A new method for video text extraction in order to overcome the weakness of existing methods was proposed.First
multi-scale text location and text region refinement in text location step were adopted.Then an improved K-means clustering method was used to segment text regions to obtain a binarized image in text segmentation step.Finally
OCR software was used to recognize text.The results of experiment show that the method can effctively used in sensitive words detection in videos from the Internet.