Khan J I, Yun D Y
Department of Electrical Engineering, University of Hawaii at Manoa, Honolulu 96822, USA.
Comput Med Imaging Graph. 1996 Jul-Aug;20(4):243-57. doi: 10.1016/s0895-6111(96)00017-1.
This paper presents an associative technique for content-based retrieval into image archive, based on a computing paradigm called Multidimensional Holographic Associative Computing (MHAC). Unlike any prior Artificial Associative Memory (AAM), MHAC has the unique ability fo focus on any subject of pixels in the sample image and retrieve learned images based on the similarity of the visual objects. In addition, MHAC is adaptive, graciously accommodative of imprecision, efficient, parallelizable, scalable and optically realizable. Together, these excellent properties of MHAC offer a promising novel approach to a content-based search into massive image archives. The paper presents the necessary transformational steps to incorporate this new mechanism into a complete image archival and retrieval system. This is the first associative search approach for content-based retrieval in image repository. The results show that this search system is capable of retrievals by using pattern objects as small as 10-15% of the query image frame at better than 90% accuracy. This demonstrates the potential of MHAC for handing contest-based image applications far beyond the capability of current associative memories. The design, methodology and performance of this system have been illustrated in this paper through its application in managing a Medical Image Archive (MEDIA).
本文提出了一种基于名为多维全息关联计算（MHAC）的计算范式，用于基于内容检索图像存档的关联技术。与任何先前的人工关联记忆（AAM）不同，MHAC具有独特的能力，能够聚焦于样本图像中任何像素主题，并基于视觉对象的相似性检索已学习的图像。此外，MHAC具有适应性，能很好地适应不精确性，高效、可并行化、可扩展且可光学实现。总之，MHAC的这些优异特性为基于内容搜索海量图像存档提供了一种有前景的新方法。本文介绍了将这种新机制纳入完整图像存档与检索系统所需的转换步骤。这是图像存储库中基于内容检索的首个关联搜索方法。结果表明，该搜索系统能够使用小至查询图像帧10%-15%的模式对象进行检索，准确率超过90%。这证明了MHAC在处理基于内容的图像应用方面的潜力，远远超出了当前关联记忆的能力。本文通过其在管理医学图像存档（MEDIA）中的应用，阐述了该系统的设计、方法和性能。
