Andrews logo

3-D (volumetric) Texture Analysis
The goal of this research is to develop new 3-D methods for analyzing texture in 3-D (volumetric) images.

Texture

Although there is no accepted mathematical definition for image texture, it can be thought loosely of as repeated patterns of pixels. The addition of noise to the patterns and their repetition frequencies result in textures that can appear to be random and unstructured. The following are examples of 2-D textured images.


Image texture can be used to: While many different computer vision applications use texture, its analysis generally falls into three categories: Texture classification is concerned with identifying the correct texture class to which a given pixel or region belongs. Texture segmentation partitions an image into separate regions of homogeneous texture. Shape recovery uses apparent distortions in surface texture to determine the shape of objects.
A key part of all texture all texture analysis algorithms is the calculation of texture features. Texture is not a property that can be defined given a single pixel value. Texture features are computed to characterize the relationship between a pixel and its neighbors. The purpose of texture features is to capture the essence of a texture with a vector of numbers. Since there is no accepted mathematical definition for texture, many different methods for computing texture features have been proposed over the years. Unfortunately, there is still no single method that works the best with all types of textures.

3-D images

2-D images are useful for capturing surfaces of objects. However, many applications (primarily in medical imaging) need to characterize the interior structures of objects. This need has led to the creation of various 3-D imaging technologies such as CT and MRI. The 3-D images these technologies produce can be viewed as a series of 2-D images (or slices) captured at different depths. While a 2-D image is a 2-D lattice of pixels, a 3-D image is a 3-D lattice of volumetric pixels (or voxels). These 3-D images are sometimes referred to as volumetic images to distinguish them from 2-D images containing 3-D rendered graphics. The following is a sample 3-D image (or volume) followed by the first five slices of the volume.

3-D Texture analysis

Volumes are often processed as a series of 2-D images. 2-D texture features are computed for pixels in each slice. Unfortunately, by processing volumes as a series of separate 2-D slices, texture information across slices is ignored. Various methods for computing 3-D texture features have been developed to include this extra texture information. These methods include the use of
Most of these methods are 3-D extensions of 2-D methods. Since texture analysis was first applied to 2-D images, there are significantly more methods to choose from for computing 2-D texture features than 3-D texture features. While many of these 2-D methods can be extended to 3-D, only a few have been. One of the focuses of my research is to extend other successful 2-D methods to 3-D and evaluate their effectiveness on 3-D images. In particular, I am interested in evaluating the usefulness of 3-D discrete cosine transform (DCT) features for classifying of texture in 3-D images. These features are computed using 3-D extension of a successful 2-D method that used 2-D DCT filters.

People

Those involved in this research include:

Publications

References

(Jafari-Khouzani et. al.) K. Jafari-Khouzani, H. Soltanian-Zadeh, K. Elisevich, and S. Patel. "Comparison of 2D and 3D wavelet features for tle lateralization." In Proc. of SPIE Medical Imaging 2004: Physiology, Function and Structure from Medical Images, volume 5369, pages 593-601, 2004.
(Kovalev et. al.) V.A. Kovalev, F. Kruggel, H.-J Gertz, and D.Y. von Cramon. "Three-dimensional texture analysis of MRI brain datasets." IEEE Trans. on Medical Imaging, 20(5):424-433,2001.
(Kurani et. al.) A.S. Kurani, D.-H Xu, J.D. Furst, and D.S. Raicu. "Co-occurance matrices for volumetric data." 7th IASTED Int'l Conf on Computer Graphics and Imaging, 2004.
(Lang et. al.) Z. Lang, R.E. Scarberry, Z. Zhang, W. Shao, and X. Sun. "A Texture-Based Direct 3D Segmentation System for Confocal Scanning Fluorescence Microscopic Images." Proc. of the 23rd Southeastern Symposium on System Theory IEEE Comput. Soc. Press, Los Alamitos, CA, 1991.
(Madabhushi et. al.) A. Madabhushi, M. Feldman, D. Metaxas, D. Chute, and J. Tomaszewski. "A novel stochastic combination of 3D texture features for automated segmentation of prostatic adenocarcinoma from high resolution MRI." Medical Image Computing and Computer-Assisted Intervention, volume 2878 of Lecture Notes in Computer Science, pages 581-591. Springer-Verlag, 2003.
(Ranguelova and Quinn) E. Ranguelova and A. Quinn. "Analysis and synthesis of three-dimensional gaussian markov random fields." Proc. of ICIP'99, Kobe, Japan, 1999.
(Reyes-Aldasoro and Bhalerao) C.C. Reyes-Aldasoro and A. Bhalerao. "Volumetric feature selection for MRI." Information Processing in Medical Imaging 2003, volume 2732 of Lecture Notes in Computer Science, pages 282-293. Sprinter-Verlag, 2003.
(Xu et. al.) D.-H Xu, A.S. Kurani, J.D. Furst, and D.S. Raicu. "Run-length encoding for volumetric texture." 4th IASTED Int'l Conf on Visualization, Imaging and Image Processing, 2004.

Please direct any questions or comments about the material on this page to   summersc AT andrews DOT edu