Features and Automatic Search at The University of Iowa

(Automatic) Search Approach – fully automatic search exploring the impact of using frames sampled at 2 second intervals rather than keyframes: • UIowa07AS01 – required ASR using noun phrases and entities • UIowa07AS02 – required non-ASR run color histogram * edge * distance • UIowa07AS03 – color histogram similarity • UIowa07AS04 – product of maximal histogram, edge, distance similarities for shot samples (one every 2 seconds) • UIowa07AS05 – product of maximal histogram, edge, distance similarities for shot samples (one every 2 seconds) boosted by NPs and entities Our best runs were for simple color histogram similarity, although the sampled interval product was very close in performance. Our worst performing run was the text only, indicating the impact that ASR and MT can have on a nonnews corpus.

[1]  Erkki Oja,et al.  Statistical Shape Features for Content-Based Image Retrieval , 2004, Journal of Mathematical Imaging and Vision.

[2]  Paul Over,et al.  Evaluation campaigns and TRECVid , 2006, MIR '06.

[3]  Paul Scheunders,et al.  Wavelets for texture analysis, an overview , 1997 .

[4]  Markus A. Stricker,et al.  Similarity of color images , 1995, Electronic Imaging.