Retrieval characteristics of an untrained holographic index for image databases

This paper describes the retrieval characteristics of an untrained holographic index for image databases. Information from multiple images can be superimposed on a single memory element in the holographic model. Images in the database that match given query images can be retrieved using the index. The untrained model minimizes maintenance overhead of the index if there are frequent additions and deletions of images in the database. We present the impact of encoding parameters on the retrieval performance of the index. We also study retrieval characteristics with both partial and full images as well as when the query image is distorted by significant noise.