J-049
An Analysis of Image Similarity Estimation based on Eye Information

◎Mario Saputra・Mutsumi Suganuma・Wataru Kameyama（Waseda Univ.）

The purpose of this research is to produce an image similarity map reflecting user subjectivity using eye information measured by eye tracker device. Based on the experiment of comparing images, an image ranking for each subject is made by using pairwise comparison method. Fifteen eye related features have been collected from ten subjects. To these features, PCA and SVD algorithms are applied to get an image similarity map. By comparing image subjective ranking with the corresponding map, we find that the SVD results are better than PCA, reflecting the subjective similarity.