Efficient local feature descriptor filtering

Abstract

The present disclosure generally relates to methods and computer program products for searching for a similar image among a plurality of stored images, and in particular to a method and computer program product used in a content based image retrieval system where roughly similar images are clustered and feature vectors for the clustered images are filtered based on a matching frequency for the feature vectors among the images in the cluster.

Description

TECHNICAL FIELD[0001]The present disclosure generally relates to methods and computer program products for searching for a similar image among a plurality of stored images, and in particular to a method and computer program product used in a content based image retrieval system where roughly similar images are clustered and feature vectors for the clustered images are filtered based on a matching frequency for the feature vectors among the images in the cluster.TECHNICAL BACKGROUND[0002]Searching for images similar to a certain image among a plurality of images is a well-used feature in today's image retrieval system. Most traditional and common methods of image retrieval utilize some method of manual image annotation which rely on that a person manually adds metadata such as keywords to the images so that retrieval can be performed over the metadata words. Manual image annotation is time-consuming, laborious and expensive.[0003]Instead, Content-based image retrieval (CBIR) system m...

AI Technical Summary

Benefits of technology

The present invention is about a way to quickly and accurately search for images that are similar to each other. It uses clusters of images that have already been grouped together to guide the filtering algorithm. This makes a very efficient system for finding similar objects, faces, buildings, and so on. The method uses one feature vector to represent the image, which is calculated based on things like color or texture. This feature vector should ideally capture the main idea of the image. By using clusters, the method is quick and efficient, and can handle large amounts of data.

Problems solved by technology

Manual image annotation is time-consuming, laborious and expensive.

One problem that often occurs in such CBIR system is the sheer computational complexity to compare all features descriptors from one image, with all the feature descriptors from all the other images.

Comparing 20000 SIFT vectors in an image, with the same amount of vectors in all images in a system, e.g. 10000 images, results in a very high computational load and a slow CBIR system.

Images