Proposal for adaptive orientation of projection planes
Résumé
The current TMC2 projection method fixes the projection planes to the following six oriented planes which are defined by their normals: (1.0, 0.0, 0.0); (0.0, 1.0, 0.0); (0.0, 0.0, 1.0); (-1.0, 0.0, 0.0); (0.0, -1.0, 0.0) and (0.0, 0.0, -1.0). [1] In the original TMC2 Patch generation process, the point cloud is decomposed into a minimum number of patches with smooth boundaries, while minimizing the reconstruction error. The clustering of the point cloud is first initialized by computing the maximal dot product of the plane normal and the point normal which is estimated using PCA (Principal Component Analysis) as described in [2]. Then followed with refining iteration by updating the cluster index to smooth the orientation of its nearest normals. This method is low complexity and efficient to use as the normals are all integers. However, the parts of the decoded point cloud with orientation other than vertical or horizontal may contains holes, with significant visual impact, because they lead to a D1 layer that has a form which is not compression friendly. This proposal is experimenting an adaptive orientation method to reduce the size of the projected 2D patch and the number of missed points for the entire point cloud while minimizing the distortion