Stereo matching algorithm for endoscopic images based on improved census transformation

In the domain of 3D reconstruction for medical imaging, binocular stereo matching technology plays a pivotal role in facilitating depth perception in endoscopic settings. However, due to the complex imaging environment of endoscopy, acquired images often exhibit challenges such as weak texture distribution on tissue surfaces, and uneven illumination reflection, which hinder the development of endoscopic 3D imaging technology. Therefore, a stereo matching algorithm based on guide filter and improved Census transform is proposed. During the matching cost computation stage, image grayscale and gradient information are fused to derive an enhanced Census matching cost, which is subsequently combined with the AD matching cost. Then, for each pixel, an adaptive cross-shaped window is built, and guided filter aggregation is carried out based on this adaptive window. Finally, the final disparity map is obtained by disparity calculation and disparity optimization. The experimental results demonstrate that the algorithm achieves higher matching accuracy on the Middlebury dataset than other stereo match algorithms.