An improved coherence diffusion method for image enhancement

Coherence diffusion is an important preprocessing step for analyzing oriented structures in the image. Previous coherence diffusion methods for image enhancement could not recognize weak boundaries. In this paper, an efficient diffusion approach is presented. A new structure tensor integrating the second-order directional derivative information is designed, which can precisely analyze complex weak edges in the image. By combining the proposed structure tensor and the classical one as complementary descriptor, the improved diffusion tensor is constructed to detect strong edges simultaneously. Furthermore, parallel AOS (Additive Operator Splitting) scheme is applied to implement numerical solution, which is faster than usual numerical approach. Promising experimental results of several real images demonstrate that the new diffusion approach can preserve important strong edges and weak edges precisely while removing the noise.