Image texture segmentation using polar S-transform and principal component analysis
The present invention relates to a method and system for segmenting texture of multi-dimensional data indicative of a characteristic of an object. Received multi-dimensional data are transformed into second multi-dimensional data within a Stockwell domain based upon a polar S-transform of the multi-dimensional data. Principal component analysis is then applied to the second multi-dimensional data for generating texture data characterizing texture around each data point of the multi-dimensional data. Using a classification process the data points of the multi-dimensional data are partitioned into clusters based on the texture data. Finally, a texture map is produced based on the partitioned data points. The present invention provides image texture segmentation based on the polar S-transform having substantially reduced redundancy while keeping maximal data variation.
1. A method for segmenting texture of multi-dimensional data indicative of a characteristic of an object comprising:
receiving the multi-dimensional data;
transforming the multi-dimensional data into second multi-dimensional data within a Stockwell domain using a rotation-invariant form of the S-transform of the multi-dimensional data;
applying principal component analysis to the second multi-dimensional data for generating texture data characterizing texture around data points of at least a portion of the multi-dimensional data; and,
partitioning the data points of the at least a portion of the multi-dimensional data into clusters based on the texture data using a classification process.
2. A method for segmenting texture as defined in claim 1 wherein the rotation-invariant form of the S-transform is a polar S-transform.
3. A method for segmenting texture as defined in claim 2 comprising:
producing a texture map based on the partitioned data points of the multi-dimensional data, wherein data points of the multi-dimensional data within a cluster corresponding to a same texture region have a same texture value assigned thereto.
4. A method for segmenting texture as defined in claim 3 wherein the texture map is produced based on a probability of each partitioned data point of belonging to at least one of the clusters.
5. A method for segmenting texture as defined in claim 3 comprising:
superimposing the texture map to the multi-dimensional data such that texture values and data values of respective data points of the texture map and the multi-dimensional data are superimposed.
6. A method for segmenting texture as defined in claim 2 comprising:
determining a modified Stockwell spectrum by integrating local spectra along a radial direction.
7. A method for segmenting texture as defined in claim 6 comprising:
transforming the multi-dimensional data into a Fourier domain.
8. A method for segmenting texture as defined in claim 7 wherein transforming the multi-dimensional data comprises performing for each local spectrum corresponding to a data point of the multi-dimensional data:
calculating a current center frequency and a corresponding orientation angle;
calculating a localizing Gaussian window at the current centre frequency;
shifting the Fourier transformed multi-dimensional data by frequency components corresponding to the current centre frequency;
producing product data by pointwise multiplying the shifted Fourier transformed multi-dimensional data with the localizing Gaussian window;
inverse Fourier Transforming the product data; and,
updating the modified Stockwell spectrum based on the inverse Fourier transformed product data.
9. A method for segmenting texture as defined in claim 8 wherein a mean of the multi-dimensional data is assigned to a respective data point in the modified Stockwell spectrum if the centre frequency corresponding to the data point is zero.
10. A method for segmenting texture as defined in claim 6 wherein applying principal component analysis comprises projecting the second multi-dimensional data onto principal components.
11. A method for segmenting texture as defined in claim 10 wherein the principal component analysis is applied along a central frequency axis of the modified Stockwell spectrum.
12. A method for segmenting texture as defined in claim 11 wherein a number of significant principal components is determined based on an accumulate sum of corresponding eigenvalues of the modified Stockwell spectrum.
13. A method for segmenting texture as defined in claim 3 wherein the multi-dimensional data are MR image data.
14. A method for segmenting texture as defined in claim 13 wherein the data points of the at least a portion of the multi-dimensional data are partitioned into data points corresponding to image pixels representing normal appearing white matter and data points corresponding to image pixels representing non normal appearing white matter.
15. A storage medium having stored therein executable commands for execution on a processor, the processor when executing the commands performing:
receiving the multi-dimensional data;
transforming the multi-dimensional data into second multi-dimensional data within a Stockwell domain using a polar S-transform of the multi-dimensional data;
applying principal component analysis to the second multi-dimensional data for generating texture data characterizing texture around each data point of at least a portion of the multi-dimensional data; and,
partitioning the data points of the at least a portion of the multi-dimensional data into clusters based on the texture data using a classification process.
16. A storage medium as defined in claim 15 having stored therein executable commands for execution on a processor, the processor when executing the commands performing:
producing a texture map based on the partitioned data points of the multi-dimensional data, wherein data points of the multi-dimensional data within a cluster corresponding to a same texture region have a same texture value assigned thereto.
17. A storage medium as defined in claim 16 having stored therein executable commands for execution on a processor, the processor when executing the commands performing:
superimposing the texture map to the multi-dimensional data such that texture values and data values of respective data points of the texture map and the multi-dimensional data are superimposed.
18. A storage medium as defined in claim 17 having stored therein executable commands for execution on a processor, the processor when executing the commands performing:
determining a modified Stockwell spectrum by integrating local spectra along a radial direction.
19. A system for segmenting texture of multi-dimensional data indicative of a characteristic of an object comprising:
an input port for receiving the multi-dimensional data;
a processor in communication with the input port for:
transforming the multi-dimensional data into second multi-dimensional data within a Stockwell domain using a polar S-transform of the multi-dimensional data;
applying principal component analysis to the second multi-dimensional data for generating texture data characterizing texture around each data point of at least a portion of the multi-dimensional data; and,
partitioning the data points of the at least a portion of the multi-dimensional data into clusters based on the texture data using a classification process;
producing a texture map based on the partitioned data points of the multi-dimensional data, wherein data points of the multi-dimensional data within a cluster corresponding to a same texture region have a same texture value assigned thereto; and,
an output port in communication with the processor for providing data indicative of the texture map.
20. A system for segmenting texture as defined in claim 19 wherein the processor comprises electronic circuitry designed for performing at least a portion of transforming the signal data into second signal data and processing the second signal data.
21. A system for segmenting texture as defined in claim 19 comprising a control port in communication with the processor for receiving control commands for controlling at least one of determining a region of interest, generation of texture data, and classification.
22. A system for segmenting texture as defined in claim 21 comprising a graphical display in communication with the processor for displaying at least the data indicative of the texture map in a graphical fashion.
23. A system for segmenting texture as defined in claim 22 wherein the graphical display comprises a graphical user interface.
24. A method for segmenting texture of multi-dimensional data indicative of a characteristic of an object comprising:
receiving the multi-dimensional data;
transforming the multi-dimensional data into second multi-dimensional data within a space-frequency domain using a rotation-invariant localized space-frequency transformation of the multi-dimensional data;
applying principal component analysis to the second multi-dimensional data for generating texture data characterizing texture around data points of at least a portion of the multi-dimensional data; and,
partitioning the data points of the at least a portion of the multi-dimensional data into clusters based on the texture data using a classification process.