Saliency detection method based on manifold Ranking and combining foreground and background features

Abstract

The invention provides a Manifold Ranking-based foreground- and background-characteristic combined saliency detection method. The method comprises the following steps: S1, obtaining image data, performing manifold ranking for foreground characteristics and background characteristics of the image data, obtaining saliency values of super-pixel portions in each image data, and obtaining saliency maps of the image data; S2, performing binarization on the saliency maps of the foreground and background characteristics of the image data, and obtaining foreground seeds of the foreground characteristics and background seeds of the background characteristics of the image data; and S3, performing manifold ranking taking the foreground and background seeds as final query nodes, performing calculations for candidate query nodes, ranking calculation results to obtain final ranking values, and obtaining the saliency maps in the image data according to the ranking values. The method fuses a global contrast degree with background characteristics taking boundary information as a reference, and is combined with a manifold ranking algorithm for several times to provide the accurate query nodes for saliency value calculations.

Description

technical field [0001] The invention relates to the field of computer image feature extraction, in particular to a saliency detection method based on Manifold Ranking and combining foreground and background features. Background technique [0002] The saliency detection of the image (Saliency Detection) simply refers to separating the salient area (foreground) in the picture from the non-salient area (background). As an image preprocessing technology, it can be used in later image segmentation, object detection, adaptive image compression, content-based image retrieval, video target detection and other fields. Saliency detection algorithms can be divided into two categories: eye movement prediction models and salient object detection models. The eye movement point detection model is mainly to detect the salient position in the image that attracts the attention of the human eye, so as to analyze and guide human attention; while the salient target detection model is to accurat...

AI Technical Summary

Problems solved by technology

[0004] However, the RC algorithm does not reflect the important influence of the spatial distribution of regional features on the saliency, that is, the more concentrated the feature distribution, the greater the saliency, and the effect is general when processing images with complex texture backgrounds

Images