Visual search modeling based on deep learning and eye tracking

Abstract

Visual elements of package design are visual cues primarily related to product salience. For example, a distinctive package design can make the product stand out from the visual clutter. Product salience has a pervasive effect on visual search performance, the process of locating a target among a set of distractors in a scene, which could frequently happen in our daily life. Many scenarios involve time pressure, which means consumers have insufficient time to collect complete information. We hypothesize that under time pressure, consumers would speed up to put more weights on the spatial search to lower localization uncertainty and avoid look back to previous examined items, leading to increasing search behavior and decreasing revisit behavior pattern. We also propose two models to predict visual search behavior by exploiting recent advances in eye-tracking technology and deep learning. Key Words: Visual search; Package Design; Time pressure; Eye tracking; Deep learning

Date of Award	2021
Original language	English
Awarding Institution	The Hong Kong University of Science and Technology
Supervisor	Ralf VAN DER LANS (Supervisor) & Bertram Emil SHI (Supervisor)