Locally regressive G-optimal design for image retrieval

Zheng-Jun Zha,Yan-Tao Zheng,Meng Wang,Fei Chang,Tat-Seng Chua
DOI: https://doi.org/10.1145/1991996.1992055
2011-01-01
Abstract:Content Based Image Retrieval (CBIR) has attracted increasing attention from both academia and industry. Relevance Feedback is one of the most effective techniques to bridge the semantic gap in CBIR. One of the key research problems related to relevance feedback is how to select the most informative images for users to label. In this paper, we propose a novel active learning algorithm, called Locally Regressive G-Optimal Design (LRGOD) for relevance feedback image retrieval. Our assumption is that for each image, its label can be well estimated based on its neighbors via a locally regressive function. LRGOD algorithm is developed based on a locally regressive least squares model which makes use of the labeled and unlabeled images, as well as simultaneously exploits the local structure of each image. The images that can minimize the maximum prediction variance are selected as the most informative ones. We evaluated the proposed LRGOD approach on two real-world image corpus: Corel and NUS-WIDE-OBJECT [5] datasets, and compare it to three state-of-the-art active learning methods. The experimental results demonstrate the effectiveness of the proposed approach.
What problem does this paper attempt to address?