立体视觉注意力引导的多视图图像 disparity 控制方法

27 浏览量更新于2024-08-26 收藏 1.17MB PDF 举报

本文是一篇发表在《显示技术杂志》(JOURNAL OF DISPLAY TECHNOLOGY)第10卷第5期的2014年5月的研究论文，标题为"立体视觉注意力引导的多视图图像深度控制"。该研究由 Jianjun Lei (IEEE会员)、Shiqiu Li、Bingren Wang 和 Kefeng Fan 以及 Chunping Hou 合著，主要探讨了在三维立体显示中深度控制（disparity control）的重要作用，目标是创建出高质量的3D图像和视频，以提供逼真的视觉体验。首先，作者深入解析了深度控制的基本原理，这涉及如何通过精确调整左右眼视差（disparity）来创造出立体感，使观众能够感知深度和空间。传统的深度控制通常依赖于固定的算法或规则，而本文则提出了一个创新的方法，即利用立体视觉注意力模型（stereoscopic visual attention model）来指导深度控制。这个模型旨在捕捉观众对画面中不同区域的注意力，通过对关键点的识别，优先处理那些引起观众兴趣和关注的部分。这种方法更注重用户体验，使得立体显示中的深度变化更加符合用户的主观感受和视觉期待。例如，如果观众正在注视画面中的某个物体，该模型会自动调整该物体周围的深度，使其在3D场景中更加突出。随后，作者将这种立体视觉注意力模型与深度控制原则相结合，设计了一种针对多视图图像的深度控制方法。这种方法能够实时根据观众的注意力动态调整各个视点之间的深度差异，从而在无须外部辅助设备的情况下实现自动立体（autostereoscopic）显示，提高观看者的沉浸式体验。实验结果展示了这一方法的有效性，通过处理多视图序列，提出的深度控制策略能够准确地反映观众的视觉焦点，使得突出区域在立体显示中得到清晰和生动的表现。因此，本文不仅提升了3D显示技术的质量，还为未来的虚拟现实（VR）和增强现实（AR）应用提供了有价值的新思路。这篇文章在3D显示技术领域具有重要的理论价值和实践意义，它探索了如何通过理解用户注意力来优化深度控制，从而增强立体视觉效果，为用户提供更加自然和身临其境的观影视觉体验。

JOURNAL OF DISPLAY TECHNOLOGY, VOL. 10, NO. 5, MAY 2014 373

Stereoscopic Visual Attention Guided Disparity

Control for Multiview Images

Jianjun Lei, Member, IEEE, Shiqiu Li, Bingren Wang, Kefeng Fan, and Chunpin

gHou

Abstract—Disparity control plays an important role in 3D

stereoscopic displays. It aims to produce high-quality 3D images

and videos which can offer realistic visual experien

ces. We propose

a disparity control method guided by stereoscopic visual attention.

Firstly, the fundamental prin ciple of disparity control is derived.

Then, a stereoscopic visual attention mo

del is introduced to get

the key-point of disparity control. Finally, the disparity control

for multiview images is conducted based on the stereoscopic

visual attention model and the princ

iple of disparity control. The

experimental results on multiview sequences demonstrate that

the proposed method is effective. The salient regions detected by

the proposed stereoscopic vis

ual attention model are displayed as

the viewer’s expectation, and the vivid 3D scenes are presented in

autostereoscopic displays.

Index Terms—Disparity control, stereoscopic visual attention,

multiview, autostereoscopic display.

I. INTRODU CTION

HREE-DIMENSIONAL (3D) display technologies,

which provide realistic and vivid visual experiences, are

receiving much attention in a number of areas, including 3D

cinema, 3DTV, and mobile phones [1], [2]. The classic 3D

movies and 3D T V systems based on conventio nal stereoscopic

vision provide a 3D experien ce throu gh special-purp ose glasses

without a motio n parallax. The m ultiview 3D techniques,

which do not require the use of special glasses, have received

widespread attention recently [3]. Among these techniques,

integral imaging and multiview autostereoscopic display are

widely researched. I n an integral imaging system, 2D elemental

images of a 3D scene are generated by a microlens array, and

each elemental image presents a different perspective of the 3D

scene. Then the recorded 2D elemental images are displayed by

an optical device placed in front of another microlens array to

reconstruct the 3D image. Thus, there is a real reconstruction o f

the light structure pro duced by the original 3D scene [4], [5]. In

multiview autostereoscopic display, multiview images obtained

Manuscript received January 28, 2014; revised February 21, 2014; accepted

March 16, 2014. Date of publication March 20, 2014; date of current version

April 14, 2014. This w ork was supported in part by the Natural Science Founda-

tion of China (No. 61271324, 60932007, 61101224, 61202266), and by Natural

Science Foundation of Tianjin (No. 12JCYBJC10400).

