Total variation based ampoule injection image denoising with universal gravity theory

摘要

Gravity edge detector based adaptive total variation denoising model (Gra-ATV) is addressed in this paper to remove noises within captured ampoule injection images. In the proposed algorithm, each image pixel is considered as a celestial body with a mass represented by its grayscale intensity. Vector sum of gravitational forces related to this pixel is calculated then. If its value is larger than the global threshold, an edge point can be achieved. To prevent the occurrence of the 'staircasing effect' and save the fine features, the regularisation term and the fidelity term of Gra-ATV model change adaptively according to whether the current operating pixel is at an edge or in flat areas. Tests and comparisons between the proposed algorithm and L_2 norm, L_1 norm based diffusion model, p-TV model and Gauss-TV model are carried out using randomly selected ampoule injection images. The experimental results show that proposed Gra-ATV model can remove noises while preserving the fine details, and higher efficiency also can be achieved.