Boundary connectivity and local comparison-combined significance testing method

Abstract

The invention belongs to the field of image processing and relates to a boundary connectivity and local comparison-combined significance testing method. The boundary connectivity and local comparison-combined significance testing method solves the problem of image significance testing. The method comprises, firstly, performing superpixel division through an SLIC (simple linear iterative cluster) algorithm, and through Harris corner detection and by means of features of local comparison, acquiring convex hulls enclosing foreground areas; secondly, eliminating background areas in the convex hulls through a clustering algorithm, taking the acquired foreground areas as absorbing nodes of a random walk model, and performing clustering propagation optimization to acquire the foreground probability of every superpixel; thirdly, through the features of boundary connectivity of every area, computing the background probability of every superpixel; finally, combining the foreground probability and the background probability of every superpixel to acquire a significance image, and inhibiting background superpixels significant value to acquire a final significance image. The boundary connectivity and local comparison-combined significance testing method can help identify the most significant parts of an image and acquire the significance image closer to a true-value image.

Description

technical field [0001] The invention belongs to the field of image processing and relates to a saliency detection method combining boundary connectivity and local contrast. Background technique [0002] The purpose of image saliency detection is to find the most salient parts in the image. The salient parts indicate which areas in the image can attract people's attention and the degree of attention. Finding salient parts efficiently and quickly can greatly improve the efficiency of image processing. Saliency detection algorithms can be divided into two categories: top-down methods and bottom-up methods. Top-down is usually aimed at specific tasks, using a supervised way to learn various features of the target, and using the learned feature information to complete the recognition of salient targets in the image. The disadvantage of this type of method is that it can only complete specific targets. And must pass the training, the expansibility is poor. The bottom-up method c...

AI Technical Summary

Problems solved by technology

In recent years, random walk models have also been used in image saliency detection. For example, Sun et al. used superpixels on the left and upper borders of images as absorbing nodes in the random walk model to calculate the initial saliency value of each superpixel. The practice is actually to use the boundary prior, but the effect is not satisfactory when the salient object appears at the edge of the image

Images