A New Assessment of Convolutional Neural Networks for Texture Directionality Detection

Published

December 1, 2023

Author(s)

Marcin Kociolek, Antonio Cardone

Abstract

Image texture analysis is ubiquitous as it finds application in many scientific fields of interest, including biomedical and material science. The automated detection of meaningful texture properties such as directionality remains a challenging task, due to the complexity of texture. Here, we build upon our past effort on the design of convolutional neural networks (CNNs) for automated and efficient texture directionality detection. As previously, CNN architectures are trained on a library of synthetic texture images with known directionality and varying perturbation levels. The present effort focuses on the enhancement of the training data through a new perturbation procedure and the use of a more diverse set of synthetic textures. Subsequently, we study the performance of new CNN architectures, such as grouped CNNs with specific properties, that are trained, validated and tested on the enhanced synthetic texture library. The resulting data yields novel insight into CNN-based texture directionality detection. Shallow and grouped CNNs show better performance than deep CNNs, unlike in our previous work. We discuss this shift in CNN performance and discuss its implications in future efforts, hence suggesting possible future work directions.

Proceedings Title

International Conference on Image Processing and Communications (IP&C)

Conference Dates

April 28-June 29, 2023

Conference Location

remote, PL

Pub Type

Conferences

Download Paper

https://doi.org/10.1007/978-3-031-41630-9_13

Local Download

Keywords

image processing, texture analysis, directionality detection

Citation

Kociolek, M. and Cardone, A. (2023), A New Assessment of Convolutional Neural Networks for Texture Directionality Detection, International Conference on Image Processing and Communications (IP&C), remote, PL, [online], https://doi.org/10.1007/978-3-031-41630-9_13, https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=936241 (Accessed December 3, 2024)