Detecting subtitles in a video signal

Abstract

A method and arrangement (300) are disclosed for detecting the presence, appearance or disappearance of subtitles in a video signal. A very high reliability can be achieved, and a marginal processing power is needed, due to the fact that most computations are already done by circuitry of an MPEG encoder (101–113) or decoder. A subtitle is detected if the complexity of the image area in which subtitles are displayed substantially exceeds the complexity of at least one other image area. Examples of properties representing the complexity are (i) the products of bit cost (b) and quantizer scale (qs) in MPEG slices, (ii) the location of the center of gravity of the spectral DCT coefficients (c), (iii) the number of macroblocks in the subtitle image area having a small motion vector (mv) versus the number of macroblocks having a large motion vector, or (iv) the fact that scene changes are not simultaneously detected in the different image areas. The arrangement can be used for commercial break detection or keyframe generation.

Dependencies are inferred from the claim text.