Removing Scattered Light in Biomedical Images via Total Variation Guided Filter

The scattered light is common in biomedical images. However, its removal is a challenging task. The challenge comes from two aspects. First, the scattered-light-free ground truth biomedical images are difficult to obtain or even unavailable. This is fundamentally different from the case in natural images where the ground truth is available. Second, although some neural network methods can remove the scattered light in biomedical images, they contain a large number of parameters, hampering their training process and the deployment in practical applications. To tackle these issues, this paper proposes a simple filter method that can effectively and efficiently remove the scattered light in biomedical images without knowing the ground truth. After analyzing the physical law behind the light scattering, we derive a novel model for biomedical images from a well-known mathematical model for the natural image de-hazing task. Then, we present a quarter-window dark channel prior for solving this model, leading to a fast filter with linear computation complexity. Finally, several numerical experiments are conducted to confirm the effectiveness of the proposed model and the efficiency of the proposed solving filter. Thanks to the effectiveness and the efficiency, the proposed method can be deployed in practical applications and achieve real-time performance.