A Compact Spatio-Temporal Fingerprint for Video Copy Detection System

Video copy detection system is used for video indexing and copy right applications. The fingerprint extraction algorithm extracts unique features of the video based on the content of the video. The key frames are extracted from the video based on the Structural Similarity Measure (SSIM). The fingerprint can be extracted from specially constructed image which is known as Temporally Informative representative Image (TIRI). The TIRI image contains both spatial and temporal information about the video. Then the TIRI-DCT algorithm extracts the fingerprints from the TIRIs. The search algorithm such as header based similarity search method searches close enough match for fingerprint of query video and fingerprint within database. The extracted fingerprint should be robust to frame rate conversion, background and intensity changes and geometric attacks such as rotation, scaling etc. The system focuses on determining pirated version of video from the video database. It consists of three stages. In key frame extraction the SSIM is used for measuring the similarity between the frames. In this way avoid more similar frames and preserve those are distinct or key frames. These key frames are used for generating TIRI images. The weighted average of all the selected frames is taken as TIRI. Then Discrete Cosine Transform (DCT) algorithm is applied to extract the features, because it can precisely capture local variations that exist in the TIRI image. From that the binary feature data or binary fingerprint is generate and stored in the fingerprint database. For fingerprint searching within the database a header based similarity search method is proposed. The system has been tested for around 50 videos. Among these test result gave a true positive rate of 97 percent. This result shows that the proposed copy detection system is more robust and discriminant.