In order to solve the problems that the active contours cannot converg e into deep indentation and poor segmentation results of images with weak edge structures,we propose an adaptive diffusion gr adient vector flow algorithm.First,in the external force field,we adopt the normalization method based on component instead of the normalization method based on vector.It can improve the ability of curve to converge into deep depression.Second,we decompos e the Laplace operator into tangential and normal components,and add two cross-correlation adaptive weighting functio ns.These two functions can make the curve adjust the diffusion process adaptively on the basis of local image features.Fin ally,we compare and analyze the segmentation results of several active contour models using quantization errors.Experimental results show that the convergence problem of deep indentation has been solved effectively.Aiming at two different images which ow n different image structures and weak boundaries, quantitative errors of the segmentation results reduce to 0.08and 0.09,r espec tively.Active contours can converge to the bottom of different deep depressions.The curves can also converge to the edges and corner s according to the local structure features of images. The results of segmentation on real images are much more accurate.