Selecting the Important Features to Classify the Archaeological Fragments by Using Statistical Tools

Abstract

Feature selection, the process of representing an object in the least dimensions, is one of the most important and difficult steps in pattern recognition. Therefore, meticulous selection of important features for classification is required. In this study, we propose a method based on Multidimensional Scaling (MDS) to reduce the dimensions of ancient ceramic fragment features. This method focuses on selecting the most important features based on the density of the grayscale image and texture. Finally, we use the Euclidean distance equation to classify objects into similar groups. With a database containing more than 300 images, the experiment achieved an impressive 90% success rate in accurately categorizing fragments as either similar or non-similar. These results demonstrate the effectiveness and promise of the proposed approach for image classification tasks, emphasizing the potential of statistical methods and image processing techniques for addressing complex computer vision challenges.

Finding on Machine Learning Approaches for Classification of Ancient Ceramics - A Systematic Literature Review

“…Some of the samples were photographed in several museums in China. ID2 [23] The proposed method was applied to two sets of images: one captured by a Nikon camera and the other obtained from the region of Independence National Historical Park in Philadelphia, dating back to the late 1700s and early 1800s. ID3 [24] There are many kinds of ancient Chinese ceramics.…”

“…Five papers (ID2, ID9, ID15, ID18, ID22) took advantage of American ceramic datasets. The dataset of ID2 was captured by a Nikon camera and the other image was obtained from the region of Independence National Historical Park in Philadelphia [23], dating back to 1700 and 1800 CE. Tusayan White Ware (from 800 to 1400 CE ) [28] was collected through the method of photographs shown in paper ID9.…”

“…The Chinese ancient ceramics cover from the 7th century BCE to the 17th century CE [22,30,33] while Thailand ceramics is relative to the period from the 13th to 15th century CE [25]. American ancient ceramics focus on the late 18th century [23,28]. Finally, Zulu ceramic vessels in South Africa were collected from the 19th century CE [35].…”

“…Table 7: Feature extraction of the reviewed studies ID Feature Extraction ID1 [22] Image segmentation of porcelain ewers based on SLIC-Ncut algorithm ID2 [23] This paper utilizes the Gray Level Co-occurrence Matrix (GLCM). This method is widely used as it provides many features, such as correlation, contrast, energy, and homogeneity based on each pixel and its neighboring pixel.…”

“…• SLIC-Ncut algorithm: it was employed in [22] (ID1) for the image segmentation of porcelain. • Gray Level Cooccurrence Matrix (GLCM) is widely used as it provides many features, such as correlation, contrast, energy, and homogeneity based on each pixel and its neighboring pixel [23] (ID2). • Data augmentation: it is a common approach in the context of image processing when working with deep learning algorithms that require large amounts of data.…”