This paper presents an effective algorithm to extract key frames from video images based on the correlation coefficients derived from the overlapping regions of adjacent images. Firstly, the polynomial model is employed to fit the change trend of correlation coefficients. Secondly, the positions of key frames are located and extracted rapidly and correctly via the high correlation of overlapping regions of adjacent key frame images. Finally, the time-effectiveness of extracting key frames based on the degree of overlap is tested in a stance. The result shows that the proposed algorithm takes only near 1/4 of time consumed by single computing of the degree of overlap, with an accuracy up to 88% and it can achieve higher accuracy for extracting key frames with different requires of degree of overlap between video images. The panchromatic image generated from the mosaic of extracted key video frames shows no apparent missing of scenes, and the new algorithm demonstrates high reliability and transferability.