Please use this identifier to cite or link to this item: http://scholarbank.nus.edu.sg/handle/10635/143315
Title: LEARNING VISUAL ATTENTION WITH DEEP NEURAL NETWORKS
Authors: SHEN CHENGYAO
Keywords: Visual Attention, Visual Saliency, Deep Neural Nework
Issue Date: 21-Jan-2016
Citation: SHEN CHENGYAO (2016-01-21). LEARNING VISUAL ATTENTION WITH DEEP NEURAL NETWORKS. ScholarBank@NUS Repository.
Abstract: VISUAL ATTENTION DESCRIBES THE ABILITY OF OUR VISUAL SYSTEM TO RAPIDLY SELECT THE MOST SALIENT PARTS IN THE SCENE AND THUS THE MOST RELEVANT DATA ABOUT THE ENVIRONMENT. THE EXISTING COMPUTATIONAL MODELS OF VISUAL ATTENTION PREDICT HUMAN EYE FIXATIONS BASED ON EARLY FEATURES SUCH AS COLOR, LUMINANCE AND EDGE ORIENTATION, AND MOST OF THEM ARE USUALLY DEFICIENT IN PREDICTING ACCURATE EYE FIXATIONS WHEN THE SCENE CONTAINS SEMANTIC CONTENTS SUCH AS FACES, TEXTS, OR OTHER OBJECTS ESPECIALLY THOSE SOCIALLY MEANINGFUL ONES. IN THIS THESIS, WE EXPLORE THE POWER OF DEEP NEURAL NETWORKS TO LEARN FEATURES UNDERLYING ATTENTION IN A WAY SIMILAR TO THE HIERARCHICAL VENTRAL PATHWAY, AND TO PREDICT EYE FIXATIONS ON NATURAL IMAGES USING THE FEATURES. SPECIFICALLY, WE DESIGN VARIOUS DEEP NEURAL NETWORK ARCHITECTURES ATTENTION PREDICTION ON NATURAL IMAGES AND WEBPAGES. MODELS INCLUDE: (1) A MULTI-LAYER SPARSE NETWORK WITH UNSUPERVISED FEATURE LEARNING, (2) A MULTI-SCALE REGRESSION NETWORK WITH SALIENC
URI: http://scholarbank.nus.edu.sg/handle/10635/143315
Appears in Collections:Ph.D Theses (Open)

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
ShenCY.pdf50.9 MBAdobe PDF

OPEN

NoneView/Download

Page view(s)

14
checked on Aug 30, 2018

Download(s)

9
checked on Aug 30, 2018

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.