next up previous
Next: Conclusion Up: Content Previous: 3D representation of raster images

Comparison of different methods used to process raster image

We achieved different success of recognition of image content using the suggested methods.

For experiments we used program written in C++ (for Windows NT), OpenGL and GLUT. Source code of the program is platform-independent to facilitate the portability to IRIX based platforms as well. Structure of the program is shown in Fig. 10.


Figure 10: Structure of the program used for experiments

Regular grid was found unusable in practice. Because relief rises from very approximate area, outline of relief is so rough that the shape can not be recognized.

Adaptive meshes seems to be the best method for haptic visualization of pictograms. Edges of the object are distinct enough and transitions among adjacent triangles smooth enough, therefore object shape is well perceivable. Only height of relief is of importance. If the height is too small, user (who has no visual information about his location in relief) can lose haptic contact with relief edge.

Heightfield method gives better results for general raster images than for pictograms. The disadvantage of pictograms is that the edges are sharp defined. Heightfield method does not take into consideration these edges and therefore the shape is partly misshapen. It makes recognition difficult. If we have the same raster image with interpolated triangles and without interpolation, success of recognition is different as well. In case of no interpolation edges are at least partly preserved.

Negative edge relief gives good results for pictograms which do not contain too narrow parts.

Dita Nemcova
Thu Apr 6 10:53:43 MET DST 2000