J. Le i is with the School of Electronic Information Engineering, Tianjin

University, Tianjin 300072, China, and also with the Department of E lectr ical

Engineering, University of Washington, Seattle, WA 98195, USA (e-mail:

jjleitg@gmail.com).

S. Li, B. Wang, an d C. Hou are with the Scho ol of Electr on ic Information

Engineering, Tianjin U niv er sity, Tianjin 300072 China.

K. Fan is w ith China E lectr on ics Stan dardization Institute, Beijing 100007,

China.

Color versions of one or more of the ﬁg ur es are available o n line at http://

ieeexplore.ieee.org.

Digital Object Identiﬁer 10.1109/JDT.2014.2312648

by an array of cam eras are displayed on parallax barrier or

lenticular autostereoscopic display. It can support the rendering

of a 3D scene from various viewpoints, and offer better access

to 3D co nten t [6]–[8].

The quality o f multiview images plays an important role in

multiview autostereoscopic display. Usually, multiview images

for autostereoscopic display can be divided into two typ es ac-

cording to the layout of camera array, namely, toed-in m ul-

tiview images and parallel multiview im ages. Original multi-

view images obtained from cameras usually have stereoscopic

distortions, wh ich may cause visual fatigue. Vergence and dis-

parity control are vitally i mportant to ensure the clear and nat-

ural display of multiview images. However, multiview images

captured by toed-in camera array invariably suffer from vertical

disparity and keystone distortion, which cannot b e el iminated

by vergence and dispa rity control [9]. Original multiview im-

ages captured by parallel camera array only have negative hor-

izontal disparity. This differs from actually viewing the scene,

thus causing all the contents to be viewed as though they are

in front of the screen. Unlike vertical disparity and keystone

distortion, the defect caused by parallel camera array can be

compensated through vergence and disparity control. Kwon et

al. proposed a binocular stereoscopic camera vergence contro l

method using disparity information by the simple image pro-

cessing [10]. However, the multiview camera array was not

mentioned. Deng et al. developed a tri-view vergence control

method using parallel cam era array with shift [11]. Although

the method gets parallel images with both positive and negative

disparities, it only suits for ﬁxed depth scenes due to its inﬂex-

ibility of shif ti ng manuall y or setting the shifting range in the

camera. In [12], we used mean shift algorithm to obtain the ob-

ject in the central region, and set it as the zero disparity object.

The d ispar ity control was then realized by shifting stereoscopic

images accordingly.

However, none of the above methods took stereoscopic visu al

attention into consideration. It has been highligh ted in previous

researches that visual attention, usu a lly modeled as saliency de-

tection, is very important when addressing 3D human visual

perception [13]. Stereoscopic visual attention model utilizin g

depth information emphasizes depth perception and simulates

human visual system more accurately [14]–[16]. In this paper,

we propose a disparity control meth od gu ided by stereoscop ic

visual attention. Firstly, we derive the fundamental p ri ncip le of

disparity con tro l. The salient regions of multiview images are

then extracted based on stereoscopic visual attention. Finally,

the center of the salient regions is set as the k ey-point of dis-

parity control, and disparity control i s achieved by shiftin g the

multiview images.

See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

下载后可阅读完整内容，剩余6页未读，立即下载

weixin_38528888

粉丝: 3
资源: 915

立体视觉注意力引导的多视图图像 disparity 控制方法

jQuery全屏视差滚动效果

结合空间域与变换域特征提取的盲立体图像质量评价.docx

安卓引导页源码

双目立体视觉视差原理

用Python实现双目立体视觉的视差估计

matlab利用提供的两幅立体图像进行视觉立体匹配操作,要求实现获得该图像的视差图

如何用Python代码实现双目立体视觉的视差估计

如何利用Python和OpenCV实现双目立体视觉中的图像匹配与视差计算？请介绍SIFT、SURF、BF和FLANN算法在该过程中的应用。

matlab绘制单镜头立体视觉

如何通过双目立体视觉原理结合散斑匹配技术提高视差计算的准确性和效率？

最新资源