Z-buffer limitation and A-buffer

Limitation of the Z-buffer

The Z-buffer stores only one surface (the nearest) per pixel. Therefore

• transparent or translucent surfaces cannot be rendered, because the surfaces behind a transparent one are discarded;
• partial pixel coverage cannot be stored, so edges are jagged (aliasing) and anti-aliasing is not possible. It also needs a large extra depth buffer.

How the A-buffer overcomes it

The A-buffer (accumulation buffer) keeps several surfaces for each pixel, so transparency and anti-aliasing are possible.

Each pixel position has a depth field and an intensity field:

• If depth ≥ 0, one surface covers the pixel: the depth field holds its depth and the intensity field its colour.
• If depth < 0, multiple surfaces contribute: the intensity field points to a linked list of surface records.

Pixel: [ depth < 0 | pointer ] -> [RGB, opacity, depth,
         area coverage %, surface id, next] -> [...] -> NULL

Each record holds RGB colour, opacity, depth, percent area coverage, surface id and a pointer to the next record.

Working: (1) polygons are scan-converted into the buffer, with subpixel masks giving the area coverage; (2) for each pixel the surfaces are sorted by depth; (3) the final colour is found by blending the colours of all contributing surfaces, using their opacity and coverage (front to back).

Hence transparent surfaces show those behind, and edge pixels get area-averaged colour (anti-aliased). The price is more memory and time than the Z-buffer.