The best texture image for Gaussian Naïve Bayes with nearest neighbor interpolation

Santoso, Irwan Budi ORCID: https://orcid.org/0000-0001-5586-9035, Utama, Shoffin Nahwa ORCID: https://orcid.org/0000-0001-9843-199X and Supriyono, Supriyono ORCID: https://orcid.org/0000-0002-4733-9189 (2024) The best texture image for Gaussian Naïve Bayes with nearest neighbor interpolation. Jurnal Nasional Teknik Elektro dan Teknologi Informasi, 13 (1). pp. 68-75. ISSN 24605719

Text 18609.pdf - Published Version Download (1MB)

Abstract

One of the factors affecting the performance of the Gaussian naïve Bayes classifier (GNBC) in texture image classification is the image size (dimensions). Image size is one of the best texture image criteria besides its pixel value. In this study, a method is proposed to obtain the size of the best texture image for GNBC by nearest neighbor (NN) interpolation optimization. The best texture image size with interpolated pixel values makes GNBC able to distinguish texture images in each class with the highest performance. The first step of the proposed method was to determine the texture image size for training through a combination of row and column sizes in the optimization process. The next important step in generating the new texture images was resizing each of the original texture images using NN interpolation. The next step was to build GNBC based on the new image from interpolation and determine the classification accuracy. The last step was to select the best texture image size based on the largest classification accuracy value as the first criterion and image size as the second criterion. The evaluation of the proposed method was carried out using texture image data from the CVonline public dataset involving several test scenarios and interpolation methods. The test result shows that in scenarios involving five classes of texture images, GNBC with NN interpolation gives the smallest classification accuracy value of 89% and the largest 100% at the best image size, 14 × 32 and 47 × 42, respectively. In scenarios involving small to large class numbers, GNBC with NN interpolation provides classification accuracy of 81.6%–95%. From these results, GNBC with NN optimization gives better results than other nonadaptive interpolation methods (bilinear, bicubic, and Lanczos) and principal component analysis (PCA).

Item Type:	Journal Article
Keywords:	image; texture; interpolation; Naïve Bayes; nonadaptive; accuracy
Subjects:	08 INFORMATION AND COMPUTING SCIENCES > 0801 Artificial Intelligence and Image Processing > 080199 Artificial Intelligence and Image Processing not elsewhere classified
Divisions:	Faculty of Technology > Department of Informatics Engineering
Depositing User:	irwan budi santoso
Date Deposited:	04 Apr 2024 10:46

Downloads

Origin of downloads

Actions (login required)

View Item