Key frame extraction, localization and segmentation of caption text in news videos

Caption text detection in news videos has become an important research topic due to its capability of providing more information and thus used to obtain keyword for annotating, summarization, building digital news library. The Key frame extraction method has been proposed which takes into account the pixel difference between 2 consecutive frames. The pixel difference will be less when the same text is present and less motion in the background. Further uniform edges present in the text region are used to differentiate them from frames having same background. Using this method 1.0 precision has been achieved. Text localization is done using the method of merging the text region horizontally and using projection profiles. 0.92 precision has been achieved. K-means algorithm performs better when text has complex background. The cluster size used is 3 and cluster having the segmented text is automatically detected and achieved an accuracy of 95.4%.