Video Tampering Detection Algorithm Based on Spatial Constraints and Stable Feature

Most traditional video passive forensics methods only utilize the similarity between adjacent frames. They usually suffer from high false detection rate for the videos with severe motion. To overcome this issue, a novel coarse-to-fine video tampering detection method that combines spatial constraints with stable feature is proposed. In the coarse detection phase, both the low-motion region and the high-texture region are extracted by using spatial constraint criteria. The above two regions are merged to obtain the regions with rich quantitative correlation, which are then used for extracting video optimal similarity features. The luminance gradient component of the optical flow is computed and considered as relatively stable feature. Then, the suspected tampered points are found by combining the above two features. In the fine detection phase, the precise tampering points are located. The similarity of the gradient structure based on the characteristics of the human visual system is utilized to further reduce the false detections. This method is tested on three public video data sets. The experimental results show that compared with the existing works, this method not only has lower false detection rate and higher accuracy for the videos with severe motion, but also has high robustness to regular attacks, such as additive noise, blur and filtering.