Visibility in Computer Graphics Toni Sellarès Unversitat de Girona
Visibility Two points of a scene are mutually visible if the line segment connecting them is unoccluded.
Visibility Visibility problems can be classified based on the problem domain
Visibility taxonomy (*) 1. visibility along a line 2. visibility from a point 3. visibility from a line segment 4. visibility from a polygon 5. visibility from a region 6. global visibility (*) J. Bittner, P. Wonka, Visibility in Computer Graphics, Journal of Environment and Planning B: Planning and Design, vol. 30. no. 5, pp 729 – 725, 2003
Visibility along a single line All visual events restricted to a single line. The visibility along a line problems can be solved by computing intersections of the given line with the scene objects.
Visibility along a single line Point-to-point visibility determines whether the line segment between two points is unoccluded, i.e. the line segment has no intersection with an object in the scene. Ray shooting: given a ray, determine the first intersection of the ray with a scene object.
Visibility from a point Is the classical visible surface determination problem. Due to its importance for image synthesis, visible surface determination (hidden-surface removal) covers the majority of existing visibilty algorithms in computer graphics.
Visibility from a region Is a most general problem. Region-to-region visibility problem determines if the two given regions in the scene are mutually visible, invisible or partially visible. Other examples: potentially visible sets, aspect graph, 3D visibility complex
Hidden-surface removal When a 3-dimensional scene description is transformed to screen coordinates by a projection transformation, more than one object may overlap the same sample in the image. Some method is needed to determine which is the visible one, so that pixels can be shaded appropriately.
Object-Space Methods Use geometric tests on the object descriptions to determine which objects overlap and where. Object space algorithms are O(n2) where n is the number of objects in the scene. Complexity of algorithms is high, few primitives are supported, and special effects often are difficult to implement Examples: painter's algorithms, depth-sorting, BSP
Image-Space Methods Many approaches take advantage of the discretized nature of the output image, computing visibility only to the precision required to decide what is visible at a particular pixel. Image space algorithms are O(nN) where n is the number of objects in the scene and N is the number of pixels. Allow more advanced effects and complex models. Examples: z-buffer, Watkin's, ray-casting
Why might a polygon (face) be invisible ? Polygon outside the field of view Polygon is backfacing Polygon is occluded by object(s) nearer the viewpoint We want to avoid spending work on polygons outside field of view or backfacing We need to know when polygons are occluded
Frustun culling Quickly reject what is not visible Conservative: never classify a visible scene element as invisible Reduce unecessary processing
Hierarchical Frustum Culling bouding volume hierarchy scene elements
Hierarchical Frustum Culling bouding volume hierarchy A B D C view frustum
Backface Removal (Backface Culling) Removes entire polygons that face away from the viewer If we are dealing with a single convex object, culling completely solves the hidden surface problem Geometric test for the visibility
Note: backface culling alone doesn’t solve the hidden-surface problem! On the surface of an object, polygons whose normals point away from the camera are always occluded: Note: backface culling alone doesn’t solve the hidden-surface problem!
Not rendering backfacing polygons improves performance Back-Face Culling Not rendering backfacing polygons improves performance Reduces by about half the number of polygons to be considered for each pixel
Occlusion For most scenes, some polygons will overlap: To render the correct image, we need to determine which polygons occlude which
Conservative occlusion testing If the box representing a node is not visible then nothing in it is either The faces of the box are projected onto the image plane and tested for occlusion occluder hierarchical representation
Testing a Node for Occlusion If the box representing a node is not visible then nothing in it is either The faces of the box are projected onto the image plane and tested for occlusion occluder hierarchical representation
Occlusion Removal Algorithms Painter’s Algorithm Hybrid BSP tree Algorithm Hybrid Z-Buffer Algorithm Image-space Ray-Casting Algorithm Object-space
Painter's Algorithm Sort all polygons according to the (farthest) z coordinates of each, When the polygon's z extents overlap, resolve the sorting ambiguities by splitting the polygons Draw (fill) the polygons in the sorted order back to front Also known as Depth-sort algorithm.
