Level Set Method Based on Improved Mumford-Shah Model Applied in Wood Cell Image Segmentation

In this study, we developed an improved method based on simplified Mumford-Shah model proposed by Chan and Vese. This method was proved to be much suitable for wood cell image segmentation. Typical softwood cell image is characterized by an obvious non-crossing boundary separating springwood and latewood. Both spring and late wood sections have gray scopes. Therefore, we changed the parameter c₁ (inside mean gray) into c_ding (core gray value of certain object, springwood or latewood). The value of c_ding was determined by transcendental information and experiments. With various c_ding settings, it came to different segmentation results. Optimized c_ding for typical softwood cell image had been gotten. Softwood cell images has similar gray characterizes, so the optimized c_ding has very important significance for other softwood cell images segmentation. Results of the experiments indicated that this improved method reduced blindness of the model, increased efficiency, improved the effectiveness of segmentation, and helped to build the foundation for classification.