Directionally weakened diffusion for image segmentation using active contours

Abstract

The active contour model, also known as the snake model, is an elegant approach for image segmentation and motion tracking. The gradient vector flow (GVF) is an effective external force for active contours. However, the GVF model is based on isotropic diffusion and does not take the image structure into account. The GVF snake cannot converge to very deep concavities and blob-like concavities and fails to preserve weak edges neighboring strong ones. To address these limitations, we first propose the directionally weakened diffusion (DWD), which is anisotropic by incorporating the image structure in a subtle way. Using the DWD, a novel external force called directionally weakened gradient vector flow (DWGVF) is proposed for active contours. In addition, two spatiotemporally varying weights are employed to make the DWGVF robust to noise. The DWGVF snake has been assessed on both synthetic and real images. Experimental results show that the DWGVF snake provides much better results in terms of noise robustness, weak edge preserving, and convergence of various concavities when compared with the well-known GVF, the generalized GVF (GGVF) snake.