Unsupervised Object Discovery from Images by Mining Local Features Using Hashing

摘要

In this paper, we propose a new methodology for efficiently discovering objects from images without supervision. The basic idea is to search for frequent patterns of closely located features in a set of images and consider a frequent pattern as a meaningful object class. We develop a system for discovering objects from segmented images. This system is implemented by hashing only. We present experimental results to demonstrate the robustness and applicability of our approach.