By characterizing pixel distance with RGB pixel value, based on graph cuts theory, an human-machine interactive fast object extraction method is proposed. Closed polygons are artificially drawn around the object as the initial contour line, a single-side variable width cyclic neighborhood is generated to avoid ineffective overlapping and repeated cutting. Meantime, an energy function is constructed and a s-t network is generated to achieve the object extraction through the minimus cost cutting of the s-t network. Repeat the above steps until converge to the best boundary of the object. Later, convenient, fast, safety-oriented, automatic and manual error correction measures are provided for local errors. Experiments show that method has the advantages of convenient human-machine interaction, efficient and complete error correction, and fast and accurate object extraction.