ABSTRACT: Artifacts are often found in archaeological excavation sites mixed with each other randomly. Therefore, classifying them manually is a difficult task and time consuming because they commonly exceed thousands of fragments. Thus, the aim of this study is to find a solution for classification of ancient pottery into groups by computer assistance. This is a preparatory stage for the next phase, which is the reconstruction of the archaeological fragments with high accuracy. To solve this problem, several steps must be taken, which are image segmentation via a proposed algorithm, and cluster the fragments into groups based on color and texture features. We proposed a novel algorithm that relies on the intersection of the RGB color between the archaeological fragments, and extraction of texture features from fragments based on Gray Level Co-occurrence Matrix (GLCM) that include Energy, Contrast, Correlation and Homogeneity. Finally, by using both proposed algorithm for classifying the color feature, and Euclidean distance for classifying the texture feature, we can classify the fragments with a high accuracy. The algorithm was tested on a pottery database, and it achieved a success rate almost 95%, so we would like to point out that by using the proposed algorithms we achieved promising results.

The archaeology is the scientific study of the last remnants of humanitarian civilization, which means studying the lives of ancient peoples by recovering and analyzing of the material culture and environmental data, which have left behind them. As well as, it provides an important information on dating, trade, technological achievements, population movements, and also for wars and conquests, this on one hand. On the other hand, the world has witnessed great development in the performance of computers as the use of image processing and pattern recognition techniques. That encouraged researchers to attempt to solve the problem of reconstruction of fractured objects from a large collection of randomly mixed fragments via proposed system instead of assembled manually, which is a tedious and time consuming task. Such as reconstructed the torn documents [1], jigsaw puzzles [2],[3], archaeological pottery fragments [4], wall paintings [5],[6] glass plate photographs [7].

Artifacts are often found in a fractured state, and the process of manual classification may require a great deal of time and tedious work. Classifying these fragments is a challenging task, especially if an artifact object consists of thousands of fragments. Hence, it is worthy to come up with an automatic solution for the cluster of the archaeological pieces accurately and reassembling them to original form. Thus, the aim of this work is to propose a new algorithm to find a solution that categorizes ancient pottery fragments into groups depending on their color and texture. This is a preparatory stage for the next phase, that represented by reconstruction the archaeological fragments. The second aim is improving the accuracy of the results. So the most important contributions of this paper assist the archaeologists to reduce both time and manual effort required, also reduce the number of employees. By depending on the intersection of RGB colors between the fragments, we obtain the color features of an artifact. As well as by depending on GLCM [8] that includes Entropy, Contrast, Correlation and Homogeneity, we obtain the texture feature. To classify the features, we proposed new algorithm for classifying the fragments according to the color feature, while we relied on Euclidean distance to classify the texture feature. After evaluation this

This paper is structured into several sections. An overview of the reconstruction of fragmented objects is presented in Section 2, the structure search is drawn in Section 3, and an important analysis is presented in Section 4. Finally, Section 5 summarizes and highlights the most important conclusions.

Numerous researches challenged the task of achieving the accurate reassembly of the archaeological fragments and returning them to their original form because of the high value in information that represents past civilizations and cultures. Ying & Gang [9] proposed an approach for the classification of ancient ceramic fragments by relying on surface texture features that were extracted by using Gabor wavelet transformation and classify by applying a non-supervision kernelbased fuzzy clustering algorithm. Their approach produces a 40 dimension eigen ector and achieved over 70%. Smith et al. [10] suggested a method for the classification of thin ceramic fragments depending on the features of color and texture on the basis of total different geometry (TVG) energies of the image. The features are classified by the fast vector quantization algorithm, and this will classify the fragment by finding the closest neighbors within the database. Their proposed method was applied on 98 fragments in different bowls, vases, and plates. The database included two groups that were very similar in color and texture, hence it is difficult to apply the classification. The result achieved by this method was 76%. The classification was correct when using the SIFT features. The proposed method classified the fragments when using TVG features with a success of 75%. Karasik & Smilansky [11] proposed a technique based on profile morphological analysis for the classification of ceramic. Their method depended on the mathematical representation of the profile for ceramic fragments on the basis of three functions (radius, tangent, and curvature). The authors employ the Cluster Analysis (CA) method to cluster and classify the fragments by using Discriminate Analysis (DA). The proposed algorithm has been applied on 358 fragments to reconstruct five types of vessels, and the results showed that the fragments were classified correctly 94.8%.

Makridis & Daras [12] focused on the automatic classification of fragments of pottery and ceramic that contain high or little textual information, also to enhance the classification accuracy, both the front and back view characteristics for the potteries were considered. Local features are extracted based on color and texture, which is converted to the global vector. The process of classifying the fragments was accomplished through the application of the KNearest Neighborhood classifier (KNN). The results after testing the model on the pottery database with a total of 62 fragments achieved a success rate 70.97%. After testing the model on the ceramic database of a total of 46 fragments, it achieved a success rate 78.26%.

Piccoli et al. [13] provided two complementary approaches for automatic classification of pottery sherds. The first one was characterized by a focus on the profile, and the other approach considers the visual surface feature. In order to extract the features, the authors relied on the points represent important information on the edge of the fragment, which will be used for the purpose of the matching process by applying the rotation invariant feature method. The second approach relied on local features that are extracted from a visual surface, such as color and texture, and also the standard deviation, michelson contrast, kirsch edge map and local binary patterns. They were classified by using K-Nearest Neighborhood and the results achieved in the overall classification accuracy according to the criteria sherd type, production technique, and Chronology were 65.99%, 93.96% and 55.65% respectively.

The proposed technique consists of several sets of procedures; each one performs a specific job, which can be summarized in the following structure, As shown in Figure 1 below. Moreover The algorithms are implemented in MATLAB R2014, and performed on a standard laptop computer (Intel Core i5 2.5GHz with 8GB RAM).