Image salient object segmentation method and device based on foreground-background mutual relation

Abstract

The embodiment of the invention provides an image salient object segmentation method and device based on a foreground-background mutual relation, and the method comprises the steps of obtaining a feature map corresponding to a training image based on a convolutional neural backbone network, and obtaining a foreground-background initial feature response according to the feature map corresponding to the training image; obtaining a mutual attention weight matrix according to the foreground-background initial feature response, and updating the foreground-background initial feature response according to the mutual attention weight matrix to obtain a foreground feature map and a background feature map; based on a cross entropy loss function and a cooperative loss function, training the convolutional neural backbone network according to the foreground feature map and the background feature map to obtain a foreground-background segmentation convolutional neural network model; and inputting the to-be-segmented image into the foreground-background segmentation convolutional neural network model to obtain a foreground prediction result and a background prediction result. According to the embodiment of the invention, the salient object is segmented from the perspective of the foreground and background cooperation, and the segmentation effect of the salient object is improved.

Description

technical field [0001] Embodiments of the present invention relate to the field of computer vision, and in particular to a method and device for segmenting salient objects in an image based on the relationship between foreground and background. Background technique [0002] Salient object detection usually aims to detect the most salient objects in a scene and accurately segment the entire contours of these salient objects. Many areas in computer vision and image processing can be enhanced by employing saliency segmentation methods, such as content-aware image editing, visual tracking, person re-identification, and image retrieval. Although a large number of methods have been proposed to detect salient objects utilizing different hand-designed features, it is still a great challenge to detect salient objects in complex scenes. [0003] In recent years, with the continuous development of convolutional neural networks, the learned features have more powerful expressive power,...

AI Technical Summary

Problems solved by technology

[0005] Embodiments of the present invention provide a method and device for salient object segmentation in an image based on the relationship between foreground and background, to solve the problem that the existing technology only relies on the feature expression of the foreground in the image, resulting in poor salient object segmentation.

Images