Abstract: An improved active contour model is presented for segmentation of medical image sequences. A restriction is applied to an internal energy function to make the length of contour as short as possible so as to prevent contour from converging at sharper points of other tissue on perpendicular bisector. An external energy function is constructed from the information pertinence of local region between the serial images to eliminate the interference of texture and capture corner points completely. Experimental results show that the improved algorithm can give better segmentation, which can detect any corner points and concave points, and avoid object edge converging at pseudo edge points.

Key words: active contour model, energy function, serial images, segmentation