Detecting organic materials with a CCD camera.

Absorption bands in the near-infrared are used to detect materials composed of organic molecules, in scenes imaged with a conventional CCD camera. A simple model of reflectance spectra (between 850 and 980 nm) is proposed and tested on a wide range of materials. An existing vision system that was designed to detect materials with high water content is tested on organic materials. The system cannot detect materials (such as cellulose and starch) that consist of chains of sugars. It is able to robustly detect materials such as fats and aliphatic plastics (in their pure form), whose molecules are essentially long chains of CH(2) and CH(3) groups. The ability of the system to detect plastic objects is limited by inorganic additives in the plastics.

Photometric stereo applied to diffuse surfaces that violate Lambert's law.

This paper investigates the use of photometric stereo (PS) with reflectance functions that are diffuse but not Lambertian. We show that, for the special case where light sources are arranged at 90° intervals around the optical axis, standard PS is not limited to Lambertian surfaces, and we define criteria for its use. A series of rough test surfaces are used as models for surface microstructure-we found that the Oren Nayar (ON) reflectance model accurately predicted the surfaces' reflectance functions. The ON model does not meet our theoretical criteria for using PS, but PS performs well in simulations if the microroughness is moderate (rms slope <0.3). When PS was applied to real surfaces, the estimated and actual slopes were highly correlated, but there were significant errors in the slope estimates for the rougher samples.

Photometric stereo with gradated extended sources for recovery of specular surfaces.

A photometric stereo technique is proposed that uses four extended sources and specular reflection to estimate surface topography. It is shown that if the intensity of incident light is weighted according to its zenith angle, then the radiance of surface facets will vary linearly with their slope. A simple system that approximates this lighting distribution is demonstrated. It is shown that surface slope in the range [-0.5, 0.5] can be recovered to within a multiplicative constant.

Detecting wet surfaces using near infrared lighting.

We propose a system to detect wet surfaces using near infrared LED lighting and a conventional CCD camera. There is a sharp fall in the transmission of light through water for wavelengths longer than 920 nm. By detecting this transition it is possible to robustly detect a film of water lying on a surface. This paper considers surfaces that are impermeable, are not hydrophilic, and are sufficiently reflective (reflectivity, rho>0.4). A system is demonstrated that can detect water on a wide range of surfaces that meet these conditions.

Shape recovery of anisotropic metal surfaces.

A method to recover the shape of metallic surfaces that have a directional microstructure is demonstrated. A simple model that shows how surface radiance varies with the azimuth of the light source is described and verified on real surfaces. The model predicts that if a light source is revolved around a flat facet then two peaks in radiance will be observed: tilting the facet against the grain of the material will cause the amplitudes of the peaks to change relative to one another; titling the facet with the grain will change the angles at which the peaks occur. It is shown that by measuring these effects, it is possible to estimate the slope of the facet within a limited range.

Evaluating Kube and Pentland's fractal imaging model.

The paper assesses the validity of a model, proposed by Kube and Pentland (1988), that relates a rough surface to its image texture. Simulation was used to assess whether a linear approximation is appropriate, and whether the optimal linear filter agrees with the predictions of Kube and Pentland's model. The predictions of the model about the image directionality were also assessed on real images. It was found that a linear model is capable of modeling the imaging process for surfaces of moderate roughness and Lambertian reflectance, and that, subject to a small modification, Kube and Pentland's model accurately predicts the relationship between surface and image spectra.