1 Abstract This year Eurecom submitted 5 runs for the High-Level Feature Extraction task. Below are brief descriptions of these submissions. • A Eurecom rerank1: this run reranks the results of the previous run Eurecom fuse base based on the prior probability of the concept in shots and videos. • A Eurecom rerank2: this run updates the concept score of the run Eurecom fuse base based on context...

In this paper, we summarize our results for the high-level feature extraction task at TRECVID 2008. Our last year’s high-level feature extraction system was based on low-level features as well as on state-ofthe-art approaches for camera motion estimation, text detection, face detection and audio segmentation. This system served as a basis for our experiments this year and was extended in severa...

The problem of unsupervised audio classification continuous to be a challenging research problem which significantly impacts ASR and Spoken Document Retrieval (SDR) performance. This paper addresses novel advances in audio classification for speech recognition. A new algorithm is proposed for audio classification, which is based on Weighted GMM Network (WGN). Two new high-level features: VSF (V...

1. Briefly, what approach or combination of approaches did you test in each of your submitted runs? A_KL1_1: A color-based image retrieval method using three kinds of image features: a global color distribution feature, a common bitmap feature and a Wavelet texture feature. Key-frames generated by our frame clustering method with threshold 5 were used as the input of the feature extraction syst...

This paper presents a system that extracts 109 musical features from symbolic recordings (MIDI, in this case) and uses them to classify the recordings by genre. The features used here are based on instrumentation, texture, rhythm, dynamics, pitch statistics, melody and chords. The classification is performed hierarchically using different sets of features at different levels of the hierarchy. W...

We present our approach to TRECVID 2006, high-level feature extraction task. We submitted one run with type ‘A’, annotating all required 39 features. The approach was based on textual information extracted from speech recogniser and machine translation outputs. They were aligned with shots and associated with highlevel feature references. A list of significant words was created for each feature...

Availability of large amounts of raw unlabeled data has sparked the recent surge in semi-supervised learning research. In most works, however, it is assumed that labeled and unlabeled data come from the same distribution. This restriction is removed in the self-taught learning algorithm where unlabeled data can be different, but nevertheless have similar structure. First, a representation is le...

We participated in one task of TRECVID 2008, that is, the high-level feature extraction (HLFE). This paper presents our approaches and results on the HLFE task. We mainly focus on exploring the data imbalance learning in this year, and propose two methods for this problem: (1) adaptive borderline-SMOTE and under-sampling SVM (ABUSVM), and (2) concept category. Our approach can be divided into t...

This paper describes experiments carried out by the UC3M team for TRECVID 2008 high-level feature extraction task. Being our first participation in TRECVID, our goal this year has been to develop a modular system to facilitate future developments and incorporation of new functionality (feature extraction and classification modules). We have basically carried out experiments with two different k...