DAN Yuya
   Department   Matsuyama University  Department of Informatics, Faculty of Informatics
   Position   Professor
Language Japanese
Publication Date 2024/06
Type
Title Performance Evaluation of Machine Learning Models using Entropy of Images
Contribution Type
Journal ITE Tech. Rep.
Journal TypeJapan
Publisher The Institute of Image Information and Television Engineers (ITE), Japan
Volume, Issue, Page 48(16),pp.17-20
Total page number 4
Author and coauthor Riko ABE and Yuya DAN
Details In machine learning with artificial neural networks, there are differences in recognition accuracy in learning models of images classified into classes, depending on the features of the dataset and test images. In particular, we compared models composed of images with high entropy with models composed of images with low entropy for the training data. The entropies used were color image entropy, which is mainly the color information of the image, and differential entropy, which is mainly information about the shape of the image. Specifically, this paper describes the results of experiments comparing the dataset CIFAR-10 with various types of entropy to capture features of images with low recognition accuracy using methods such as back propagation.
ISSN 2424-1970