Image segmentation is a fundamental and challenging problem in image processing and often a vital step for high level analysis. Considering of the inefficient curve evolution against weak boundary and intensity heterogeneous images, an improved level set segmentation framework based on image directional gradient is proposed. In this framework, we divide the evolution processing into two stages: the evolution of the image background and foreground, and the evolution of the image regions based on the chosen directional gradient. Compare to the other local information based active contour evolution algorithm: Local Binary Fitting (LBF) model, this algorithm may improve efficient of curve evolution in a large extent. Extensive experiments on synthetic and real images are provided to evaluate our method, showing the segmentation of the blurry boundary and intensity heterogeneous images may achieve more accuracy results.