Colour-based relevance feedback for image retrieval
Wai-Chee Low, Tat‐Seng Chua · 2002
Colour is an important attribute for image matching and retrieval, and relevance feedback (RF) has been found to improve the retrieval effectiveness in information retrieval systems significantly. We propose a new approach for colour-based RF for image retrieval. The approach makes use of the Multi-Interval Discretization Algorithm to discretize the range of colour histogram values into discrete colour intervals. The probabilistic 0.5 formula is adapted to identify significant colour intervals which are then used to refine the initial query. In addition, knowledge is extracted from the relevance judgement information and represented as a colour interval decision tree. The decision tree is used to aid in classifying the images, retrieved using the modified query in subsequent retrieval, into the relevant and non-relevant sets. Extensive tests on a large image collection were conducted to demonstrate the effectiveness of our proposed RF approach.