Extraction of image similarity by clustering of word vectors made from attached text and its evaluation

摘要

This paper proposes a method to determine a similarity measure between images by using texts which are attached to images. A text is transformed a word vector. A clustering algorithm is applied to both a set of word vector and a set of vectors made from images independently. Then each pair data of image and text makes a linkage between clusters of two kinds. The proposed similarity is defied in a circle across clusters of two kinds by through of partition of image. Entry of a single image or text into the circle realizes an image retrieval.