University of Paris 6 at TRECVID 2005: High-Level Feature Extraction

In this paper, we present the methodology we use in the NIST TRECVID'2005 evaluation. We have participated in the High-level Feature Extraction task. Our approach is founded on Fuzzy Decision Trees through the Salammbô software. 1 Structured Abstract Summary Here we present the contribution of the University of Paris 6 at TRECVID 2005 [1]. It concerns only the High-Level Feature Extraction task. The approach focuses on the use of Fuzzy Decision Trees (FDT) and is based on a rather simple image description. In the following, we start with a short summary of the used method and up from Section 3, our approach is detailed. First, we describe the particularities of our image descriptors. Then we explain how we performed the training (Section 4) and classi cation (Section 5). Before concluding, the submitted runs are discussed in details(Section 6). 1.1 Brief Description of the Runs Here is the general information about all the runs (more information can be found in the rest of this paper): The task: High-Level Feature Extraction. The feature: #40. Map: Segment contains video of a map. Type: A system trained on TRECVID development collection data, and common annotation of such data. Data used: XML les that describes the cutting into shot of each video (Master shot references by [5]), All of the image les representing keyframes, Annotations les for devel keyframes. Pre-treatment: Each keyframe was cut in 5 regions (see 3.1), HSV histogram was computed for each piece of a keyframe (see Section 3.1), Temporal information about each shot was extracted from the XML les. (see Section 3.2). Training: The Fuzzy Decision Tree (FDT) learning method was used in each run (see Section 4). Table 1 shortly di erentiates the submitted runs. The columns of the table describe: