Minimizing human-machine interactions in automatic image retrieval

The self-organizing tree map (SOTM) has been successfully applied to image retrieval with automatic relevance feedback, but may provide inaccurate relevance identification when dealing with a complex image database. We developed an improved SOTM method which treats relevance identification as a one (the relevant class) vs. many (the irrelevant classes) problem. Experimental results demonstrated the superior performance of the proposed approach.