An approach to video text localization based on gradient discrete cosines transform

Video text information plays an important role in semantic-based video analysis, indexing and retrieval. The grade of gray in the video text and background is different, so an approach to video text localization using gradient discrete cosines transform (DCT) is proposed. Firstly, the video frame is divided into N×N sub-blocks, and we can get the DCT coefficients from every sub-block. After that, the gradient operator value considered as block intensity is calculated, and we deal with it using smooth filter and morphology processing. Lastly, the horizontal and vertical projection can be obtained, the number of text line can be got and the text region is marked with text box. The experimental results show that the proposed method is efficient on the localization of the static and rolling video text, especially the accuracy and integrality of character localization with the fewer strokes.