The invention discloses a foreground extraction method and device and belongs to the field of computer vision. The method comprises the steps: foreground extraction is performed on an image sequence, and an initial foreground block mass sequence; the amount of pixels contained by the foreground block mass is utilized to distinguish overlapping foreground block masses and non-overlapping foreground block masses in the initial foreground block mass sequence; first pre-cutting is performed on the overlapping foreground block masses, and a candidate cutting result is obtained; the non-overlapping foreground block masses are utilized to perform second pre-cutting, and a prediction cutting result is obtained; a distance between the candidate cutting result and the prediction cutting result are calculated, and a candidate cutting result corresponding to the shortest distance serves as a final cutting result; and a final foreground block mass sequence is outputted. According to the invention, a problem of hard-separated foreground and big foreground extraction error when overlapping and background interference occur on different foregrounds in the prior art is solved, the overlapping foregrounds can be effectively cut, and effects of foregrounds in an image sequence are precisely extracted.