Multi-scale mutual feature convolutional neural network for depth image denoise and enhancement
Xuan Liao, Xin Zhang · 2017
RGB-D images captured by consumer camera can provide pair-wise color and depth information but depth image usually contains strong noise and large holes. Due to different modalities of RGB-D, the intensity-guided depth enhancement easily causes artifacts and blurred edges. To solve this problem, we propose the multi-scale mutual feature convolutional neural network (MSMF-CNN) to learn essential mutual features of RGB-D. This end-to-end framework has two stages, i.e., mutual feature learning and depth regeneration. Firstly, we design two parallel separated subnetworks as intensity and depth mutual feature generator. Specifically, we design a multi-scale mutual feature generator to enforce the depth mutual feature learning and incorporate various structure characteristic. Secondly, depth transformation sub-network combines two mutual features to recover clean depth image. We tested our model on two well-known datasets in terms of PSNR, visual effect and computing speed. Compared with state-of-art algorithms, our method provides better performance of all three criteria.