¡@

A feature-based sequence-to-sequence alignment method was proposed by Caspi et al., which can align multiple sequences in space-time. But this method can not be very effective and precise in spatial alignment, due to a small number of moving objects in the scene, limited coverage of the trajectories and the short temporal duration of the trajectories. In order to solve this problem, we proposed a new sequence-to-sequence alignment method, combining the method proposed by Caspi et al. with a feature-based image registration method. In this paper, we made a comparison between our method and that of Caspi et al. Experimental results show that the presented method improves alignment precision both on X-axis orientation and on Y-axis orientation. The PSNR value is approximately improved by 13.5dB. At the same time, the time needed by Our-method will be slightly increased but can be ignored. When the results of the alignment are applied in space-time super-resolution reconstruction, the effect also proves the validity of our-method.

Keywords: sequence-to-sequence alignment, image registration, super-resolution reconstruction, psnr, alignment accuracy, alignment time

Retrieve PDF document (201205_12.pdf)

Received June 17, 2010; revised August 12, 2010; accepted September 27, 2010.
Communicated by Chun-Lin Huang.
^* This paper was supported in part by the National Natural Science Foundation of China (No. 61071161) Science Research Program of Sichuan Provincial Department of Education of China (11ZB073), Science Research Program of Visual Computing and Virtual Reality Key Laboratory of Sichuan Province (PJ201113) and the Education Ministry¡¦s Key Research Project on Science and Technology of China (No. 107094).
⁺ Corresponding author.