Research On Algorithm Of Video Tampering Detection

The intelligent video surveillance system is widely used in many fields and its correlativetechnology has become an attractive hotspot to researchers all over the world in recent years.With the requirements of intelligent video surveillance system growing, tamperings insurveillance system are happening more and more. These tamperings make image of systemchanged acutely, which has affected subsequent flows of system such as object detection, objectclassification, object tracking and so on. Sometimes the tampering can make the system losefunction, which may result in enormous economic losses and give an alarm for safety of publicand society.This dissertation makes research on tamperings of intelligent video surveillance system indetail. The contributions obtained can be summarized in the following aspects:1. Types of tampering are analyzed and summarized. The primary tamperings of systemcontain seven types such as camera turned, dithering of camera, covered camera, defocused,luminance error (over exposure and under exposure), color error and noise. According to cameraturned or not, they are divided into tampering of moving and tampering of no moving. A flowchart of tampering detection and classification is proposed. Meanwhile, image is analyzed in twoaspects of space domain and frequency domain, which contains features of color, edge, texture,Fourier transformation, wavelet transformation and so on. All of these supply a base fordetection and classification of tampering.2. A method of tampering detection which uses structure of two buffers is improved.Because of structure of two buffers, the method of Evan decreases rate of false alarm.Meanwhile time of detection is increased and real time operation is lost. Based on analyzingdisadvantage of Evan’s method, histogram of grads and flow of detection are improved. With thesame condition, time of detecting the same frame is compared between Evan’s method and theimproved method. The experimental results show that the improved method not only acceleratesthe speed of detection but also has excellent performance, which improves range of application.With the capacity of buffers growing, time of detection is reduced more and more.3. A method based on global motion estimation is proposed to detect and classify tamperingof moving. When tampering of moving happens which includes camera turned and dithering ofcamera, there is a displacement between two neighboring frames. The displacement is got bymethod of phase correlation based on Fourier transformation. In order to reduce computationalcomplexity, method of phase correlation is applied to four regions of every frame. Meanwhile,gray-histogram is used to compute difference between two neighboring frames. There isn’ttampering of moving when the difference is bigger than the number set before. Based ondetection, two thresholds are computed. One is the sum of absolute value of several continuousdisplacements, the other is absolute value of sum of several continuous displacements.Classification is realized due to difference between two thresholds. The experimental resultsshow that the proposed method has excellent performance of detection and classification, whichbasically meeting the requirement of the surveillance system. 4. A method based on image content is proposed to detect and classify types of tampering ofno moving. The tampering of no moving contains covered camera, defocused, luminance error(over exposure and under exposure), color error, noise and so on. Five features are proposedaccording to content of tampering images. Based on temporal difference, a way of combiningfive features is used to detect tampering of no moving and threshold of detection is updatedadaptively. Five rates between value of features and value of corresponding thresholds arecomputed. Classification is realized due to the maximal rate. The experimental results show thatthe proposed method has excellent performance of detection and classification, which suppliesconvenience for subsequent flows of intelligent video surveillance system.Finally, the research work for this thesis is summarized and the further research topics anddirections in the future of tampering of detection and classification in intelligent videosurveillance system are discussed.