Similar-Video Retrieval via Learned Exemplars and Time-Warped Alignment
- Cite this paper as:
- Horie T., Moriwaki M., Yokote R., Ninomiya S., Shikano A., Matsuyama Y. (2014) Similar-Video Retrieval via Learned Exemplars and Time-Warped Alignment. In: Loo C.K., Yap K.S., Wong K.W., Beng Jin A.T., Huang K. (eds) Neural Information Processing. ICONIP 2014. Lecture Notes in Computer Science, vol 8836. Springer, Cham
New learning algorithms and systems for retrieving similar videos are presented. Each query is a video itself. For each video, a set of exemplars is machine-learned by new algorithms. Two methods were tried. The first and main one is the time-bound affinity propagation. The second is the harmonic competition which approximates the first. In the similar-video retrieval, the number of exemplar frames is variable according to the length and contents of videos. Therefore, each exemplar possesses responsible frames. By considering this property, we give a novel similarity measure which contains the Levenshtein distance (L-distance) as its special case. This new measure, the M-distance, is applicable to both of global and local alignments for exemplars. Experimental results in view of precision-recall curves show creditable scores in the region of interest.
KeywordsSimilar-video retrieval exemplar time-bound affinity propagation M-distance numerical label
Unable to display preview. Download preview PDF.