Title A Methodology of Enhancing the Accuracy of Image Classification with CNN
Authors 이상현(Lee, Sang-Hyun) ; 노녕(Lu, Ning)
DOI https://doi.org/10.5659/JAIK.2020.36.9.15
Page pp.15-22
ISSN 2733-6247
Keywords CNN; Artificial Neural Network; Image Classification; XOR; Deep Learning
Abstract This research proposes the method of using the Convolutional Neural Network (CNN) more efficiently in the classification of architectural images. The characteristics of images are not considered in the conventional methods of classifying architectural images. In this research, architectural images are first categorized primarily as two-dimensional images such as facades or three-dimensional images such as perspectives before classifying the types of the images. We compared the cases, where the types of architectural images are classified without being categorized as two- or three-dimensional images, with the cases where they are classified after the preliminary categorization is conducted. In this research, it was confirmed that the latter had higher classification accuracy. This implies that for image classification using the CNN, it could be effective to classify the types of architectural images after categorizing the basic preliminary delimitation based on characteristics of the images. We anticipate that the outcome of this research for architectural images will also apply to classification of other types of images.