Dmesh: Fast Depth-Image Meshing and Warping

Abstract

In this paper we present a novel and efficient depth-image representation and warping technique called DMesh which is based on a piece-wise linear approximation of the depth-image as a textured and simplified triangle mesh. We describe the application of a hierarchical multiresolution triangulation method to generate adaptively triangulated depth-meshes efficiently from reference depth-images, discuss depth-mesh segmentation methods to avoid occlusion artifacts and propose a new hardware accelerated depth-image rendering technique that supports per-pixel weighted blending of multiple depth-images in real-time. Applications of our technique include image-based object representations and the use of depth-images in large scale walk-through visualization systems.

Dynamic feathering: minimising blending artefacts in view-dependent rendering

“…Such scores could be defined in terms of the magnitude of the angle between camera i and the rendering camera by computing for example the inverse of this angle (see [5]) or its cosine (see [17]). More sophisticated measures which take into account the Euclidian distance between image centres (see [6,13]), or even the camera resolution and field-of-view (see [2,21]) could also be considered. This choice is independent of the proposed feathering algorithm.…”

“…This is a simple strategy which typically works well for scenes where there are no extreme variations in zoom factors. More sophisticated selection strategies incorporating distance between image centres (see [6,13]), camera resolution and field-of-view [2,21] could also be considered. This was not necessary for the type of scene considered in this paper.…”

“…This approach is called soft z-buffer in [17] or ǫ-z-buffer in [13]. Such rendering is done in two passes.…”

“…These weights are then combined with other per pixel blending weights stored in the alpha channel of each rendered view by enabling alpha blending, and rendering them on a screen size quad. The per pixel blending weights could correspond to surface sampling weights as in [17], or per pixel quality measures as in [13]. In our implementation we weight each pixel with its opacity value which was pre-computed by a matting algorithm [11].…”

“…In [22,21], the uniform sampling of the blending field is replaced by a non-uniform sampling, where sampling density increases with surface curvature, thus enhancing blending quality. In [13], Pajarola et al use adaptive depth-image meshing to reduce the complexity of the representation and thereby speed-up rendering time. Hardware acceleration is used to efficiently compute per pixel blending weights for each input image, first rendering separately each input image using a soft z-buffer algorithm, then combining the separate outputs using alpha blending of quadrilaterals texture mapped with these images.…”