Visible light image saliency prediction method, device, equipment and storage medium

Abstract

The application provides a visible light image saliency prediction method, device and equipment and a storage medium, and relates to the technical field of computer vision and visual perception. The method generates a basic visual feature representation through a multi-scale feature extraction process without introducing high-dimensional feature interaction when processing an input visible light image; a center-surround modulation mechanism with an entropy gate is introduced in the feature extraction process to adaptively modulate the feature response difference between the center region and the surrounding region at different spatial scales; through an adaptive grouping saliency selection processing mode with controlled parameter scale, the feature channels are divided into multiple groups, and competitive selection and weighted fusion are performed in the grouping dimension to sequentially distribute the multi-peak saliency candidate response; finally, through grouping feature fusion and saliency mapping, a saliency prediction result spatially aligned with the input image is generated. The application is suitable for natural scene visual saliency processing tasks under limited computing resources.

Images