Surface normal smoothing for superpixels in noisy depth images

Abstract

A new smoothing algorithm to obtain surface normals for superpixels in the noisy depth image is proposed. For smoothing surface normals, the plane normal that indicates a surface characteristic is obtained by considering the inherent geometric similarity. This process results in smoothed surface normals which are free from noise of the depth image. The superpixels method with suggested smoothing algorithm outperforms in higher quality than the state-of-the art oversegmentation algorithms. The proposed technique can also be applied to any smoothing algorithms using surface normals for 3D features. Experimental results to demonstrate the effectiveness of the proposed smoothing surface normal technique is provided.