National Engineering Laboratory on Information Content Analysis Techniques

Shanghai, China

National Engineering Laboratory on Information Content Analysis Techniques

Shanghai, China
SEARCH FILTERS
Time filter
Source Type

Zhao Y.,Shanghai JiaoTong University | Sun T.,National Engineering Laboratory on Information Content Analysis Techniques | Jiang X.,National Engineering Laboratory on Information Content Analysis Techniques | Wang S.,National Engineering Laboratory on Information Content Analysis Techniques
Proceedings - 2016 9th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics, CISP-BMEI 2016 | Year: 2016

Human action/interaction recognition has wide appliances in video surveillance. Spatial and temporal information representation is the key issue of this topic. A framework called Long-term Residual Recurrent Network (LRRN) for human interaction recognition is proposed in this paper. The framework has an advantage of incorporating spatial and temporal features. Spatial feature is generated from Residual Network (ResNet). Temporal feature is learned from Long Short Term Memory (LSTM). The spatial-temporal feature representation learned automatically from LRRN is more expressive than hand-crafted counterparts. Optical flow image sequences are utilised to reduce static background interference. Experiments are conducted on BIT-interaction and UT-interaction datasets. The results show excellent performance in accuracy compared with prior traditional methods, achieving a state-of-art accuracy of 90% and 98.33% respectively. © 2016 IEEE.


Dong Y.,Shanghai JiaoTong University | Jiang X.,Shanghai JiaoTong University | Jiang X.,National Engineering Laboratory on Information Content Analysis Techniques | Sun T.,Shanghai JiaoTong University | And 2 more authors.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2017

Steganographic Channel Model (SCM) is hard to build for different steganography algorithms in different embedding domains. Thus, theoretical analysis for some important factors in steganography, such as capacity, distortion, is hard to obtain. In this paper, to avoid introducing significant distortion into HEVC video file, a novel HEVC SCM is presented and analyzed. It is firstly proposed that the distortion optimization method in this SCM should be applied on coding efficiency instead of visual quality. According to this conclusion, a novel coding efficiency preserving steganography algorithm based on Prediction Units (PUs) is proposed for HEVC videos. The intra prediction modes of candidate PUs are taken as cover. This algorithm was tested on the dataset consisting of 17,136 HD sequences. The Experimental results prove the correctness of the previous conclusion and the practicability of the proposed channel model, and show that our algorithm outperforms the existing HEVC steganography algorithm in capacity and perceptibility. © Springer International Publishing AG 2017.


Zheng J.,Shanghai JiaoTong University | Sun T.,Shanghai JiaoTong University | Sun T.,National Engineering Laboratory on Information Content Analysis Techniques | Jiang X.,Shanghai JiaoTong University | And 2 more authors.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2017

Detection of double video compression plays an important role in video forensics. However, existing methods rarely focused on H.264 videos and are unreliable to provide detection results for static-background videos with fast moving foregrounds. In this paper, an effective double compression detection scheme based on Prediction Residual of Background Regions (PRBR) is proposed to overcome these limitations. Firstly, the mask of background regions in each frame is obtained by applying Visual Background Extractor (VIBE). VIBE is an efficient and robust background subtraction algorithm, which can distinguish the background and foreground of each frame at pixel level. Then, the PRBR feature is designed to characterize the statistical distribution of average prediction residual within the background mask. After that, the Jesen-Shannon Divergence is introduced to measure the difference between the PRBR features of the adjacent two frames. Finally, a periodic analysis method is applied to the final feature sequence for double H.264 compression detection and estimation of the first Group Of Pictures (GOP). Eighteen standard testing sequences captured by fixed cameras are used to establish the double compression dataset. Experiments demonstrate the proposed scheme can achieve better performance compared the-state-of-art methods. © Springer International Publishing AG 2017.


Shen M.,Shanghai JiaoTong University | Shen M.,National Engineering Laboratory on Information Content Analysis Techniques | Jiang X.,Shanghai JiaoTong University | Jiang X.,National Engineering Laboratory on Information Content Analysis Techniques | And 2 more authors.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2017

Since there is an increasing demand of security and safety assurance for public, anomaly detection has been a great focus in the field of intelligent video surveillance analysis. In this paper, a novel method is proposed for anomaly detection through the analysis of the pedestrian behavior with motion-appearance features and the dynamic changes of the behavior over time. Locality Sensitive Hashing (LSH) functions are used in the method to finally detect the abnormal behaviors. Compared to the relative works, the main novelties of this paper mainly includes: (1) the pedestrians in the image are segmented with the method of Robust Principal Component Analysis (RPCA) in the preprocessing step; (2) in order to describe the dynamic changes of behavior, the Dynamics of Pedestrian Behavior (DoPB) feature on Riemannian manifolds is proposed as the individual descriptor; (3) during the detection process, the Adaptive Anomaly Weight (AAW) with block-based optical flow tracking is used to measure the anomaly saliency. Experimental results and the comparisons with state-of-the-art methods demonstrate that the proposed method is effective in anomaly detection and localization. © Springer International Publishing AG 2017.


