Abstract:

Depth images have some problems such as edge mismatching, invalid pixels and noise, which are induced by the measuring distance principle. We present a depth image enhancement method based on the improved anisotropic diffusion algorithm. Firstly, we correct the position relationship between the depth image and the color image, and choose more frames to preprocess the multi-frame mean filtering according to the time continuity. Secondly, we build a depth image model with 4-neighborhood by introducing the idea of weights to the color image. The anisotropic diffusion is conducted on the depth image under color image guidance and the holes are filled. Finally, we use the improved adaptive median filtering smooth algorithm to remove image noise. Experimental results show that the proposed method can effectively repair the black holes composed of invalid pixels in the original depth image and remain the detailed information of the object’s edge in the depth image while suppressing the noise.

Key words: color image guidance, anisotropic diffusion, holes filling, depth image denoising