method, was developed in two steps which Adaboost.M1 is employed to identify NEs and Adaboost.M2 is employed to classify NEs. There are many methods using multiple classifiers as voters and summing up their results, among which, ensemble methods are those which generate multiple hypotheses using the same base learner. We developed an ensemble consisting of 50 members (classifiers) based on decision stump to implement the weak learner. Since only 17% of the text data is composed of name entity labels, we had to deepen the tree while restricting pruning. We exploited tokenizing, part of speech (POS) tagging, and base phrase chunking (BPC) to overcome linguistic obstacles in Arabic including Meaning ambiguity, Optional diacritics, Complex morphology and Nonstandard written text. Moreover, using a statistical technique, the most frequently used words extracted as key words. Results show that performance of the method is better than decision tree as the base classifier. An overall F-measure value of 86.85 obtained which is better than base line about 20% and CART decision tree about 12%. Since CA corpus consists of simpler linguistic patterns compared to MSA, we applied the proposed approach on ANERCorp as Modern Standard Arabic corpus. Results show that the proposed model outcome on CA corpus is about 19% better than MSA. This result is due to the fact that there are plenty of NEs entered to MSA from other languages. These proper names do not have specific patterns and do not exist in the gazetteer. In addition, many NE’s are not distributed uniformly in ANERcorp which considerably reduces the results accuracy.      Named entity recognition (NER) Ensemble learning Boosting method Classical Arabic Language 2017 9 01 59 74 http://jsdp.rcisp.ac.ir/article-1-295-en.pdf 10.18869/acadpub.jsdp.14.2.59

16-392 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Seyyed Ali Sadati Tileboni seyyedali.sadati@nit.ac.ir Hamid Jazayeriy jhamid@nit.ac.ir Mojtaba Valinataj m.valinataj@nit.ac.ir Graph coloring is a way of coloring the vertices of a graph such that no two adjacent vertices have the same color. Graph coloring problem (GCP) is about finding the smallest number of colors needed to color a given graph. The smallest number of colors needed to color a graph G, is called its chromatic number. GCP is a well-known NP-hard problems and, therefore, heuristic algorithms are usually used to solve it. GCP has many applications such as: bandwidth allocation, register allocation, VLSI design, scheduling, Sudoku, map coloring and so on. We try genetic algorithm (GA) and chaos theory to solve GCP. We proposed a heuristic algorithm called CMHn to implement multi-point crossover operation in GA. To generate initial population, a fast greedy algorithm is used. In this algorithm, the degree of each node and the number colors in its neighbor is used to assign a color to each node. Mutation operation in GA is used to explore the search space and scape from the local optima. In this study, a chaotic mutation operation is presented to select some vertices and change their color.  The crossover and mutation parameters in the proposed algorithm is tuned based on some experiment. To evaluate the proposed algorithm, some experiment is conducted on DIMACS data set. Among DIMACS sample graphs, DSJ, Queen, Le450, Wap are well-known challenging samples for graph coloring. The proposed algorithm is executed 10 times on each sample and the best, worst and mean results are reported. Results show that the proposed algorithm can effectively solve GCP and have comparable outcome with the recent studies in this field. The proposed method outperforms other algorithms on very large graphs (Wap graphs).    Graph coloring problem genetic algorithm heuristics multi-point crossover chaotic mutation 2017 9 01 75 96 http://jsdp.rcisp.ac.ir/article-1-392-en.pdf 10.18869/acadpub.jsdp.14.2.75

16-434 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Ali Jamshidi jamshidi@shirazu.ac.ir Mehran Yazdi yazdi@shirazu.ac.ir Maryam Manafi jamshidi801@gmail.com Compression can be done by lossy or lossless methods. The lossy methods have been used more widely than the lossless compression. Although, many methods for image compression have been proposed yet, the methods using intelligent skipping proper to the visual models has not been considered in the literature. Image inpainting refers to the application of sophisticated algorithms to replace lost or corrupted parts of the data so that visual difference cannot be inferred from the reconstructed image. In this paper, first we review some of the image inpainting algorithms and some of the image compression techniques using the inpainting algorithms, we propose a new inpainting based image compression algorithm that can improve the compression rate considerably. We present image compression system based on the proposed parameter-assistant image inpainting method to more deeply exploit visual redundancy inherent in color images. We have shown that with carefully selected dropped regions and appropriately extracted parameters from them, dropped regions can be satisfactorily restored using the proposed PAI algorithm. Accordingly, our compression scheme has a higher coding performance compared with traditional methods in terms of the perceptual quality. To best represent the target region for inpainting, an effective region classifier is required. A generic solution is to study the distribution of each image region and find the best match among the candidates in the predefined model class. For simplicity, in our scheme, an entire image divided into three categories: gradated, structural, and non-featured, at non-overlapping block level of size S×S. The classification is performed based on edge content and color variance in each block. Simulation results show that our proposed method has reasonable visual quality in comparison with the other proposed image compression algorithms.     Image Inpainting Image Compression Intelligent Information Removing Coding 2017 9 01 97 114 http://jsdp.rcisp.ac.ir/article-1-434-en.pdf 10.18869/acadpub.jsdp.14.2.97

