Title: DCT-based rotational invariant unsupervised texture segmentation

Authors: T. Ray; P. Patra

Addresses: Research & Development, Tata Steel, Jamshedpur, Jharkhand, 831001, India ' Department of Instrumentation and Control, Automation Division, Tata Steel, Jamshedpur, Jharkhand, 831001, India

Abstract: Texture segmentation mainly focuses on dividing an object into different textures or regions with particular local properties. Usually, unsupervised texture segmentation is more challenging than supervised texture segmentation. In unsupervised texture segmentation, feature-based technique is more effective than model-based methods due to simplicity and fastness. The present paper proposes an unsupervised texture segmentation based on mask-based DCT as a feature-based technique. The proposed features are interesting as compared to plain wavelet and conventional DCT features for textures with and without rotation. Texture image database ranges from Brodatz album to natural scene. The proposed method performs well with the addition of white Gaussian noise. Moreover, post-processing on feature images is not required.

Keywords: discrete cosine transform; DCT; rotational invariance; wavelet transform; unsupervised texture segmentation; automatic texture category determination.

DOI: 10.1504/IJIT.2021.10047656

International Journal of Instrumentation Technology, 2021 Vol.3 No.1, pp.30 - 56

Received: 02 Apr 2021
Accepted: 18 Jun 2021

Published online: 26 May 2022 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article