A heuristic based on the intrinsic dimensionality for reducing the number of cyclic DTW comparisons in shape classification and retrieval using AESA

摘要: Cyclic Dynamic Time Warping (CDTW) is a good dissimilarity of shape descriptors high dimensionality based on contours, but it computationally expensive. For this reason, to perform recognition tasks, method reduce the number comparisons and avoid an exhaustive search convenient. The Approximate Eliminate Search Algorithm (AESA) relevant indexing because its drastic reduction comparisons, however, algorithm requires metric distance that not case CDTW. In paper, we introduce heuristic intrinsic allows use CDTW AESA together in classification retrieval tasks over these descriptors. Experimental results show that, for dimensionality, our proposal optimal practice significantly outperforms search, which only alternative them tasks.