作者: Hans-Peter Kriegel , Peter Kunath , Alexey Pryakhin , Matthias Schubert
DOI: 10.1007/978-3-540-77409-9_15
关键词: Representation (mathematics) 、 Computer science 、 Similarity (network science) 、 Feature (computer vision) 、 Pattern recognition 、 Artificial intelligence 、 Similarity heuristic 、 Multimedia 、 Semantic similarity 、 Probability distribution 、 Data mining 、 Precision and recall 、 Normalized compression distance
摘要: In modern multimedia databases, objects can be represented by a large variety of feature representations. order to employ all available information in best possible way, joint statement about object similarity must derived. this paper, we present novel technique for multi-represented estimation which is based on probability distributions modeling the connection between distance value and similarity. To tune these distribution functions model each representation, propose bootstrapping approach maximizing agreement distributions. Thus, capture general notion implicitly given relationships our does not need any training examples. experimental evaluation, demonstrate that new offers superior precision recall compared standard measures real world audio data set.