Interactive image segmentation method based on weak link constraints

Abstract

The invention provides an interactive image segmentation method based on weak link constraint. The interactive image segmentation method mainly comprises the following steps: 1) performing SLIC super-pixel segmentation preprocessing on a target image; 2) providing paired weak link constraint information by a user; 3) constructing a similarity matrix based on the Euclidean distance of the superpixel center point and the Euclidean distance of the RGB space average value; 4) on the basis of a Bayesian framework, constructing a clustering algorithm model based on representative points by maximizing the likelihood probability of the image; and 5) optimizing the target function by using an EEM algorithm, and then completing an interactive image segmentation task. The weak link constraint and thestrong link constraint are combined together; on the basis of superpixel segmentation. Paired link constraint information is embedded into a clustering model based on representative points, in the optimization process, suboptimal representative points are expanded into an optimization space, a representative point set is modified according to energy loss, image segmentation is completed, and theacquisition cost of link constraints is greatly reduced.

Description

technical field [0001] The invention relates to the technical field of image segmentation, in particular to an interactive image segmentation method based on weak link constraints. Background technique [0002] Image segmentation is an important part of artificial intelligence, computer vision and other fields. However, in many cases, the results of unsupervised image segmentation algorithms lack semantic interpretation and cannot outline a complete target prospect. For example, treat clothes of different colors and materials as different classes, or treat partially overlapping mountains as one class. In order to solve such problems, image preprocessing with superpixel strategy has gradually become a research hotspot in image processing. Superpixel segmentation is an over-segmentation algorithm that treats a set of pixels with similar colors and adjacent positions in an image as a whole, that is, superpixels. Therefore, the complexity of image processing is greatly reduced...

AI Technical Summary

Problems solved by technology

[0004] At present, the algorithms dealing with link constraints all have different limitations. For example, the CEM algorithm (Constraint EM) embeds the Gaussian mixture model into the maximum expectation algorithm, and controls the must-link through an improved maximum expectation algorithm model. At the same time, an invisible The data cluster (cluster) variable is embedded into the Markov Random Field (MRF) to control the cannot-link; however, the CEM algorithm requires the total number of data clusters to be input in advance; the SAP algorithm (Subspace AffinityPropagation) algorithm is a subspace based on representative point aggregation However, this algorithm adds an iterative step, thus reducing the efficiency of the algorithm; another semi-supervised algorithm based on representative points, SSAP (Semi-Supervised Affinity Propagation), adds several variable nodes (meta-point); adjust the similarity matrix to handle linkage constraints

Images