16-409 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 nooshin_bigdeli@yahoo.com hamedjabbarie@gmail.com Dispersion of nanoparticles in nanostructures is one of the most important indicators designed to verify the effectiveness of proposed methods in the synthesis of nanomaterials. In the recent years, various methods have been suggested for the synthesis of nanostructures in which the Scanning Electron Microscopy (SEM) has been used to show the quality of the nanomaterial. The SEM images of nanoparticles contain structural, chemical and morphological information with high resolution in nanometer scale of nanomaterials. One of the challenges in the quality of dispersion’s nanostructures is detection of agglomeration degree. In some SEM images of nanoparticles, the particles have speeded uniformly and not aggregately. In some of the other SEM images, their particles are agglomerated. Also, there are a few SEM images of nanoparticles that their particles aren’t very aggregate or diffused. If the SEM images of nanoparticles with their particles speeded uniformly, are called good images, and the images with their aggregate particles are called bad images, and the images with their particle dispersion between good and bad images, are called average images, the nanomaterials could be classified in categories of good, average, and bad images. In this paper, a new algorithm has been provided to classify nanostructures using SEM images of nanoparticles. For this purpose, these images were transformed to time series at first (the time series extracted are unique for each SEM image of nanoparticles) and their specifications were investigated through time series analysis methods. Then, statistical specifications of these series were extracted. Six statistical specifications have been extracted for classification of nanostructures. These specifications are as follows: standard deviation, first and second kurtosis, interquartile range, the criterion of Pearson, and skewness. The extracted specifications were used as inputs of a fuzzy inference system for classifying microscopic images of nanostructures into three groups: good, average and bad. This algorithm has been tested on 65 nanoparticles microscopic images with identical size and resulted precision above 93 percent indicated validity of this algorithm.   SEM image of nanoparticles Time series analyses Statistical features Fuzzy logic 2017 9 01 115 130 http://jsdp.rcisp.ac.ir/article-1-409-en.pdf 10.18869/acadpub.jsdp.14.2.115

16-448 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Mohammad Mikaeili Mikaeili@shahed.ac.ir foroogh Najafi foroogh.najafi.c@gmail.com As a Brain computer interface system, BCI P300 Speller tries to help disabled people and patients to regain some of their lost ability with allowing communication via typing. The ability of personalization is one of the most important features in a BCI system, so the typing language as a personalization factor is an important feature in a BCI speller. Most prior researches on P300 Speller has focused on displaying English alphabet and there were only few studies made on other languages such as Chinese. In this research, we present a P300 Speller system, based on RCP, for Persian (Farsi) character input. RCP (Row or Column Paradigm) was introduced by Farwell and Donchin at 1988, and since then it has been considered as a benchmark in P300 BCI speller research. As a result, in this study also, Row or column paradigm was selected as the base stimulation pattern in P300 speller system. In order to evaluate the Persian row or column paradigm performance, we recorded EEG signals from volunteered subjects while the stimulation pattern was being displayed. It should be noted that the test was explained to each subject before testing, and for more experience and in order to reduce the error, each subject participated in an experiment test before attending the main test. These EEG signals were recorded from 8 channels based on ‘‘Fz’’, ‘‘Cz’’, ‘‘P3’’, ‘‘Pz’’, ‘‘P4’’, ‘‘O1’’, ‘‘Oz’’ and ‘‘O2’’ site in accordance to the International 10–20 system electrode placement system and by using Science Beam co.’s EEG recording device. The sample rate was 1 KHz which was down sampled to 250Hz. After recording, the EEG signals were filtered using a band passed filter And for classification, Linear discriminate analysis was used in combination with K-fold validation method for classifier training. As performance determination, we calculated accuracy and bit rate for the mentioned system based on recorded data from volunteers and reached the average accuracy of 88.21% and bit rate of 6.74 (bits/minute) (we use Linear LDA classifier for classification and the total trial number was set to 15). Furthermore, in this research performance was measured for different trial number and final results demonstrated that this system can achieve high average accuracy of 80.06% and average bit rate of 42.43 (bits/minute) by using only 2 repetitions.   Brian computer interface systems BCI P300 Speller P300 wave LDA classifier 2017 9 01 131 140 http://jsdp.rcisp.ac.ir/article-1-448-en.pdf 10.18869/acadpub.jsdp.14.2.131

