PATSI - Photo Annotation through Similar Images

Michal Stanek, Oskar Maier, Halina Kwasnicka
(Wroclaw University of Technology, Poland)

Abstract:

Automatic Image Annotation (AIA) is an important open problem in computer vision field, it can improve the performance of images retrieval from large image collections by text queries. A family of baseline methods proposed by Makadia et al. (2008) share the assumption that visually similar images are likely to share the same annotations. The annotation process relies on transferring labels from a number of the nearest neighbors. Handicaps are the need to manually set the neighborhood size and especially the restriction of the annotation length to a fixed, predetermined size. We present a method for Photo Annotation through Finding Similar Images (PATSI) based on the hypothesis that similar images should share a large part of the annotations inspired by Makadia et al. (2008). We incorporate the nearest neighbor approach and keep our method as simple as possible. Our method is designed towards solving the problem of choosing the appropriate length of annotation assigned to the target image. Additionally we propose a transfer parameter optimization method which tunes the resulting word count associated with the image, eliminating the need of manually setting the parameters. On top of this we investigate a number of interesting and representative transfer functions and PATSIs sensitivity to a variety of distance measure and feature set selections. We also present experimental results for benchmark data sets.