A Cross-Domain Visual Retrieval Method Based on Saliency Detection

Abstract

The invention discloses a cross-domain visual retrieval method based on saliency detection. First, use the boundary connection value of each superpixel area to assign different salience values ​​to each area to obtain the main target area; then perform multi-scale processing on the target image in the database, extract features from the main target area, and obtain the target image feature template; Feature extraction and linear classifier training are performed on the main target area of ​​the query image, and the optimized query image feature template is obtained through a large number of negative sample iterative training; in the final retrieval, according to the relationship between each target image feature template and the query image feature template Matching degree, return the area with the highest response score as the final retrieval result. The invention reduces the influence of the background region on the retrieval result by detecting the salience of the main body region, effectively improves the retrieval accuracy and efficiency in the cross-domain visual retrieval, and has good robustness.

Description

technical field [0001] The invention belongs to the technical field of image processing and computer vision, and in particular relates to a cross-domain visual retrieval method based on saliency detection. Background technique [0002] Cross-domain Visual Retrieval (Cross-domain Visual Retrieval) is one of the very promising technologies in the field of computer vision. With the rapid development of imaging sensor performance and the continuous enrichment of types, the means of acquiring images of the same thing are becoming more and more diverse, and the number of various types of images is also growing exponentially. In order to make full use of these digital resources, it is often necessary to match and retrieve cross-domain images of the same thing acquired under different imaging conditions or different carriers. For example: the retrieval of oil paintings to natural photos of the same building on the Internet, the police need to match the sketches of suspects with the...

AI Technical Summary

Problems solved by technology

In recent years, visual retrieval technology has been continuously improved and developed, but there are not many researches on visual retrieval algorithms between cross-domain images.

In 2008, the Second Artillery Equipment Research Institute proposed a region-based matching retrieval algorithm and a feature-based matching retrieval algorithm for cross-domain images presented by different sensors (visible light, infrared, radar), but these two methods are only applicable to The retrieval of three specific domain images has a limited scope of application and is not suitable for retrieval of cross-domain images in complex scenarios

In 2011, Carnegie Mellon's research team proposed a data-driven cross-domain matching retrieval method, which uses the concept of machine learning to train and optimize feature vectors, but the single feature vector extraction method and the increase in scene complexity will greatly reduce Accuracy of matching searches

Although this method has improved the retrieval accuracy, the interference of the complex background often causes the target to be wrongly retrieved as the background area.

This happens mainly because the existing cross-domain retrieval technology does not take into account the different importance of the target area and the background area in the image for retrieval.

Images