16-406 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Mohammad Amin Tolou Beidokhti M.a.Tolou.b@Gmail.com Alireza Ahmadyfard ahmadyfard@shahroodut.ac.ir Document images produced by scanners or digital cameras usually have photometric and geometric distortions. If either of these effects distorts document, recognition of words from such a document image using OCR is subject to errors. In this paper we propose a novel approach to significantly remove geometric distortion from document images. In this method first we extract document lines from document using morphological operators. Then, extracted document lines are divided into a number of equal size column strips.  This allows to assume that each segment of line document is not curved. Each extracted document line segment is aligned horizontally. For this purpose, a segment line of document is rotated at different angels and for each rotation horizontal projection is obtained. The rotation angle with maximum peak at the corresponding projection signal is selected to align the line segment, horizontally. In order to estimate the geometrical distortion, for each document line a reference point is extracted from each line segment. These points indicate the position of a document line at starting column of line segments. Using reference points of a document line a polynomial function is fitted to each document line. At the end, geometric distortion for each part of the document is eliminated using a perspective transformation. This transformation is estimated based on the extracted polynomial function. To increase the stability of the proposed method for short text lines, the curve of adjacent text lines of longer length is used. A post processing stage is required after applying perspective transformation on document patches. Since this transformation is a continuous mapping but it is applied on digital images. To remove this distortion from the result, the consistency of each pixel value with the value of neighboring pixels are considered to correct the value of inconsistence pixels. The proposed method is implemented on Persian and English databases and has been compared with the existing methods. The results indicate the efficiency and accuracy of the proposed method in elimination of geometric distortions.   Geometric distortion document processing perspective Transformation Optical character recognition (OCR) 2017 9 01 141 158 http://jsdp.rcisp.ac.ir/article-1-406-en.pdf 10.18869/acadpub.jsdp.14.2.141

16-453 2026-08-13 10.1002

Signal and Data Processing JSDP 2538-4201 2538-421X 10.66224/jsdp 2017 14 2 Arash Chaghari a.chaghari@tabrizu.ac.ir Mohammad-Reza Feizi-Derakhshi mfezi@tabrizu.ac.ir Imperialist Competitive Algorithm (ICA) is considered as a prime meta-heuristic algorithm to find the general optimal solution in optimization problems. This paper presents a use of ICA for automatic clustering of huge unlabeled data sets. By using proper structure for each of the chromosomes and the ICA, at run time, the suggested method (ACICA) finds the optimum number of clusters while optimal clustering of the data simultaneously.To increase the accuracy and speed of convergence, the structure of ICA changes. As in different applications, there is a need for data clustering which the number of clusters is not known before it is necessary to have methods that can cluster data without knowing the correct prediction of the number of clusters. In the other words, the proposed algorithm requires no background knowledge to classify the data.  In addition, the proposed method is more accurate in comparison with other clustering methods based on evolutionary algorithms. In Imperialist Competitive Algorithm, firstly steps should be taken to increase search rates and explore possible solution while approaching to the global optimal response the steps should be reduced to ensure that the algorithm is not lost and it is not in the local optimal manner. For this purpose and improvement of imperialist competitive algorithm, mutation rate and revolution operator's operation rate are determined dynamically. DB and CS are cluster validity Indexes. In this paper, DB and CS cluster validity measurements are used as the objective function. To demonstrate the superiority of the proposed method, the average of fitness function and the number of clusters determined by the proposed method is compared with three automatic clustering algorithms based on evolutionary algorithms. The partitional clustering algorithms are based on three powerful well-known optimization algorithms, namely the genetic algorithm, the particle swarm optimization and differential evolutionary algorithm. Partitional Clustering Automatic Clustering Imperialist Competitive Algorithm (ICA) 2017 9 01 159 169 http://jsdp.rcisp.ac.ir/article-1-453-en.pdf 10.18869/acadpub.jsdp.14.2.159