Article: Hierarchical two-pathway autoencoders neural networks for skyline context conceptualisation Journal: International Journal of Information and Decision Sciences (IJIDS) 2020 Vol.12 No.4 pp.299 - 327 Abstract: In this paper, we proposed a novel hierarchical two-pathway autoencoders architecture to transform a local information based on skyline scene representation, into nonlinear space. The first pathway is intended for the transformation of the geometric features extracted from the horizon line. The second pathway is applied after the first one to joint the colour information under the skyline to the transformed geometric features, and to get the landscape context conceptualisation. To evaluate our suggested system, we constructed the SKYLINEScene database containing 2,000 images of rural and urban landscapes, with a high degree of diversity. In order to investigate the performance of our HTANN-Skyline, many experiments were carried out using this new database. Our approach shows its robustness in skyline context conceptualisation and enhances the classification rates by 1% compared to the AlexNet architecture; and by more than 10% compared to the hand-crafted approaches based on global and local features. Inderscience Publishers - linking academia, business and industry through research

Title: Hierarchical two-pathway autoencoders neural networks for skyline context conceptualisation

Authors: Ameni Sassi; Wael Ouarda; Chokri Ben Amar; Serge Miguet

Addresses: Research Groups in Intelligent Machines (REGIM-Lab.), University of Sfax, ENIS, BP 1173, 3038, Sfax, Tunisia ' Research Groups in Intelligent Machines (REGIM-Lab.), University of Sfax, ENIS, BP 1173, 3038, Sfax, Tunisia ' Research Groups in Intelligent Machines (REGIM-Lab.), University of Sfax, ENIS, BP 1173, 3038, Sfax, Tunisia ' LIRIS, Université de Lyon, UMR CNRS 5205, Université Lumière Lyon 2, 5 av. Mendès-France, Bât C, N 123, 69676, Bron, Lyon, France

Abstract: In this paper, we proposed a novel hierarchical two-pathway autoencoders architecture to transform a local information based on skyline scene representation, into nonlinear space. The first pathway is intended for the transformation of the geometric features extracted from the horizon line. The second pathway is applied after the first one to joint the colour information under the skyline to the transformed geometric features, and to get the landscape context conceptualisation. To evaluate our suggested system, we constructed the SKYLINEScene database containing 2,000 images of rural and urban landscapes, with a high degree of diversity. In order to investigate the performance of our HTANN-Skyline, many experiments were carried out using this new database. Our approach shows its robustness in skyline context conceptualisation and enhances the classification rates by 1% compared to the AlexNet architecture; and by more than 10% compared to the hand-crafted approaches based on global and local features.

Keywords: deep neural network; autoencoder; scene categorisation; skyline; curvature scale space; features transformation; classification; horizon line; hierarchical; skyline context conceptualisation.

DOI: 10.1504/IJIDS.2020.110447

International Journal of Information and Decision Sciences, 2020 Vol.12 No.4, pp.299 - 327

Received: 29 Dec 2018
Accepted: 12 Mar 2019
Published online: 20 Oct 2020 *

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

Title: Hierarchical two-pathway autoencoders neural networks for skyline context conceptualisation

Keep up-to-date