In this chapter, we present a novel fast and robust watermarking scheme for three different standard video in RGB uncompressed AVI format in DWT domain using a newly developed SLFN commonly known as Extreme Learning Machine (ELM). The embedding is carried out by using scene detection. The LL4 sub-band coefficients of frames constitute the dataset to train the ELM in millisecond time. The output of the ELM is used to embed a binary watermark in the video frames using a pre-specified formula. The resultant video exhibits good visual quality. Five different video processing attacks are executed over signed video. The extracted watermarks from the signed and attacked video yield high normalized correlation (NC) values and low Bit Error Rate (BER) values. This indicates successful watermark recovery and the embedding scheme is found to be robust against these common attacks. It is concluded that the proposed watermarking scheme produces best results due to optimized embedding facilitated by fast training of the ELM. The proposed scheme is found to be suitable for developing real time video watermarking applications due to its low time complexity.
|Title of host publication||Extreme Learning Machines 2013: Algorithms and Applications|
|Editors||Fuchen Sun, Kar-Ann Toh, Manuel Grana Romay, Kezhi Mao|
|Place of Publication||Switzerland|
|Number of pages||17|
|Publication status||Published - 2014|
|Name||Adaptation, Learning, and Optimization|