Comparing with Other Databases

CUReT

CUReT (Columbia-Utrecht Reflectance and Texture Database) developed at Columbia University and Utrecht University has collaborated in an extensive investigation of the visual appearance of real-world surfaces. It comprises three texture databases:

• BRDF database (bi-directional reflectance distribution function;
• BRDF parameter database;
• BTF (bi-directional texture function) database.

We have to note that, for all measurements of a selected texture sample, the light source remains fixed. However as the camera is mounted on a tripod, with its optical axis parallel to the floor of the lab. The relative positions of texture sample, illuminant and camera used in CUReT database are shown on the left. The measurement setup for CUReT database is not suitable for photometric stereo, as for a given texture sample at a certain orientation we need to capture the different images by fixing the position of camera and moving the light source, not by fixing the light source and moving camera.

---

OUTex

OUTex (University of Oulu Texture database), is a framework for the empirical evaluation of texture classification and segmentation algorithms. At this time, the collection of 319 surface textures are captured by well defined variations to a given reference in terms of illumination directions, surface rotations and spatial resolutions.

However, the OUTex database cannot used to provide registered photometric stereo data. For example, although texture images of 3D surfaces are taken under three different illumination directions, all of the three involving illumination directions are lying on the same plane (coplanar or collinear), shown on the left. In the other words, all of these three illumination conditions only vary in the change of slant angle (s), the tilt angle (t) is constant. Therefore we can not correctly resolve the surface partial derivatives from these images using photometric stereo, as the inverse of the lighting matrix in the photometric stereo solution does not exist when the three illumination vectors lie in the same plane.

---

PhoTex

PhoTex database is a texture database of rough surface. It holds images of rough surfaces that have been illuminated from various directions. Images that allow the use to calibrated the image transfer function, and measure the noise in the process are also held in the database. The main variables in the database are azimuth and zenith of the illumination. In few cases, the surface sample is also rotated. This database therefore mainly focuses on the changes of illumination condition rather than the surface rotation. In fact, although it is a photometric texture database that is mostly close to our requirement, it does not provide full surface rotation samples.

---

VisTex

VisTex (Vision and Modeling Group at the MIT Media Lab) collection has been assembled and maintained to assist in the development of more robust computer vision algorithms and their comparison on a common set of data. It does not conform to rigid frontal plane perspectives and studio lighting conditions. Their lighting conditions include daylight, artificial-florescent and artificial-incandescent. Moreover, some of the lighting conditions are imprecise. For example, descriptions are given as “daylight, direct and from right”.

Therefore, considering the limitations of the VisTex database with unknown illumination directions, we can not use them with our 3D texture analysis scheme, because we have to recover the surface properties from several images with the known and controlled light conditions using photometric stereo.

---

MeasTex

MeasTex (the MEASurement of TEXture classification algorithms), an image database and quantitative measurement framework for image texture analysis algorithms. It is not only a texture database, but also a frame work for the quantitative measurement of texture algorithms targeted on a number of texture testing suites, and an implementation of some major well-known texture classification paradigms.

Although a number of texture sets in MeasTex have been compiled by other texture databases such as the Brodatz texture database and the VisTex database, most of their natural textures are 2D texture rather than 3D texture. We do not know what the exact illumination direction is when the real texture is captured. Therefore this database is not suitable for our 3D surface scheme.

---

Brodatz

The “Brodatz texture database” is derived from the Brodatz album. It has a relatively large number of classes (112 classes), and a small number of examples for each class. Although the Brodatz texture database has become the standard for evaluating texture algorithms, it is not suitable for use by our classification scheme.

Our experiment requires that the photometric stereo algorithms are able to run on textures taken under different and controlled lighting and perspective. The original Brodatz texture database is not suitable for our experiments as it can not provide photometric stereo image sets for each texture classes and it does not provide true surface rotation.