Can Eye Movement Improve Prediction Performance on Human Emotions Toward Images Classification?

Simple item page
dc.contributor.authorPasupa, Kitsuchart
dc.contributor.authorSunhem, Wisuwat
dc.contributor.authorLoo, Chu Kiong
dc.contributor.authorKuroki, Yoshimitsu
dc.date.accessioned2026-08-06T10:15:07Z
dc.date.available2026-08-06T10:15:07Z
dc.date.issued2017-01-01
dc.description.abstractRecently, image sentiment analysis has become more and more attractive to many researchers due to an increasing number of applications developed to understand images e.g. image retrieval systems and social networks. Many studies aim to improve the performance of the classifier by many approaches. This work aims to predict the emotional response of a person who is exposed to images. The prediction model makes use of eye movement data captured while users are looking at images to enhance the prediction performance. An image can stimulate different emotions in different users depending on where and how their eyes move on the image. Two image datasets were used, i.e. abstract images and images with context information, by using leave-one-user-out and leave-one-image-out cross-validation techniques. It was found that eye movement data is useful and able to improve the prediction performance only in leave-one-image-out cross-validation.
dc.identifier.citationLecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics, 10637 LNCS, 830-838, 2017
dc.identifier.doi10.1007/978-3-319-70093-9_88
dc.identifier.issn03029743
dc.identifier.other2-s2.0-85035087555
dc.identifier.urihttps://dspace.kmitl.ac.th/handle/123456789/7229
dc.sourceLecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics
dc.subjectAbstract image
dc.subjectEmotion classification
dc.subjectEye movement
dc.subjectImage with context information
dc.titleCan Eye Movement Improve Prediction Performance on Human Emotions Toward Images Classification?
dc.typeConference Paper

Files

Collections

Research