Salient target detection method based on refined spatial consistency two-stage graph

Abstract

The invention discloses a salient target detection method based on a refined spatial consistency two-stage graph, and mainly solves a problem that a salient target cannot be completely and consistently detected in a complex scene in the prior art. The implementation scheme of the invention is that the method comprises the steps: 1, segmenting an input image, and obtaining a number of superpixels;2, constructing a first stage graph; 3, constructing a weighted joint robust sparse representation model; 4, solving an optimal solution of the weighted robust sparse representation model; 5, calculating a significant value of each node in the first stage graph; 6, refining the spatial consistency and constructing a second stage graph; 7, calculating the significant value of each node in the second stage graph through a popular ranking model; 8, calculating the significant value of each superpixel of the input image; 9, outputting a pixel-level saliency map of the input image. The method is better in foreground segmentation and background inhibition effect, can achieve the complete and consistent detection of the salient target in a complex scene, and can be used for an image preprocessingin computer vision.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to a salient target detection method, in particular to a salient target detection method based on a refined spatial consistency two-stage graph, which can be used in the image preprocessing process in computer vision. technical background [0002] Salient object detection aims to detect objects in the scene that attract the attention of the human eye and are significantly different from the surrounding area, and segment the object from the scene. As an important image preprocessing method, salient object detection has been widely used in image processing fields such as image segmentation, image restoration, and object recognition. [0003] Existing object detection methods can be divided into two categories: one is supervised, task-specific top-down salient object detection methods, and the other is unsupervised, task-independent bottom-up While the salient object detection ...

AI Technical Summary

Problems solved by technology

However, this method only considers the consistency of the neighborhood space, that is, the current node is connected to its adjacent nodes, and essentially only performs two simple calculations on a graph

Since only the spatial consistency of the neighborhood is considered, there may be "holes" inside the salient objects in the obtained saliency map, and the salient objects cannot be detected completely and consistently.

In addition, the consistency of neighborhood space can generally describe the relationship between nodes in simple scenes, but in complex scenes, such as the foreground and background are very similar or the background is very complex, it is difficult to describe the relationship between nodes well , so it is often impossible to detect salient targets

Images