Efficient Watermarking for High Efficiency Video Coding

Abstract

According to the report of the Cisco, around 73% of the global traffic of data is in the form of High Definition (HD) and Ultra High Definition (UHD) quality videos. HEVC provides good visual output with 50% more compression accuracy contrasted with that of the previous H.264 / AVC encoding format. Consequently, nowadays, HEVC is hugely used for transmission of HD and UHD videos over the Internet. Therefore, video security concerns such as content protection and authentication of HEVC encoded videos have emerged an important area of research. Over the last few decades, digital watermarking has been used to deter copying, preserve integrity of content, track video transmission and dissemination. The present work is based on developing efficient watermarking schemes for authentication and protection of HEVC encoded videos by addressing the drawbacks of existing literature. The main features of the authentication scheme are that it should detect the malicious tampering in the contents of the video and must resist the non-malicious contents preserving manipulations that are done for transcoding the encoded videos. The main drawback of the existing authentication schemes is that they cannot resist the transcoding processing, i.e. re-compression with different Quantization Parameter (QP). To overcome this problem, in the present thesis, an authentication scheme is designed for HEVC, which can help shield the sanctity of the video during transcoding. In new scheme, invariant features, i.e. the relationship between the signs of negative and positive coefficients in blocks, is used to generate the authentication code. These features do not change after re-compression of the video with the same or different QP values. On the other hand, the generated authentication code is fragile to any malicious alterations such as insertion and deletion of objects in the contents of the video. The generated code is inserted into the video by altering the magnitudes of negative and positive coefficients. The integrity of the video is verified at the decoder end in two steps. The integrated authentication code is retrieved from the video during the first process. In the second step, the authentication code is re-generated from the contents of the video. Finally, the extracted and re-generated authentication codes are compared to detect any malicious tampering done in the videos. There are many schemes in existing literature for the protection of HEVC encoded videos. The compression efficiency of the HEVC is further improved by introducing new functionalities in the transformation process of the standard. For this purpose, a hybrid transformation is applied to the residual error to compact its maximum energy into fewer non-zero coefficients. In hybrid transformation, multiple transforms, namely Discrete Cosine Transform (DCT) and Discrete Sine Transform (DST) are applied on the single block of residual error. The existing single transformation based protection schemes are not effective for enhanced HEVC in contexts of resistance to attacks and barely noticeable modification throughout visual quality. Firstly, the existing schemes are not resilient to re-encoding with various QP values. This is due to the coefficients generated using different kernels of DCT and DST transform that have varying features with regards to stability and amount of energy contained in each coefficient. In the case of DCT, the first coefficient is the DC coefficient, which contains the maximum energy, and stability of the coefficients decreases from low to high frequency. On the other hand, the DST generated coefficients do not follow this phenomenon of stability and energy compaction. Moreover, some of the kernels used in hybrid transformation are not symmetric. Therefore any distortion induced by altering the coefficients of these transforms cannot be nullified and are spread to the neighbouring blocks. This mostly results in recognizable artifacts in the video after embedding the watermark. To overcome these problems, hybrid transforms oriented watermarking scheme is developed for the protection of enhanced HEVC video using Iframes. The developed I-frame protection scheme embeds the watermark by altering the residual error blocks. For this, first, the residual error is transformed using a symmetric transform, which is selected based on the hybrid transformation process used by the encoder. Then, the magnitudes of the transform coefficients generated have been adjusted whereas maintaining the same signs of coefficients, as any changes in the signs of the coefficients degrade the quality drastically. Currently, there is extensive use of Low Delay (LD) configuration for the compression of the video that are employed in the video applications specifically surveillance and conferencing system. The LD configuration is the primary video coding structure, of that every first frame is an I-frame while the subsequent frames being P-frame. So, a P-frame based protection scheme is proposed for video applications with low bit-rate. The above I-frame based protection scheme does not give optimal results for P-frames. In P-frames, dual-mode (intra and inter) is used in the prediction process. The blocks in P-frames are predicted from spatial and temporal neighbours. This results in the synchronization error, which is due to the changes in the prediction mode of the blocks after the re-compression, i.e. the prediction mode is changed from intra to inter and vice-versa. The blocks for embedding the watermark are selected using a robustness threshold to fix this issue of synchronization error. The robustness threshold is computed on the fly by exploiting the magnitude of the Quantized Transform Coefficients (QTC) in the P-frame’s intra-and interluma blocks. Further, the embedding of the watermark in P-frames results in a significant increase in bit-rate of video. For controlling this increase in bit-rate, the pair of coefficients with non-zero magnitudes are altered to embed the watermark.