Multi-scale diffusion salient target detection method based on background and target prior

Abstract

The invention discloses a multi-scale diffusion salient target detection method based on background and target priors. The multi-scale diffusion salient target detection method comprises the steps of:firstly, segmenting an image into super-pixels at different scales by utilizing a simple linear iterative clustering algorithm; secondly, regarding the periphery of the image as a background prior, and calculating a Euclidean distance between each pixel and background super-pixels in a CIELAB color space to obtain a background saliency map; thirdly, using target property as prior information to obtain a foreground saliency map; fourthly, calculating background saliency and target saliency of each super-pixel on each scale by means of Bayesian inference, so as to obtain a saliency map fusing the foreground and background priors; fifthly, selecting a manifold sorting method to propagate the saliency of each super-pixel into the whole image to obtain a spatially optimized saliency map; and finally, constructing a pixel-level saliency map through weighted summation of the saliency values at different scales. The experimental results show that the multi-scale diffusion salient target detection method disclosed by the invention can detect the salient targets more effectively than the conventional methods on four kinds of common reference data sets.

Description

technical field [0001] The invention relates to a multi-scale diffusion salient target detection method based on background and target prior, and belongs to the technical field of image salient target detection. Background technique [0002] When processing a huge amount of input information, human vision uses the attention mechanism to filter out some of the most valuable data for priority processing. Inspired by this, the computer establishes a saliency detection model to automatically select the most interesting part of the image scene to reduce the complexity of subsequent analysis and the amount of processing calculations. Early saliency models tended to focus on human vision. Salient object detection has received more attention because it can obtain a relatively complete overall object and has a wide range of applications in image segmentation, object recognition, and image retrieval. [0003] In recent years, the surrounding regions of images have acted as background...

AI Technical Summary

Problems solved by technology

[0005] The technical problem to be solved by the present invention is to provide a multi-scale diffusion salient target detection method based on background and target priors, which solves the problems of errors and inaccurate single-scale detection that may be caused when the target appears on the edge

Images