Low-cost image indexing for massive database

Abstract

According to the trend in the modern society that utilizes various image related services and products, the amount of images created by diverse personal and industrial devices is immeasurably voluminous. The adoption of very uncommon or unique identifiers or index attributes with admissible storage requirement and adequate data representation enables massive image databases to process demanded operations effectively and efficiently. For the last decades, various content-based image retrieval techniques have been studied and contributed to support indexing in large digital image databases. However, the complexity, high processing cost of the content-based image retrieval techniques might create inefficiency regarding the configuration of a high-performance image database even though satisfying their own objectives. Moreover the indexing methods with the property of low cardinality might need additional indexing in order to provide strong uniqueness. In this paper, we present identifier generation methods for indexing which are efficient and effective in the perspective of cost and indexing performance as well. The proposed methods exploit the distribution of line segments and luminance areas in an image in order to compose identifiers with high cardinality. From the experimental evaluation, we've learned that our approaches are effective and efficient regarding processing time, storage requirement and indexing performance.