Relevance Feedback pada Temu Kembali Citra Menggunakan Bayesian Network

Abstract

Query results not always match with user's desire. The key issue in relevance feedback is how to utilize the feedback information to improve the retrieval performance. This research proposed a method of relevance feedback using the Bayesian network model. Research data consists of 550 images which are grouped into 11 classes. The information concern in this model are histogram-l 62 for color infonnation, gradient for shape information and co-occurrence matrix for texture infOlmation. This research used Rocchio as relevance feedback algorithm. Re-weighting in feature vector used inter-weight and intra-weight updating theory. The weights that have been updated then it was used as posterior probability, and also used as a rank ofrelevant image. Evaluation of retrieval result is computed using average value of precision, the results showed that average value of precision for all classes is increasing about 4.7959% on Bayesian network model with relevance feedback compared with Bayesian network model without relevance feedback. Through the results, this research can improve the retrieval perfonnance in CBIR.