CASIA OpenIR  > 毕业生  > 硕士学位论文
Alternative TitleResearch in automatic video shot detection
Thesis Advisor卢汉清
Degree Grantor中国科学院自动化研究所
Place of Conferral中国科学院自动化研究所
Degree Discipline模式识别与智能系统
Keyword镜头检测 比值直方图 光照不变测度 速度和精度 Shot Detection Ratio Histogram Illumination Invariant Measures Speed And Accuracy
Abstract镜头切换的自动检测一直是数字视频处理中一个重要的课题。无论是Marc Davis 的Media Stream或者是CMU的Video Skimmer都要涉及。 尽管有许多人在该领域研究过,仍然有要解决的问题。 其一是渐变检测问题。双门限方法是一种常用的方法,但阈值难于选择而且阈值 严重影响了算法的性能。渐变检测的主要思路是通过累积比较的办法。我们的算 法思想是先检测渐变的结束处,在以结束处为参考帧反向累积比较来寻找渐变开 始的地方。只用一个阈值就可检测渐变。然而,检测到的也不可避免地带有摄像 机目标大幅运动。实际上,这也是很多算法的不足之处。 其二是光照变化问题。Kong把Funt的颜色比值直方图引入镜头检测作为一种 光照不变测度。我们认为没有必要进行如此精细地比较。因此我们提出了另一种 光照不变测度。与颜色比值直方图相比,它计算简单但是仍然有效地解决光照变 化问题。首先通过差分计算出图像的边缘部分,二值化处理后,把背景面积/边 缘面积作为光照不变测度(ABEI)。 其三讨论了把比值直方图和一般直方图结合起来进行镜头检测的方法。该方法在 速度和精度间寻求了一种平衡:在提高速度的同时而没有牺牲太多的精度。该算 法的特点是首先使用一股直方图来寻找可能的切换处,然后再利用比值直方图排 除那些假的切换。该算法的有效性得到了实验的证实,同时也分析了算法的失败 之处。 其四总结了我们在该领域的研究并提出了对未来工作的设想。设想部分包括了小 波一个来使各种方法优势互补,扬长避短的框架。该框架的建立也是本着速度和 精度平衡的原则。设法使算法速度加快而不牺牲精度,同时该框架还是开放的, 它可以融合新的技术。
Other AbstractAutomatic shot shift detection is always a hot topic in the digital video processing. Both Marc Davis's Media-stream and CMU's video skimmer need that technology. Though it had been studied in the past by many researchers, there are still many issues to address. 1. It is difficult to select the thresholds in the twin-comparison method. The main idea in gradual transition detection is to use accumulated comparison. Our method detects the end of the transition frst and then in the reversal direction, the start of the transition is located with the accumulated comparison with the end frame of the transition. Only one threshold is needed to detect gradual transition. Unfortunately, some large camera and objects motions may also be included. Actually this is the case for many algorithms. 2. Illumination variation. Kong introduces Funt's color ratio histograrn into video segmentation as an illumination invariant measures. We do not think it is necessary to make such accurate calculations. Therefore we proposed another illumination invariant measure. Compared with color ratio histogram, it is simpler in computing while remaining the same performance in dealing with illumination variation. First the edges of the image are extracted by differencing. We take it as the illumination measure the area of the background/objects in the binary images generated after thresholding the edges. The new illumination invariant measures are called ABEl. 3. How to combine the Ratio Histogram with General Histogram in shot detections? Our method seeks a balance between speed and accuracy: the speed is improved however accuracy is guaranteed. The algorithm first locates possible shot boundaries with general histogram as image content similarity measurement. Then false alarms are discarded with ratio histogram as image content sirnilarity measurement. The validation of the algorithm has been confirmed by some test sequences. Meanwhile the failures of the algorithm are also addressed. 4. Summary of our study in this field and our suggestions for further work. One of the suggestions is a frame to combine the advantages of the algorithms while trying to avoid the disadvantages. The frame is also based on the principal of the balance between speed and accuracy. Try to speed the algorithm while maintaining the accuracy. At the same time, the frame is open in the sense that any new technology can be fused into it.
Other Identifier598
Document Type学位论文
Recommended Citation
GB/T 7714
李大龙. 视频镜头自动检测的研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所,2001.
Files in This Item:
There are no files associated with this item.
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[李大龙]'s Articles
Baidu academic
Similar articles in Baidu academic
[李大龙]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[李大龙]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.