Salient target detection method based on edge guidance and multilevel feature dynamic aggregation

Abstract

The invention belongs to the field of computer vision, and provides a saliency target detection method based on edge guidance and multi-level feature dynamic aggregation, which comprises the following steps of: 1) performing initial multi-level feature extraction and coding on an RGB (Red, Green and Blue) image by utilizing a VGG-16 trunk framework; 2) strengthening the multi-level features by using a multi-level feature dynamic aggregation module, and generating an initial saliency map; 3) generating a high-quality edge information graph from low-level features in the backbone network by using an edge prediction module; and 4) performing efficient fusion on the initial saliency map and the edge information map to enhance the edge information of the saliency target, and generating a high-performance final saliency map. Compared with the prior art, the saliency target detection method based on edge guidance and multi-level feature dynamic aggregation has the advantages that the expression ability of multi-level features is greatly improved and optimized by integrating two modules, and the quality of the saliency map is further improved.

Description

Technical field: [0001] The invention relates to the field of image processing, in particular, to a saliency target detection method based on edge guidance and dynamic aggregation of multi-level features. Background technique: [0002] The statements in this section merely relate to the background art related to the present invention, and do not necessarily constitute prior art. [0003] With the rapid development of multimedia data technology, network technology, camera equipment and digital storage technology, the capacity of digital image data obtained by people is increasing. There is also a higher demand for image information processing. Inspired by the human visual system, salient object detection came into being. [0004] According to whether the feature selection is based on human prior knowledge, salient target detection can be roughly divided into two categories, one is the traditional stimulus-driven salient target detection, and the other is data-driven deep le...

AI Technical Summary

Problems solved by technology

However, they are all static feature fusion strategies, and it is impossible to judge the pros and cons of the feature fusion, and the pros and cons of the features will directly lead to the performance of the saliency map.

At the same time, although the existing methods can locate salient objects more accurately, the edge information of the segmented salient objects is still not rich enough.

Images