MeasTex: Texture Images

• Brodatz images
• VisTex images
• MeasTex images
• "Ohanian and Dubes" images

Not all of the Brodatz images are suitable for texture analysis research. For example, the analysis of non-homogeneous textures introduces a level of complexity beyond the analysis of homogenous textures. Yet the analysis of homogeneous textures is still an open field of research. Hence, we have contrained MeasTex to the analysis of homogeneous textures. However, not all of the Brodatz images contain homogeneous textures.

For copyright reasons, we do not make these images available for anonymous copying.

Images of the Brodatz textures, for research purposes, are available from USC Signal and Image Processing Institute.

VisTex Images

The VisTex images are from the Vision Texture (VisTex) database maintained by the Vision and Modelling group at the MIT Media Lab. The full database contains images representative of real-world textures under practical conditions (lighting, perspective etc).

Two image sets are currently available covering homogeneous textures and multi-texture scenes. Both sets are distributed as raw PPM files with annotations declaring image content, lighting conditions, and perspective. The homogeneous textures (called Reference Textures) are distributed as both 128x128 pixel and 512x512 pixel images and the multi-texture scenes are distributed as both 192x128 pixel and 786x512 pixel images.

The full distribution is approximately 165 Megabytes (compressed) and may be obtained via ftp.

MeasTex Images

A number of image sets have been compiled by the authors to supplement the above sets. These sets contain examples of artificial and natural textures. We use the term natural here to mean textures which occur in the real world. The natural images have been chosen for their homogeneity of texture. Each image has a size of 512x512 pixels and is distributed in raw PGM format.

Artificial Textures

The artificial textures are generated by the makeBrick program (available in the MeasTex distribution), which creates brick-like textures. The program initially creates a stone texture background by randomly bombarding a blank image with gaussian "blobs". The "blobs" may be elongated and rotated. Brick textures may then be created by either overlaying the stone texture with a mortar pattern OR transferring randomly cut bricks from the stone texture and placing in a new image, separating the bricks with mortar. Thus, three texture sets are available

• bomb - normal background texture
• lattice - overlaying normal background texture with mortar
• mortar - separating randomly cut bricks with mortar

The artificial brick textures used in the MeasTex test suites are available with the MeasTex distribution in two tar'ed and compressed files ; one containing bomb and mortar, and the other containing lattice textures.

Natural Textures

A database of natural textures has been compiled (and is still evolving). The images have been obtained from 6"x4" colour photographs taken with a 35 mm camera. Each photograph was scanned at 256 dpi and stored in PPM format. The distributed images are 2"x2" (512x512 pixels) areas which have been cropped from these full-size images and converted to PGM format. An annotation header, similar to the VisTex annotations, has been used to label each image. A comprehensive database of grasses (most labelled with species) make up a large proportion of the textures in this set. Also significant is a number of landscaping materials (gravel, mulch, sand, rocks, bark etc). The following table has been compiled from the header information. Example images of each texture have been compiled.

	> gunzip -c meastex_newtexRawA.tar.gz | tar xvf -
	> gunzip -c meastex_newtexRawB.tar.gz | tar xvf -
	> gunzip -c meastex_newtexRawC.tar.gz | tar xvf -
	> gunzip -c meastex_newtexRawD.tar.gz | tar xvf -
OR
	> gunzip meastex_newtexRawA.tar.gz
	> tar xvf meastex_newtexRawA.tar
	> gunzip meastex_newtexRawB.tar.gz
	> tar xvf meastex_newtexRawB.tar
	> gunzip meastex_newtexRawC.tar.gz
	> tar xvf meastex_newtexRawC.tar
	> gunzip meastex_newtexRawD.tar.gz
	> tar xvf meastex_newtexRawD.tar

A comparative study of a number of texture classification algorithms was presented by Ohanian and Dubes (1992). Five test problems were constructed from examples of four image types ; fractal, Markov random fields, leather, and painted surface. These texture problems are contained in the Ohanian and Dubes test suite in the MeasTex framework.

The complete original texture images may be obtained from the meastex_ohandube.tar.gz file in the MeasTex distribution.

Philippe P. Ohanian and Richard C. Dubes (1992), "Performance Evaluation for Four Classes of Textural Features", Pattern Recognition, v25(8), pp819-833