A Video Saliency Detection Model in Compressed Domain
2014-01-03 20:13
302 查看
A Video Saliency Detection Model in Compressed Domain
Last Edit 2014/1/3
文献信息:
Y. Fang, W. Lin, Z.Chen, C. Tsai, and C. Lin, “A Video Saliency Detection Model in CompressedDomain,”
Circuits and Systems for Video Technology, IEEE Transactions on,vol. PP, no. 99, pp. 1-1, 2013
文献的内容:
主要的内容利用压缩信息(DCT系数,Motion Vector)分别提取出空域特征,时域特征,计算出spatioal saliency,temporal saliency,最终将两都融合(提出了新的融合方法)见下图。
计算过程是在8*8的blocklevel下进行的,MPEG4 ASP。
第一部分:A. Feature Extraction from Video Bitstream (特征提取)
1) DCT Coefficient and Motion Vector Extraction (DCT系数提取)
DCT系数主要是从I帧提取,64个系数。
运动向量是从P,B帧内提取。
2) Feature Calculation Based on DCT Coefficients (特征计算)
在一个8*8的块内包含1个直流系数(DC coefficient) 63个交流系数(AC coefficient)
AC系数主要是用来计算纹理特性的,而且主要是用前9个AC系数(因为前9个系数能够表示一个block的绝大部分能量)
纹理特征T:
3) Motion Feature Calculation Based on Motion Vectors(运动特征)
下面是一个Motion Feature提取的示意图:
第二部分B. Saliency Detection in Compressed Domain
1)Static and Motion Saliency Map Calculation
这个部分还是利用center-surround,并利用高斯模型来模拟这个机制。
分别求出各个特征的值后,下一步要做的就是把各个值融合在一起,最终求出图像的显著性;
怎么个融合呢?
2) Final Saliency Map Calculation
静态显著性值与动态显著性值的融合:
这个系数的计算就不去管了~~~~
跟上一篇文献对比,这篇文献的计算过程还是将spatial saliency 和temporal saliency 这两个过程分开来。
Last Edit 2014/1/3
文献信息:
Y. Fang, W. Lin, Z.Chen, C. Tsai, and C. Lin, “A Video Saliency Detection Model in CompressedDomain,”
Circuits and Systems for Video Technology, IEEE Transactions on,vol. PP, no. 99, pp. 1-1, 2013
文献的内容:
主要的内容利用压缩信息(DCT系数,Motion Vector)分别提取出空域特征,时域特征,计算出spatioal saliency,temporal saliency,最终将两都融合(提出了新的融合方法)见下图。
计算过程是在8*8的blocklevel下进行的,MPEG4 ASP。
第一部分:A. Feature Extraction from Video Bitstream (特征提取)
1) DCT Coefficient and Motion Vector Extraction (DCT系数提取)
DCT系数主要是从I帧提取,64个系数。
运动向量是从P,B帧内提取。
2) Feature Calculation Based on DCT Coefficients (特征计算)
在一个8*8的块内包含1个直流系数(DC coefficient) 63个交流系数(AC coefficient)
AC系数主要是用来计算纹理特性的,而且主要是用前9个AC系数(因为前9个系数能够表示一个block的绝大部分能量)
纹理特征T:
3) Motion Feature Calculation Based on Motion Vectors(运动特征)
下面是一个Motion Feature提取的示意图:
第二部分B. Saliency Detection in Compressed Domain
1)Static and Motion Saliency Map Calculation
这个部分还是利用center-surround,并利用高斯模型来模拟这个机制。
分别求出各个特征的值后,下一步要做的就是把各个值融合在一起,最终求出图像的显著性;
怎么个融合呢?
2) Final Saliency Map Calculation
静态显著性值与动态显著性值的融合:
这个系数的计算就不去管了~~~~
跟上一篇文献对比,这篇文献的计算过程还是将spatial saliency 和temporal saliency 这两个过程分开来。
相关文章推荐
- Moving Region Detection in Compressed Video运动检测论文翻译
- Salient Motion Detection in Compressed Domain
- Splicing MPEG Video Streams in the Compressed Domain(翻译)
- detection in video and image
- 【计算机视觉】<前景检测>ViBe - a powerful technique for background detection and subtraction in video sequences
- saliency detection by forward and backward cues in deep-cnn
- Automatic fall detection of human in video using combination of features译文
- Factors in Finetuning Deep Model for Object Detection with Long-tail
- how to use trained object detection model in tensorflow
- ViBe算法:ViBe - a powerful technique for background detection and subtraction in video sequences
- Factors in Finetuning Deep Model for Object Detection with Long-tail Distribution解读
- Violence Detection in Video Using Computer Vision Techniques
- 转-Region-based Saliency Detection and Its Application in Object Recognition
- 显著性检测论文解析2——Visual Saliency Detection Based on Bayesian Model, Yulin Xie, ICIP2011
- VISUAL SALIENCY DETECTION BASED ON BAYESIAN MODEL
- 【分享】视频视觉显著度数据集和评测方法(A dataset and evaluation methodology for visual saliency in video)
- ###《Video Event Detection by Inferring Temporal Instance Lables》
- ### 《Video Event Detection by Inferring Temporal》
- Traditional Saliency Reloaded: A Good Old Model in New Shape
- 【论文】Efficient compressed sensing based object detection system for video surveillance application