Denoising infrared structured light DIBR signals using 3D morphological operators

The advent of low cost infrared structured light cameras (such as MS Xbox Kinect sensors) has brought the need for effective denoising algorithms that improves the quality and the accuracy of the reconstructed 3D scene. The overall noise depends on diverse factors and leads to significant alterations on object borders in the depth signal. It is possible to exploit a color signal acquired by a standard RGB camera to correct these alterations and interpolate the missing values. The paper presents a denoising and interpolation strategy that adopts 3D morphological operators to smooth and regularize volumes generated from the structured light cameras. Experimental results show that both the number of valid points and the quality of the warped views improve with respect to other regularizing approaches.