基于眼动追踪的个性化图像推荐研究

项目名称： 基于眼动追踪的个性化图像推荐研究

项目编号： No.61473243

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 自动化技术、计算机技术

项目作者： 池哲儒

作者单位： 香港理工大学深圳研究院

项目金额： 79万元

中文摘要： 个性化图像推荐旨在根据用户的意图和偏好给用户提供更感兴趣的图像检索结果。个性化图像推荐目前是图像检索的一个热点，它能为互联网用户带来更舒适的图像搜索体验。与传统的通过鼠标键盘获取用户反馈的方法不同，本研究提出通过对用户基本无干扰的眼动追踪来实现个性化图像推荐，目的是通过眼动数据分析用户观看图像时的隐性反馈信息，从而推测用户的意图和偏好，为用户提供更好的图像检索结果。首先，拟设计人性化的图像检索界面，并且利用眼动仪记录用户在自由和受控两种不同模式下观看图像的眼动数据；然后，针对眼动数据特征提取困难，拟利用特征学习性能优异的深度神经网络从眼动数据中提取有用的特征；接着，结合图像的眼动特征和视觉特征，对图像进行相关性分类和获取用户隐性反馈；最后，向用户作出个性化图像推荐。本研究还将对所建系统进行全面的性能评估。

中文关键词： 眼动追踪；个性化图像推荐；深度神经网络；眼动数据处理；相关反馈

英文摘要： Personalized image recommendation aims to provide an user with better image retrieval results according to the user's intention and preference. Personalized image recommendation is a hot topic of image retrieval. It can offer more comfortable experience of image searching service for the Internet users. In contrast to the traditional methods that collect users' feedback using a keyboard and/or a mouse, this proposal explores non-invasive eye tracking for performing personalized image recommendation. The purpose is to indentify an user's intention and preference by analying the user's eye tracking data generated when he or she is browsing images, such that better image retrieval results could be generated to the user. First of all, an user-friendly image retrieval interface is designed. The eye movement data of an user generated when he or she is browsing images are recorded by an eye tracker with little interference to the user's viewing process. In light of that the features of eye tracking data are difficult to extract, we propose to exploit deep neural networks which are powerful for feature learning to extract eye movement features from eye tracking data. After that, the eye movement features and the visual features of images are combined for performing relevance image identification and obtaining the user's implicit feedback. At last, personalized image recommendation is conducted.This study will also comprehensively evaluate the performance of our proposed personalized image recommendation system.

英文关键词： eye tracking;personalized image recommendation;deep neural networks;eye tracking data processing;relevance feedback