A saliency-based technique for advertisement layout optimisation to predict customers’ behaviour

Authors: Bruno, A., Lancette, S., Zhang, J., Moore, M., Ward, V.P. and Chang, J.

Journal: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

Volume: 12662 LNCS

Pages: 495-507

eISSN: 1611-3349

ISBN: 9783030687892

ISSN: 0302-9743

DOI: 10.1007/978-3-030-68790-8_39

Abstract:

Customer retail environments represent an exciting and challenging context to develop and put in place cutting-edge computer vision techniques for more engaging customer experiences. Visual attention is one of the aspects that play such a critical role in the analysis of customers behaviour on advertising campaigns continuously displayed in shops and retail environments. In this paper, we approach the optimisation of advertisement layout content, aiming to grab the audience’s visual attention more effectively. We propose a fully automatic method for the delivery of the most effective layout content configuration using saliency maps out of each possible set of images with a given grid layout. Visual Saliency deals with the identification of the most critical regions out of pictures from a perceptual viewpoint. We want to assess the feasibility of saliency maps as a tool for the optimisation of advertisements considering all possible permutations of images which compose the advertising campaign itself. We start by analysing advertising campaigns consisting of a given spatial layout and a certain number of images. We run a deep learning-based saliency model over all permutations. Noticeable differences among global and local saliency maps occur over different layout content out of the same images. The latter aspect suggests that each image gives its contribution to the global visual saliency because of its content and location within the given layout. On top of this consideration, we employ some advertising images to set up a graphical campaign with a given design. We extract relative variance values out the local saliency maps of all permutations. We hypothesise that the inverse of relative variance can be used as an Effectiveness Score (ES) to catch those layout content permutations showing the more balanced spatial distribution of salient pixel. A group of 20 participants have run some eye-tracking sessions over the same advertising layouts to validate the proposed method.

http://eprints.bournemouth.ac.uk/35890/

Source: Scopus

A Saliency-Based Technique for Advertisement Layout Optimisation to Predict Customers’ Behaviour

Authors: Bruno, A., Lancette, S., Zhang, J., Moore, M., Ward, V.P. and Chang, J.

Editors: Bimbo, A.D., Cucchiara, R., Sclaroff, S., Farinella, G.M., Mei, T., Bertini, M., Escalante, H.J. and Vezzani, R.

Conference: DeepRetail2020 ICPR International Workshops and Challenges

Dates: 12-15 January 2021

Journal: Pattern Recognition. ICPR International Workshops and Challenges - Virtual Event, January 10-15, 2021, Proceedings, Part II

Volume: 12662

Pages: 495-507

Publisher: Springer

DOI: 10.1007/978-3-030-68790-8_39

Abstract:

Customer retail environments represent an exciting and challenging context to develop and put in place cutting-edge computer vision techniques for more engaging customer experiences. Visual attention is one of the aspects that play such a critical role in the analysis of customers behaviour on advertising campaigns continuously displayed in shops and retail environments. In this paper, we approach the optimisation of advertisement layout content, aiming to grab the audience’s visual attention more effectively. We propose a fully automatic method for the delivery of the most effective layout content configuration using saliency maps out of each possible set of images with a given grid layout. Visual Saliency deals with the identification of the most critical regions out of pictures from a perceptual viewpoint. We want to assess the feasibility of saliency maps as a tool for the optimisation of advertisements considering all possible permutations of images which compose the advertising campaign itself. We start by analysing advertising campaigns consisting of a given spatial layout and a certain number of images. We run a deep learning-based saliency model over all permutations. Noticeable differences among global and local saliency maps occur over different layout content out of the same images. The latter aspect suggests that each image gives its contribution to the global visual saliency because of its content and location within the given layout. On top of this consideration, we employ some advertising images to set up a graphical campaign with a given design. We extract relative variance values out the local saliency maps of all permutations. We hypothesise that the inverse of relative variance can be used as an Effectiveness Score (ES) to catch those layout content permutations showing the more balanced spatial distribution of salient pixel. A group of 20 participants have run some eye-tracking sessions over the same advertising layouts to validate the proposed method.

http://eprints.bournemouth.ac.uk/35890/

https://doi.org/10.1007/978-3-030-68790-8_39

Source: Manual