多特征组合的深度图像分割算法

计算机工程与科学

多特征组合的深度图像分割算法

谭志国1，2,欧建平1,张军1,沈先耿2

（1.国防科技大学电子科学学院，湖南长沙 410073;2.武警警官学院信息通信系,四川成都 610213）

收稿日期:2017-02-04 修回日期:2017-04-25 出版日期:2018-08-25 发布日期:2018-08-25
基金资助:
国家自然科学基金（61471370，61471371）;博士后科学基金（2012M512168）

Multi-feature combined depth

image segmentation algorithm

TAN Zhiguo1，2，OU Jianping1，ZHANG Jun1，SHEN Xiangeng2

(1.College of Electronic Science,National University of Defense Technology,Changsha 410073;

2.Department of Information & Communication,Armed Police College of PAP,Chengdu 610213,China)

Received:2017-02-04 Revised:2017-04-25 Online:2018-08-25 Published:2018-08-25

摘要/Abstract

摘要：

深度图像直接反映景物表面的三维几何信息，且不受光照、阴影等因素的影响，对深度图像处理、识别、理解是目前计算机视觉领域研究的热点和重点之一。针对深度图像信息单一且噪声较大的特点，提出一种基于组合特征的阈值分割算法，实现对深度图像数据的有效分割。算法首先通过梯度特征对图像进行Otsu阈值分割；在此基础上，分别在不同分割区域内利用深度特征进行Otsu多阈值分割，得到候选目标；然后，在空域上利用像素的位置特征对候选目标进行分割、合并与去噪，最终得到图像分割的结果。实验结果表明，该方法能有效克服深度图像中噪声的影响，得到的分割区域边界准确，分割质量较高，为以后的室内对象识别和场景理解工作奠定了较好的基础。

关键词: 深度场景理解, 深度图像分割, Otsu阈值, 梯度特征, 深度特征

Abstract:

Depth image directly reflects the threedimensional geometric information of the scene surface and is not affected by factors such as light and shadow. Processing, recognizing, and understanding depth images are currently one of the hot topics and focuses in the field of threedimensional computer vision. Aiming at the problem that the depth image information is single and the noise is large, a threshold segmentation algorithm based on combined features is proposed to realize effective segmentation of depth image data. The algorithm first performs Otsu threshold segmentation on the image by using gradient features. On this basis, Otsu multithreshold segmentation is performed using depth features in different segmented regions to obtain candidate targets. Then, in the spatial domain, the depth feature is used to segment, merge, and denoise the candidate targets, thus finally obtaining the segmentation results. Experimental results show that this method can effectively overcome the influence of noise in depth images, the obtained boundary of the segmentation area is accurate, and the segmentation quality is high, which lays a good foundation for future indoor object recognition and scene understanding.

Key words: depth scene understanding, depth image segmentation, Otsu threshold, gradient feature, depth feature

谭志国1，2,欧建平1,张军1,沈先耿2. 多特征组合的深度图像分割算法[J]. 计算机工程与科学.

TAN Zhiguo1，2，OU Jianping1，ZHANG Jun1，SHEN Xiangeng2.

Multi-feature combined depth

image segmentation algorithm

[J]. Computer Engineering & Science.

[1]	陈巩, 李占利, 朱莉. 多尺度深度特征融合的个人信用风险预测[J]. 计算机工程与科学, 2023, 45(12): 2265-2273.
[2]	魏伟一, 赵毅凡, 陈帼. 基于深度特征提取和DCT变换的图像复制粘贴篡改检测[J]. 计算机工程与科学, 2023, 45(01): 163-170.
[3]	周燕, 潘丽丽, 陈蓉玉, 邵伟志, 雷前慧. 结合自适应融合网络与哈希的图像检索算法 [J]. 计算机工程与科学, 2021, 43(09): 1616-1622.