1. Run-Length Encoding for Texture Classification
Dong-Hui Xu
Visual Computing Research Seminar
CTI, DePaul University

2. Topics of Discussion Problem statement Motivation Background
Run-Length Matrices and the Eleven Run-Length features
Preliminary Results
Future Work
References

3. Problem Statement
We want to develop a texture vocabulary that defines the different human body tissues in terms of low-level texture descriptors.

4. Motivation
Our hope is that our classification of tissues will help radiologists detect irregularities (ex. Tumors) in the tissues of the human body sooner.
Earlier detection can help save lives.

5. Background Q. What is texture?
A. Texture is the term used to characterize the surface of a given object or region. It is described as fine, coarse, grained, smooth, etc,

6. Background: Examples of Textures
These images are taken from Brodatz Textures. They are benchmarks
that researchers use in order to test if their algorithms are working properly.

7. Background Basic concepts for texture:
Texture primitives – maximum contiguous set of constant-gray-level pixels
Three features can be defined for textures:
Tone of texture (Gray-Level) – Based mostly on pixel intensity properties in the primitive
Structure of texture (Direction) – Spatial relationship between texture primitives
Length of the primitive (long = coarse and small = fine)

8. Ways to Characterize Texture
Co-occurrence matrices
Discrete Wavelet Transform
The Power Spectrum features
Run-Length encoding

9. Definitions for gray level runs
Galloway proposed the use of a run-length matrix for texture feature extraction
For a given image:
A gray level run is defined as
A set of consecutive, collinear pixels having the same gray level
Length of the run is
The number of pixels in the run

10. Definition of Run-Length Matrices
The run-length matrix p (i, j) is defined by specifying direction.
0 °, 45 °, 90 °, 135 °
and then count the occurrence of runs for each gray levels and length in this direction
Dimension corresponds to the gray level (bin values) and has a length equal to the maximum gray level (bin values) n
(j) dimension corresponds to the run length and has length equal to the maximum run length (bin values). j i 1 2 3 4 5 6 8
0 °

11. Definition of Run-length Features
Short Run Emphasis
nr is the total number of runs in the image.
M is the number of gray levels (bins)
N is the number of run length (bins)
The number of runs of pixels that have gray level i and length group j is represented by p (i, j)
SRE feature measures the distribution of short runs
The SRE is highly depend on the occurrence of short runs and is expected large for fine textures.

12. Definition of Run-length Features (Continued)
Long Run Emphasis
LRE feature measures distribution of long runs
The LRE is highly depend on the occurrence of long runs and is expected large for coarse structural textures.

13. Definition of Run-length Features (Continued)
Low Gray-Level Run Emphasis
Measures the distribution of low gray level values
High Gray-Level Run Emphasis
Measures the distribution of high gray level values

14. Definition of Run-length Features (Continued)
Short Run Low Gray-Level Emphasis
Short Run High Gray-Level Emphasis
Long Run Low Gray-Level Emphasis
Long Run High Gray-Level Emphasis
Measures the joint distribution of run and gray level distribution

15. Run-length Features (Continued)
Gray-Level Non-uniformity
Measures the similarity of gray level values through out the image
The GLN is low if the gray levels are alike through out the image.
Run Length Non-uniformity
Measure the similarity of the length of runs through out the image
The RLN is low if the run lengths are alike through out the image.

16. Run-length Features (Continued)
Run Percentage
Measures the homogeneity and the distribution of runs of an image in a given direction.
The RP is the highest when the length of runs is 1 for all gray levels.

17. Result
Run-length features for one slice:

18. Results
Run run-length application on segmented images and the four quadrants of the segmented images
4 directions (0°, 45°, 90° and 135°)
calculate 11 descriptors from the run-length matrices

19. Results (Backbone - Sample)

20. Results (Backbone_P1)

21. Results (Backbone)

22. Results
Correlation Coefficients for Run-Length Descriptors

23. Future Work Investigate run-length matrices for volumetric data
Run run-length application over more patient images.
Use neural networks and statistic analysis technique to identify patterns for each organ.
Build a texture vocabulary that defines the different human body tissues in terms of low-level texture descriptors.

24. References
S.A. Karkanis On the Importance of Feature descriptors for the Characterisation of Texture.
Xiaoou Tang Texture Information in Run-Length Matrices