Automatic video-text extraction is an important field in video content comprehension.In complex background

conventional extraction methods can not robustly locate text nor eliminate interference from background.A robust video-text extraction method is proposed in this letter.First

an unsupervised paradigm based on wavelet is applied to obtain candidate text region.Second

traversing line and its aptitude spectrum is introduced

and text refinement is carried out by extracting traversing line features.At the end

according to the stroke connectivity

MFVL is introduced and applied to obtain key-points in strokes

from those key-points region growth and geometrical restriction is performed to obtain binary text image.Experiment demonstrates that our scheme can extract text with different color

font

size in complex image

and is better than method in Refs. .