TV Logo Recognition for Internet Videos Based on Selective Gabor Filter Bank

Recognizing TV logo in internet videos is a very challenge task,because the logo region is vulnerable to blurring and background changes dramatically especially for the hollow-out or semitransparent logos.Aimed at solving these problems,a Gabor filter bank-based approximate invariant feature is proposed to construct robust logo features against blurring and background changes.A new logo matching model is built on some certain fitted ellipsoids of the feature cluster’s surfaces,so that complex classification may be replaced by simple determining the position relationship between logo feature and ellipsoids,which can dramatically speed up logo matching and enable TV logo recognition with one single frame.Experiments show that the proposed method can efficiently recognize all kinds of logos and achieves 97.3% F1 measurement on average.