News video is a very important video source. Caption in a news video can help us to understand the semantics of video content directly. A caption localization and extraction approach for news video will be proposed. This approach applies a new Multi-Frame Average (MFA) method to reduce the complexity of the background of the image. A time-based average pixel value search is employed and a Canny edge detection is performed to get the edge map. Then, a horizontal scan and a vertical scan on this edge map are used to obtain the top, bottom, left and right boundaries of the rectangles of candidate captions. Then, some rules are applied to confirm the caption. Experimental results show that the proposed approach can reduce the background complexity in most cases, and achieves a high precision and recall. Finally, we analyze the relationship between background variation of frame sequence and detection performance in detail.