Classification Based on Machine Learning Methods for Identification of Image Matching Achievements

Abstract

Classification is one method in image processing. Image processing to search for similar images or with similarity ownership is called image matching or image matching. In the measurement of image matching, the original and fake logo objects are used. Identification of similarity manually with the help of human vision is not necessarily precise and it is difficult to obtain accurate results. Based on this, the identification of image matching of the original and fake logos automatically requires an application, in order to obtain precise and more accurate results. Identification of image suitability is determined through the image segmentation process, and feature extraction is based on the statistics of Red-Green-Blue (RGB), Hue-Saturation-Value (HSV), feature extraction of area, perimeter, eccentricity, and tangent distance measurements. The purpose of this study includes the identification of the achievement of image-matching logo images with comparisons of accuracy between various machine learning methods. The use of machine learning methods in this study includes the k-Nearest Neighbor (kNN), Random Forest (RF), and Multilayer Perceptron (MLP) methods. The use of the dataset includes eighteen training data and eight logo image testing data, divided into genuine and fake classes. The results of the measurement of the accuracy value obtained a value of seventy-five percent with the kNN method or the RF method, while the MLP method obtained an accuracy value of eighty-seven point five percent. Based on these results, it can be concluded that the MLP method with the highest accuracy value was chosen as a classification model from machine learning to identify the achievement of image matching on the original and fake logos. For further development, the system can be developed using other methods or a combination of different methods, in order to obtain better accurate results.