Building image segmentation method for augmented reality application

Abstract

The invention discloses a building image segmentation method for augmented reality application. The method comprises the steps that building image features which comprise a gradient angle entropy feature, a color entropy feature and a line feature are designed; training samples are selected, and improved K-means is used to cluster training samples to select samples in disjoint cluster sets as training samples; Adaboost decision classifier training is carried out, and selected samples are input to train an Adaboost decision classifier; and a test image is input into the Adaboost decision classifier, and a building segmentation result is output. Compared with a traditional segmentation method, the method provided by the invention has the advantages that a building image feature description method is designed, and the gradient angle entropy is simplified, so that the segmentation efficiency is significantly improved; the classifier based on an Adaboost decision tree is designed to removethe sample redundancy; and a sample selection method is designed, and the final classifier model is trained to quickly and accurately segment building, non-building and non-artificial structures containing building images.

Description

technical field [0001] The invention belongs to the field of computer vision and image processing, and relates to a building image segmentation method for augmented reality applications. Background technique [0002] Modern architecture is an artificial structure that conforms to certain scientific laws and aesthetic structures. In terms of color characteristics, buildings do not have a uniform color range; non-artificial structures, such as plants, land, sky, etc., have a relatively uniform color range. In terms of texture features, the surface texture of buildings has a high degree of uniformity and repetition; non-artificial structures are more chaotic and do not have unity, that is, the entropy of images of artificial structures is low, while the entropy of images of non-artificial structures is high. In terms of shape features, buildings, especially modern buildings, have a large number of straight-line edges and rectangular structures, and most of the straight-line fe...

AI Technical Summary

Problems solved by technology

The normalized cutting algorithm solves the problem of separating a single node of the minimum cut algorithm very well, and the segmentation results are quite satisfactory, but its disadvantage is that the calculation is huge and there is no systematic convergence point, so later scholars put forward many Improved methods, such as dividing the image into blocks, performing normalized cutting within the block; sparse matrix, removing points in the similarity matrix

Normalized segmentation is not detailed enough for the segmentation of complex areas, trees and building stairs are divided into the same area; and for the number of segmented blocks, it has a great impact on the segmentation effect, so normalized segmentation is not suitable for building and non-building segmentation

Images