Abstract: In order to extract caption region in digital video, an algorithm is provided which first detects text event and gets edges, then makes a size restrict to the edges and eventually wipes off non-text regions according to the textual energy. The overlaying of detected horizontal edges and vertical edges enhances the text edges; the size restrict of edges helps to wipe off the non-text edges. The conception of pixel density α is presented in a threshold scope (αmin≤α≤αmax) and used as an auxiliary measure to wipe off texture-like regions. Eventually image projection is applied to get text regions. The combination of these methods guarantees the performance of this algorithm. Our experiments show that this arithmetic has satisfactory performance of correctness and computing speed.