Etter Solutions Research Group TRECVID 2007

Etter Solutions Research Group participated in the TRECVID conference for the first time in 2007. We submitted five runs in the area of fully automatic search. Fully Automatic Search Runs F_A_1_ESRG_A1_1 The first run is a required baseline, using only ASR/MT features. A sliding window of three shots is used to create a “bag of words” representation of the shot. F_A_1_ESRG_KN2_2 The second run is a required baseline, using no text from the ASR/MT output. This run uses visual keypoint features with shot description expansion using a semantic network. F_A_2_ESRG_AN3_3 The third run is an optional run combining ASR/MT features with shot term expansion using a semantic network. F_A_2_ESRG_AKN4_4 The fourth run is an optional run using a weighted fusion model with the ASR/MT and visual keypoint features. Shot description expansion using a semantic network is applied to both feature sets. F_A_2_ESRG_AKNF5_5 The fifth run is an optional run using a weighted fusion model with the ASR and ...

• Type: Conference
• Year: 2007
• Where: TRECVID
• Authors: David Etter
• Keywords: Semantic Network, Shot Description Expansion, TRECVID 2007, TRECVID 2008, Visual Keypoint Features