A multi-step approach for partial similarity search in large image data using histogram intersection

Abstract

Listen

Translate article

We investigate the problem of retrieving partially similar images from a large image database. The region-based image retrieval technique is a method of retrieving partially similar images and has been proposed as a way to efficiently process queries in an image database In region-based image retrieval, region matching is indispensable for computing the partial similarity between two images because the query processing is based upon regions instead of the entire image. A naive method of region matching is a pairwise comparison between regions; this causes severe overhead and deteriorates the performance of query processing. In this paper, we focus on the development of a filtering function for the reduction of overall search time in region-based image retrieval, which is of special importance in the case of retrieving partially similar images from a large image database. To prune irrelevant images in a database, we introduce a correct and efficient similarity function by using the Histogram Intersection, which is needed for a crude selection based on a lower bounding property. Subsequently the result is refined by the pairwise region comparison between the query image and selected images. We have performed extensive experiments on synthetic and real image data to evaluate our proposed method. The experimental results reveal that our proposed technique achieves a significant pruning of up to 99% of irrelevant images and is up to 22 times faster than pairwise comparison, where the number of bins is set at 100.