Signature-based search algorithm
S.A. Khakoo · International Conference on Acoustics, Speech, and Signal Processing · 2003
Most existing search algorithms for motion compensation have been specifically tailored to exploit the interframe redundancy in image sequences that contain small amounts of motion. With large amounts of motion, the performance of these algorithms is significantly reduced. A class of signature-based search algorithms is proposed as an alternative to these algorithms. A specific example, called the frequency labeling search algorithm, is developed in detail. Experimental results show that this algorithm gives compression rates 3-5 times that achieved by other algorithms in the presence of substantial amounts of motion. The accuracy of the technique can be adjusted to suit different applications by changing the number of coefficients used in the signature and the size of the candidate set. While it is computationally more expensive than other approaches, the cost of the frequency labeling technique falls within the range reasonable for real-time applications.>