Hidden Surface Determination - Hidden Surface Removal Algorithms

Hidden Surface Removal Algorithms

Considering the rendering pipeline, the projection, the clipping, and the rasterization steps are handled differently by the following algorithms:

• Z-buffering During rasterization the depth/Z value of each pixel (or sample in the case of anti-aliasing, but without loss of generality the term pixel is used) is checked against an existing depth value. If the current pixel is behind the pixel in the Z-buffer, the pixel is rejected, otherwise it is shaded and its depth value replaces the one in the Z-buffer. Z-buffering supports dynamic scenes easily, and is currently implemented efficiently in graphics hardware. This is the current standard. The cost of using Z-buffering is that it uses up to 4 bytes per pixel, and that the rasterization algorithm needs to check each rasterized sample against the z-buffer. The z-buffer can also suffer from artifacts due to precision errors (also known as z-fighting), although this is far less common now that commodity hardware supports 24-bit and higher precision buffers.
• Coverage buffers (C-Buffer) and Surface buffer (S-Buffer): faster than z-buffers and commonly used in games in the Quake I era. Instead of storing the Z value per pixel, they store list of already displayed segments per line of the screen. New polygons are then cut against already displayed segments that would hide them. An S-Buffer can display unsorted polygons, while a C-Buffer requires polygons to be displayed from the nearest to the furthest. Because the C-buffer technique does not require a pixel to be drawn more than once, the process is slightly faster. This was commonly used with BSP trees, which would provide sorting for the polygons.
• Sorted Active Edge List: used in Quake 1, this was storing a list of the edges of already displayed polygons. Polygons are displayed from the nearest to the furthest. New polygons are clipped against already displayed polygons' edges, creating new polygons to display then storing the additional edges. It's much harder to implement than S/C/Z buffers, but it will scale much better with the increase in resolution.
• Painter's algorithm sorts polygons by their barycenter and draws them back to front. This produces few artifacts when applied to scenes with polygons of similar size forming smooth meshes and backface culling turned on. The cost here is the sorting step and the fact that visual artifacts can occur.
• Binary space partitioning (BSP) divides a scene along planes corresponding to polygon boundaries. The subdivision is constructed in such a way as to provide an unambiguous depth ordering from any point in the scene when the BSP tree is traversed. The disadvantage here is that the BSP tree is created with an expensive pre-process. This means that it is less suitable for scenes consisting of dynamic geometry. The advantage is that the data is pre-sorted and error free, ready for the previously mentioned algorithms. Note that the BSP is not a solution to HSR, only a help.
• Ray tracing attempts to model the path of light rays to a viewpoint by tracing rays from the viewpoint into the scene. Although not a hidden surface removal algorithm as such, it implicitly solves the hidden surface removal problem by finding the nearest surface along each view-ray. Effectively this is equivalent to sorting all the geometry on a per pixel basis.
• The Warnock algorithm divides the screen into smaller areas and sorts triangles within these. If there is ambiguity (i.e., polygons overlap in depth extent within these areas), then further subdivision occurs. At the limit, subdivision may occur down to the pixel level.