Painter's Algorithm Does polygon P obscures polygon Q? Do the polygons' x extents overlap? Do the polygons' y extents overlap? Is P entirely on the opposite side of Q's plane from the viewpoint? Is Q entirely on the same side of P's plane from the viewpoint? Do the projections of the polygons onto screen not overlap?
Painter’s Algorithm Some cases in which Z extents of polygons overlap
Binary Space Partitioning Tree Algorithm Very efficient for a static group of 3D polygons as seen from an arbitrary viewpoint Correct order for Painter’s algorithm is determined by a suitable traversal of the binary tree of polygons: BSP Tree
BSP Tree
BSP Tree A B C
BSP Tree Draw BSP Tree function draw(bsptree tree, point eye) if tree.empty then return if ftree.root(eye) < 0 draw (tree.right) rasterize(tree.root) draw(tree.left) else draw (tree.left) draw(tree.right) tree.root tree.left tree.right
BSP Tree rasterize(C) rasterize(B) rasterize(A) rasterize(A)
BSP Tree Code works for any view Tree can be pre-computed Requires evaluation of fplane of the triangle(eye)
BSP Tree Construction The binary tree is constructed using the following principle: For each polygon, we can divide the set of other polygons into two groups One group contains those lying in front of the plane of the given polygon The other group contains those in the back The polygons intersecting the plane of the given polygon are split by that plane
Summary: BSP Trees Pros: Cons: Simple, elegant scheme Only writes to framebuffer (i.e., painters algorithm) Thus very popular for video games (but getting less so) Cons: Computationally intense preprocess stage restricts algorithm to static scenes Worst-case time to construct tree: O(n3) Splitting increases polygon count Again, O(n3) worst case
Z-Buffer Algorithm Color values are stored in a frame buffer. We also need a z-buffer, of the same size as the frame buffer, to store z value of each image pixel. Frame-Buffer Z-Buffer
Z-Buffer Algorithm -far -far -far Key Observation: Each pixel displays color of only one polygon, ignores everything behind it. Don’t need to sort polygons, just find for each pixel the closest triangle. Z-buffer: one fixed or floating point value per pixel. -far -far -far -far -far -far -far -far -far zbuffer framebuffer 35
Z-Buffer Algorithm -far -.1 -.2 Key Observation: Each pixel displays color of only one polygon, ignores everything behind it. Don’t need to sort polygons, just find for each pixel the closest triangle. Z-buffer: one fixed or floating point value per pixel. -.3 -.4 -.5 -.6 -.7 -.8 -far -far -far zbuffer framebuffer 36
Z-Buffer Algorithm -far -.1 -.2 Key Observation: Each pixel displays color of only one polygon, ignores everything behind it. Don’t need to sort polygons, just find for each pixel the closest triangle. Z-buffer: one fixed or floating point value per pixel. -.3 -.4 -.3 -.1 -.7 -.8 -far -far -far zbuffer framebuffer 37
Z-Buffer Algorithm -far -.1 -.2 Key Observation: Each pixel displays color of only one polygon, ignores everything behind it. Don’t need to sort polygons, just find for each pixel the closest triangle. Z-buffer: one fixed or floating point value per pixel. -.3 -.4 -.3 -.1 -.7 -.8 -far -far -far zbuffer framebuffer 38
Z-Buffer Pseudo-code For each rasterized fragment (x,y) If z > zbuffer(x,y) then framebuffer(x,y) = fragment color zbuffer(x,y) = z
Ray Casting Find nearest surface along the view ray (ray shooting). eye for i = 0 to nRows-1 for j = 0 to nCols-1 ray = genRay(eye, pixel(i,j)) castRay(ray)
Ray Casting Cast ray(ray) for every polygon in the scene intersect ray with polygon store the color at point of intersection and the distance from ray origin sort the intersection points according to distance draw_pixel(color)
Ray Casting Find nearest surface along the view ray (ray shooting). eye for i = 0 to nRows-1 for j = 0 to nCols-1 ray = genRay(eye, pixel(i,j)) castRay(ray)
Ray Casting V eye W U
Ray Casting V U eye W