Xu Q.,Shanghai JiaoTong University | Sun T.,Shanghai JiaoTong University | Sun T.,National Engineering Laboratory on Information Content Analysis Techniques | Jiang X.,Shanghai JiaoTong University | And 2 more authors.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2017

During the process of video forgery detection, double compression is a significant evidence. A novel scheme based on the Sequence of Number of Prediction Unit of its Prediction Mode (SN-PUPM) is proposed to conduct double compression detection on videos under HEVC standard, together with estimation on GOP structures. Number of PU with three kinds of prediction mode (INTRA, INTER and SKIP) is firstly extracted from each frame inside a given video sequence. Then the SN-PUPM is calculated by Absolute Difference Values from adjacent three frames in original extracted features and filtered with Twice Averaging Filter to reduce noises induced by the process. Then, an initiative Abnormal Value Classifier is trained with SVM to label I-P frames and have a final sequence for double compression detection and GOP analysis. Nineteen original YUV sequences are adopted for dataset in experiments. Results have demonstrated better performance in HEVC double compression than previous method adapted to HEVC. © Springer International Publishing AG 2017.


Chao J.,Shanghai JiaoTong University | Jiang X.,Shanghai JiaoTong University | Jiang X.,National Engineering Laboratory on Information Content Analysis Techniques | Sun T.,Shanghai JiaoTong University | And 2 more authors.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2013

In this paper, a novel video inter-frame forgery detection scheme based on optical flow consistency is proposed. It is based on the finding that inter-frame forgery will disturb the optical flow consistency. This paper noticed the subtle difference between frame insertion and deletion, and proposed different detection schemes for them. A window based rough detection method and binary searching scheme are proposed to detect frame insertion forgery. Frame-to-frame optical flows and double adaptive thresholds are applied to detect frame deletion forgery. This paper not only detects video forgery, but also identifies the forgery model. Experiments show that our scheme achieves a good performance in identifying frame insertion and deletion model. © 2013 Springer-Verlag.


Cheng D.,Shanghai JiaoTong University | Sun T.,Shanghai JiaoTong University | Sun T.,National Engineering Laboratory on Information Content Analysis Techniques | Sun T.,New Jersey Institute of Technology | And 2 more authors.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2013

In recent researches, image classification of objects and scenes has attracted much attention, but the accuracy of some schemes may drop when dealing with complicated datasets. In this paper, we propose an image classification scheme based on image sparse representation and multiple kernel learning (MKL) for the sake of better classification performance. As the fundamental part of our scheme, sparse coding method is adopted to generate precise representation of images. Besides, feature fusion is utilized and a new MKL method is proposed to fit the multi-feature case. Experiments demonstrate that our scheme remarkably improves the classification accuracy, leading to state-of-art performance on several benchmarks, including some rather complicated datasets such as Caltech-101 and Caltech-256. © 2013 Springer-Verlag.


Xu K.,Shanghai JiaoTong University | Jiang X.,Shanghai JiaoTong University | Jiang X.,National Engineering Laboratory on Information Content Analysis Techniques | Sun T.,Shanghai JiaoTong University | Sun T.,National Engineering Laboratory on Information Content Analysis Techniques
Proceedings - International Conference on Image Processing, ICIP | Year: 2015

A novel method for human action recognition is proposed in this paper. Traditional spatial-temporal interest point detectors are easily affected by hair, face, shadow, clothes texture or the shake of camera. Inspired by the use of points distribution information, we propose a point selection method to select representative points (denoted by the 'pose points'), which use HOG human detector and contour detector to select the points on human pose edges. The pose points carry both local gradient information and global pose information. 3D-SIFT scale selection method and novel descriptors called body scale and motion intensity feature are also studied. The descriptors calculate the width scale of different levels of human body and count motion intensity of activity in five directions. The descriptors combine spatial location with the moving intensity together and are used for further classification with SVMs. Experiments have been conducted on benchmark datasets and show better performance than previous methods, which achieved 99.1% on Weizmann dataset and 95.8% on KTH dataset. © 2015 IEEE.


He P.,National Engineering Laboratory on Information Content Analysis Techniques | Jiang X.,Shanghai JiaoTong University | Sun T.,National Engineering Laboratory on Information Content Analysis Techniques | Wang S.,National Engineering Laboratory on Information Content Analysis Techniques
Journal of Visual Communication and Image Representation | Year: 2016

Videos captured by stationary cameras are widely used in video surveillance and video conference. This kind of video often has static or gradually changed background. By analyzing the properties of static-background videos, this work presents a novel approach to detect double MPEG-4 compression based on local motion vector field analysis in static-background videos. For a given suspicious video, the local motion vector field is used to segment background regions in each frame. According to the segmentation of backgrounds and the motion strength of foregrounds, the modified prediction residual sequence is calculated, which retains robust fingerprints of double compression. After post-processing, the detection and GOP estimation results are obtained by applying the temporal periodic analysis method to the final feature sequence. Experimental results have demonstrated better robustness and efficiency of the proposed method in comparison to several state-of-the-art methods. Besides, the proposed method is more robust to various rate control modes. © 2015 Elsevier Inc. All rights reserved.


He P.,Shanghai JiaoTong University | He P.,National Engineering Laboratory on Information Content Analysis Techniques | Sun T.,Shanghai JiaoTong University | Sun T.,National Engineering Laboratory on Information Content Analysis Techniques | And 4 more authors.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) | Year: 2015

In this paper, we propose a novel scheme to detect double MPEG-4 compression with block artifact analysis. An adaptive measurement of block artifact in decompressed frames is proposed and then combined with the Variation of Prediction Footprint (VPF) in an effective way. Based on such measurement, periodic analysis is used to detect double compression. The proposed scheme is verified on several publically available standard videos and compared with the state-of-the-art method. Experimental results demonstrate that it has more robust detection capability. © Springer International Publishing Switzerland 2015.

Loading National Engineering Laboratory on Information Content Analysis Techniques collaborators
Loading National Engineering Laboratory on Information Content Analysis Techniques collaborators