SCENE CHANGING AND TEXT DETECTION METHOD OF COMPRESSED NEWS IMAGES

摘要

PURPOSE: A scene changing and a text detection method of compressed news images is provided to reduce the time required for decoding images by detecting the image changing position without decoding a compressed video. CONSTITUTION: The scene changing and a text detection method of compressed news images includes following steps. At first, an image changing probability is detected by applying a brightness difference( Y) and a color correlation(pn) to a pseudo-neural network(S13). For all P-frames which are included in n-th GOP(group of picture) from which an image changing probability is detected, a number of intra-macro blocks is divided by a number of forward estimation macro blocks so as to attain a P-frame coefficient(Rp) which is required in image changing detection. For all B-frames which are included in n-th GOP from which an image changing probability is detected, a number of reverse estimation macro blocks is divided by a number of forward estimation macro blocks so as to attain a B-frame coefficient(Rb) which is required in image changing detection. The P-frame coefficient and the B-frame coefficient are used(S16) so as to detect an exact image changing frame from the GOP(S17). And, a special frame by applying the frames to a special edition effect detection algorithm when a correct image changing frame is not detected(S19).