Sparsely opaque three-dimensional textures: Towards understanding images of plants

Adlai A Waksman, University of Pennsylvania


Plants, which play a major role in natural scenes, consist primarily of piecewise opaque three-dimensional textural patterns. Surprisingly, very little work seems to have been done on the recovery or description of such patterns from single views. This dissertation studies these patterns, including both real vegetation (houseplants) and synthetic patterns. We emphasize the estimation of properties of the pattern from single, approximately orthographic images. In particular, we have investigated the following questions: (1) How far into a texture can one see from a given viewpoint? What is the probability that a given viewing ray will pass all the way through it? We show that the pattern's "transparency" depends on the thickness of the volume it occupies, the density of the opaque texture elements (texels), their distribution of areas, and their distribution of slants, but is independent of the texels' shapes or their tilt distribution. (2) What information can we infer about the distribution of sizes and orientations of the texels, and what can we conclude about the predominant direction of illumination? We show that qualitative properties of the texel orientation and size distributions can be recovered by examining run length histograms derived from cross-sections of an image of the texture. We also show how the light source direction is constrained by the shape of the texture's gray level histogram. (3) What can we conclude about the gray level histogram of a texture in an image taken from a given viewpoint under given illumination conditions? We study such histograms, and show that the variation (with light source direction) in the gray level histograms of real plant images resembles that of synthetic histograms derived from 3D textures in which the distribution of texel orientations is uniform. (4) An example of an application to real plants: how can we distinguish images of a plant that needs watering from images of the plant in good condition? We show that simple geometric and colorimetric methods can measure stem flaccidity and leaf pallor, which can indicate the thirstiness of a plant.

Subject Area

Computer science

Recommended Citation

Waksman, Adlai A, "Sparsely opaque three-dimensional textures: Towards understanding images of plants" (1996). Dissertations available from ProQuest. AAI9628021.