Depth Map Reconstruction and Enhancement With Local and Patch Manifold Regularized Deep Depth Priors

The depth map captured by sensors (e.g., the time of flight (ToF) and Kinect) is often prone to low resolution, degradation, noise, poor quality. This paper proposes a novel model for robust estimation RGB-D images through local nonlocal manifold regularizations. first stage called deep prior (DDPM), inspired partly (DDP) model, that convolutional neural network (CNN) integrated with regularization term. neighboring relationships between pixels color are employed promote smoothing in results. Laplacian Eigenmap technique used modeling produces over-smooth map. To improve quality reconstructed image, was suggested, where similarity corresponding image determined characterizing their matching aspects. These objectives aggregated within an optimization problem. Moreover, extract edges better considering visual characteristics, structured low-rank Hankel approximation adopted eliminate degradations, highly promoted sharp points. Three types degradations were handling this work, containing undersampling, ToF-like, Kinect-like degradations. Experimental results indicate proposed method outperformed state-of-the-art restoration techniques on standard benchmark images, terms well-known criteria like PSNR.