Multi Color Channel vs. Multi Spectral Band Representations for Texture Classification - Archive ouverte HAL Access content directly
Conference Papers Year : 2021

Multi Color Channel vs. Multi Spectral Band Representations for Texture Classification

Abstract

Texture and color are salient visual cues of human perception and are widely used in many image analysis applications. Multi color spaces (MCS) approaches enrich the color texture representation and improve the performances of color texture classification applications. In these approaches, textures are represented by color texture features computed from descriptors that are extracted from images coded in several color spaces. When spectral characteristics of the texture of materials need to be analyzed, hyperspectral imaging (HSI) devices are chosen to address industrial applications that conventional color imaging is unable to solve. This paper aims to evaluate the contribution of HSI in the performances of texture classification methods compared to color imaging. For this purpose, we propose to extend the MCS approach to HSI in order to extract relevant spectral texture features computed from images of different spectral bands. Since these approaches both require to process high-dimensional data, they need to reduce the dimensionality of the feature space by selecting the most discriminating features, leading to a multi color channel (MCC) representation and a multi spectral band (MSB) representation respectively. This paper presents a unified representation of textures contained in color or hyperspectral images and compares the MCC and MSB representations for classification issues. Experimental results carried out on two hyperspectral texture databases show that the MCC representation is able to outperform the MSB ones.
No file

Dates and versions

hal-03150845 , version 1 (24-02-2021)

Identifiers

Cite

Nicolas Vandenbroucke, Alice Porebski. Multi Color Channel vs. Multi Spectral Band Representations for Texture Classification. 25 th International Conference on Pattern Recognition, ICPR 2021, Jan 2021, (Virtual) Milano, Italy. pp.310-324, ⟨10.1007/978-3-030-68790-8_25⟩. ⟨hal-03150845⟩
70 View
0 Download

Altmetric

Share

Gmail Mastodon Facebook X LinkedIn More