Visual query in medical image archive

Abstract

This paper presents a technique for searching an image archive with visual examples. The technique described in this work is specifically targeted for searching anatomical image archives where it is difficult to obtain a symbolic description of the objects of interest. This technique is based on a new computing method called multidimensional holographic associative computing (MHAC). MHAC introduces object based search in associative computing. Unlike conventional associative computing methods, this new method can perform a search on the basis of scene objects or features present in a sample image. It also retains the usual advantages of associative computing, that is, their computational efficiency, and highly parallel and distributed nature of computation. This paper presents the design, query mechanism and performance of an experimental medical image archive (MEDIA) search system based on this new attention modulated computing technique. The MEDIA search system can retrieve from cue objects as small as 15-10% of the frame size of the sample query image, which far exceeds the capability of previous associative memories.