Indexing of Image Databases Using Untrained 4D Holographic Memory Model

Abstract

An interesting application of the holographic memory model is the indexing of multi-media databases for efficient retrieval of matching patterns. Two dimensional holographic memory models have been previously investigated for indexing collections of gray scale images. In this paper, we describe the implementation of a four dimensional holographic memory model that can be used for indexing colour images and video frames. We have investigated the model using both simulations and real image data to determine the effective operational range of the 4D holograph. Our experiments show that a high level of retrieval accuracy is achieved even if fifty percent of pixels in the query image are distorted by noise. The results provide the basis for designing a robust indexing scheme for retrieving matching images from large image databases.