- Book Chapter
7
- 10.1007/3-540-45925-1_7
A Bayesian Method for Content-Based Image Retrieval by Use of Relevance Feedback
- Jan 01, 2002
- Tao Ju-Lan + 1 more +1
This paper proposes a new Bayesian method for content-based image retrieval using relevance feedback. In this method, the problem of content-based image retrieval is first formulated as a two-class classification problem, where each image in the database can be classified as or nonrelevant with respect to the query and the goal is to minimize the misclassification error. Then, the problem of image retrieval is further transferred into a simpler problem of ranking each image in the database by using a similarity measure that is basically a likelihood ratio. Here, the likelihood of the relevant class is modeled by a mixture of Gaussian distribution determined by the positive samples, and the likelihood of the non-relevant class is assumed to be an average of Gaussian kernels centered at negative samples. The experimental results have indicated that the proposed method has potential to become practical for content-based image retrieval.
Read more