关键词: 图像去噪, 各向异性扩散, 图像特征增强, 特征指标, 梯度变差指标

Abstract:

This paper presents an image feature enhancement diffusion model. Since the coherence-enhancing anisotropic diffusion model, proposed by J.Weickert, often induces false edges in slippy regions and can not preserve the detail features effectively, the new model poses two characteristic indexes and one gradient variance index to finely describe much more image information than the previous work, such as corners and isolated noises except with slippy regions and edges, and defines eigenvalues based on the classification results such that the new diffusion tensor has large eigenvalues along both the gradient direction and edge direction in the slippy regions and at isolated noises, but has small eigenvalues along the two directions at corners, and has small eigenvalue along the gradient direction and large eigenvalue along the edge direction at edges. So it can remove noises efficiently and at the same time enhance edges and detail features. Theoretical analysis and numerical experiments show the effectiveness of the proposed model.

Key words: image denoising, anisotropic diffusion, image feature enhancement, characteristic index, gradient variance index