Synergistic saliency detection method for rgbd images based on multi-kernel enhancement and saliency fusion

Abstract

The invention discloses a RGBD image cooperative saliency detection method based on multi-core enhancement and saliency fusion. The specific steps are: (1), segment a group of color images with common salient objects into several regions, and calculate the saliency map of a single RGBD image; (2), select samples, and select the class synergy among the optimal clustering results For the class with the largest saliency value, the region in this class is used as a positive sample, and the region with a saliency value lower than the threshold of a single image is used as a negative sample; (3), different training sets are generated by random sampling of images to Learn multiple different models to obtain a co-saliency map based on multi-kernel enhancement; (4), combine the co-saliency map based on multi-kernel enhancement obtained in step (3) and the base co-saliency map obtained in step (2) Linear fusion is used to obtain the fused co-saliency map, and the quality of each fused co-saliency map is evaluated, and the quality evaluation is used as the weight for adaptive fusion to obtain the final co-saliency map of the RGBD image.

Description

technical field [0001] The present invention relates to an image collaborative saliency detection method, in particular to an RGBD image (RGB color image and its corresponding depth image) collaborative saliency detection method based on multi-core enhancement and saliency fusion, aiming to obtain from a group of images with common saliency Co-salient objects are detected in RGBD image sets of objects. Background technique [0002] With the vigorous development of the Internet and multimedia technology, people can obtain image information from more and more fields such as surveillance video, social network, news reports, etc. Therefore, how to capture the key information in the image is the focus of people's attention today. Saliency detection is a challenging field that aims to extract regions of the main objects of interest in images or videos, while co-salient object detection, an emerging branch of visual saliency, aims to collect Extract common salient objects. Co-sa...

AI Technical Summary

Problems solved by technology

In 2015, Fu et al. proposed an object-based collaborative segmentation algorithm that uses depth information, which can effectively deal with noise images that have no common objects or appear more than once. Using several existing RGB saliency detection models with better performance, collaborative saliency The saliency model and the RGBD saliency model are combined according to the rank constraints to obtain the final RGBD co-saliency map, but the correlation of depth information between different RGBD images is ignored.

But in terms of RGBD collaborative saliency detection, there is still a lot of room for development

Images