Multi-class target cooperative positioning method based on graph regularization multi-view feature embedding

Abstract

The invention relates to a multi-class target cooperative positioning method based on image regularization multi-view feature embedding, and the method comprises the steps: carrying out the fusion ofa cooperative saliency map of each input image through employing a method based on image regularization multi-view feature embedding, so as to obtain a fusion saliency map with regularization items; performing coarse segmentation on the fused saliency map by using the learned foreground priori knowledge and a GrabCut algorithm; and finally, refining the segmentation through the connectivity of thegraph to obtain a target cooperative positioning result. According to the invention, the positioning precision of multiple types of targets can be improved.

Description

technical field [0001] The invention relates to the technical field of target positioning in images, in particular to a multi-type target collaborative positioning method based on graph regularization and multi-view feature embedding. Background technique [0002] Separating objects of interest from the background is a key step in many computer vision tasks, such as object-based image retrieval, image classification, object recognition, etc. When working with a single image, human annotation or visual saliency are common methods to extract the foreground. However, image annotation is time-consuming and labor-intensive. When dealing with images with cluttered backgrounds or indistinct foregrounds, the saliency of a single image may have limited performance gains. Therefore, a variety of methods for simultaneously processing images, such as co-saliency, co-segmentation, and co-localization, have emerged in recent years. The common idea of ​​these methods is to compensate fo...

AI Technical Summary

Problems solved by technology

On the one hand, a set of images may have multiple objects at the same time, but existing methods tend to locate the dominant object rather than all objects

On the other hand, with the increase in the number of images or the complexity of image scenes, existing localization methods often cannot provide effective and accurate localization results

Images