色彩加权相关法的立体匹配

需积分: 9 100 浏览量更新于2024-09-04 收藏 497KB PDF 举报

"立体匹配算法的研究，通过色彩加权相关性、分层信念传播和遮挡处理来解决" 在计算机视觉领域，立体匹配是一项基础且重要的研究主题，它涉及到多个图像之间对应像素的深度估计，从而实现三维场景的理解。《Stereo Matching with Color-Weighted Correlation, Hierarchical Belief Propagation, and Occlusion Handling》这篇论文提出了一种新的算法，旨在处理立体匹配问题，特别关注了不连续性、遮挡和视差处理。该算法基于能量最小化框架构建了一个全局匹配的立体模型。全局能量包含两个主要部分：数据项和平滑项。数据项首先通过色彩加权相关性进行近似计算，这意味着算法会考虑像素的颜色信息来提高匹配的准确性。这种色彩权重的引入有助于在相似颜色区域找到更准确的对应点，尤其是在低纹理区域，颜色信息往往是区分特征的关键。随后，为了处理遮挡和低纹理区域，论文中采用了一种分层循环信念传播算法进行细化。分层信念传播是一种优化方法，能够逐级处理图像的细节，对于复杂场景中的遮挡和边缘，它可以更好地恢复正确的视差信息。通过重复应用此算法，系统能逐步识别并修正遮挡区域的错误匹配，提升匹配质量。实验结果在Middlebury数据集上进行了验证，证明了所提出的算法在性能上优于其他现有方法。Middlebury数据集是立体匹配领域的标准测试集，包含各种复杂的场景和挑战，因此这一结果具有很高的可信度。这篇论文为立体匹配带来了创新性的方法，结合了色彩信息、分层优化和遮挡处理，对提高立体匹配的精度和鲁棒性做出了贡献。这些技术对于自动驾驶、机器人导航、虚拟现实等依赖于三维信息的应用有着深远的影响。未来的研究可能会进一步探索如何将这些方法与其他先进的深度学习技术相结合，以提升立体匹配的自动化程度和实时性能。

Stereo Matching with Color-Weighted Correlation, Hierarchical Belief

Propagation and Occlusion Handling

Q`ıngxi´ong Y´ang Liang Wang Ruigang Yang Henrik Stew´enius David Nist´er

Center for Visualization and Virtual Environments

Department of Computer Science, University of Kentucky

http://www.vis.uky.edu/

∼

liiton/

Abstract

In this paper, we formulate an algorithm for the stereo

matching problem with careful handling of disparity, dis-

continuity and occlusion. The algorithm works with

a global matching stereo model based on an energy-

minimization framework. The global energy contains two

terms, the data term and the smoothness term. The data

term is ﬁrst approximated by a color-weighted correlation,

then reﬁned in occluded and low-texture areas in a repeated

application of a hierarchical loopy belief propagation algo-

rithm. The experimental results are evaluated on the Mid-

dlebury data set, showing that our algorithm is the top per-

former.

1. Introduction

Stereo is one of the most extensively researched topics in

computer vision. Stereo research has recently experienced

somewhat of a new era, as a result of publically available

performance testing such as the Middlebury data set [

11],

which has allowed researchers to compare their algorithms

against all the state-of-the-art algorithms.

In this paper, we describe our stereo algorithm, which

is currently evaluating as the top performer on the Mid-

dlebury data set. The algorithm springs from the popular

energy minimization framework that is the basis for most

of the algorithms on the Middlebury top-list, such as graph

cuts [

4, 10] and belief propagation [12, 13]. In this frame-

work, there is typically a data term and a smoothness term,

where the data term consists of the matching error implied

by the extracted disparity map, and the smoothness term

encodes the prior assumption that world surfaces are piece-

wise smooth.

However, the algorithm presented in this paper departs

somewhat from the normal framework, in that in the ﬁnal

stages of the algorithm, the data term is updated based on

the current understanding of which pixels in the reference

image are occluded or unstable due to low texture.

The paper is organized as follows: Section

2 gives a

high-level overview of the approach. In Section

3 we then

give the detailed equations for all the building blocks. Sec-

tion 4 reports results supporting the claims that the algo-

rithm is currently the strongest available on the Middlebury

data set. Section

5 concludes.

2. Overview of the Approach

The algorithm can be partitioned into three blocks, ini-

tial stereo (Figure 1), pixel classiﬁcation (Figure 2) and iter-

ative reﬁnement (Figure

3). In the initial stereo, see Figure

1, the correlation volume is ﬁrst computed. A basic way

to construct the correlation volume is to compute the abso-

lute difference of luminances of the corresponding pixels in

the left and right images, but there are many other meth-

ods for correlation volume construction. For instance, Sun

et al. [

12] use Birchﬁeld and Tomasi’s pixel dissimilarity

[1] to construct the correlation volume, and Felzenszwalb

[

6] suggests to smooth the image ﬁrst before calculating the

pixel difference. In this work, we are using color-weighted

correlation to build the correlation volume, in a similar man-

ner as was recently described by Yoon and Kweon [

17]. The

color-weighting makes the match scores less sensitive to oc-

clusion boundaries by using the fact that occlusion bound-

aries most often cause color discontinuities as well. The

initial stereo is run in turn with both the left and the right

image as reference images. This is done just to support

a subsequent mutual consistency check (often called left-

right check) that takes place in the pixel classiﬁcation block.

Functions E

and E

deﬁning the smoothness costs in the

left and right reference images, respectively, are determined

based on the color gradients in the input images. The left

and right smoothness costs and the left and right correlation

costs are then optimized using two separate hierarchical be-

lief propagation processes. The hierarchical belief propaga-

tion is performed in a manner similar to Felzenszwalb [

6],

resulting in the initial left and right disparity maps D

(0)

and

下载后可阅读完整内容，剩余7页未读，立即下载

tf_zxq

粉丝: 4
资源: 2

色彩加权相关法的立体匹配

立体匹配经典文献

Group-wise Correlation Stereo Network.pdf

stereomatching-v2-2001数据集

Guo-Group-Wise-Correlation-Stereo-Network-CVPR-2019-paper

PatchMatch Stereo - Stereo Matching with Slanted Support Windows

stereo-matching-master_立体匹配_stereomatching_源码

Stereo-Matching-Research

Segment-tree based cost aggregation for stereo matching with enhanced segmentation

bp-stereo matching-Matlab

A stereo matching algorithm using multi-peak candidate matches and geometric constraints

最新资源