System and method for illumination invariant image segmentation

1. An automated, computerized method for processing an image, comprising the steps of:

providing an image file depicting the image;

identifying a boundary in the image;

calculating a representation of the boundary extending to segments of the image at either side of the boundary;

performing feature calculations on the representation, the feature calculations comprising performance of a series of spatio-spectral tests on pixels in the representation, to generate data indicating a likelihood as to whether the boundary is a material boundary, wherein the series of spatio-spectral tests are selected from a group of tests for measuring characteristics of the boundary including a fine similarity of illumination change along the boundary, a gross similarity of illumination change, local continuity in illumination conditions, a gross match analysis of a BIDR model prediction, overall qualities of a bi-illuminant, a degree of variation of color values, an indication of the quantity of information relevant to the nature of the boundary change, a spatial quality of changes in appearance, linearity predicted by the BIDR model and a combination of the measured characteristics from the group; and

classifying the boundary as caused by a material change, as a function of the feature calculations.

2. The method of claim 1 wherein the step of identifying a boundary in the image is carried out by identifying spatio-spectral information for the image, utilizing the spatio-spectral information to identify single material token regions, and utilizing the single material token regions to identify a boundary in the image.

3. The method of claim 1 wherein the step of calculating a representation of the boundary is carried out by generating a pad representation of the boundary.

4. The method of claim 3 wherein the pad representation comprises an arrangement of pads distributed along either side of the boundary, and each one of the pads comprises an array of contiguous pixels.

5. The method of claim 1 wherein the step of calculating a representation of the boundary is carried out by generating a strip token representation of the boundary.

6. The method of claim 1 wherein the step of classifying the boundary as caused by a material change, as a function of the feature calculations, is carried out by performing a computer learning technique.

7. The method of claim 6 wherein the computer learning technique comprises a technique selected from the group consisting of a neural network, a support vector machine, a Bayesian decision tree and boosting.

8. An automated, computerized method for processing an image, comprising the steps of:

providing an image file including pixels forming the image;

identifying a boundary in the image;

selecting representative pixels of the image relevant to the boundary;

performing a feature calculation on the representative pixels, the feature calculation comprising performance of a series of spatio-spectral tests on the pixels, to generate data indicating a likelihood as to whether the boundary is a material boundary, wherein the series of spatio-spectral tests are selected from a group of tests for measuring characteristics of the boundary including a fine similarity of illumination change along the boundary, a gross similarity of illumination change, local continuity in illumination conditions, a gross match analysis of a BIDR model prediction, overall qualities of a bi-illuminant, a degree of variation of color values, an indication of the quantity of information relevant to the nature of the boundary change, a spatial quality of changes in appearance, linearity predicted by the BIDR model and a combination of the measured characteristics from the group; and

classifying the boundary as caused by a material change, as a function of the feature calculation.

9. The method of claim 8 wherein the step of identifying a boundary in the image is carried out by identifying spatio-spectral information for the image, utilizing the spatio-spectral information to identify single material token regions, and utilizing the single material token regions to identify a boundary in the image.

10. The method of claim 8 wherein the step of selecting representative pixels of the image relevant to the boundary is carried out by generating a pad representation of the boundary.

11. The method of claim 10 wherein the pad representation comprises an arrangement of pads distributed along either side of the boundary, and each one of the pads comprises an array of contiguous pixels.

12. The method of claim 8 wherein the step of selecting representative pixels of the image relevant to the boundary is carried out by generating a strip token representation of the boundary.

13. The method of claim 8 wherein the group of tests includes a log of gradient match, dynamic time warping, a log space matching, a multi-scale ratio matching, a pad linearity test, a reasonable direct/ambient illuminant pair test, a degree of variation test, a sharpness of strip token test, a linearity of strip token test and an image change test.

14. The method of claim 8 wherein the step of classifying the boundary as caused by a material change, as a function of a feature calculation is carried out by performing a computer learning technique.

15. The method of claim 14 wherein the computer learning technique comprises a technique selected from the group consisting of a neural network, a support vector machine, a Bayesian decision tree and boosting.

16. A computer system which comprises: a CPU; and a memory storing an image file containing an image; the CPU arranged and configured to execute a routine to identify a boundary in the image, calculate a representation of the boundary extending to segments of the image at either side of the boundary, perform feature calculations on the representation, the feature calculations comprising performance of a series of spatio-spectral tests on pixels in the representation, to generate data indicating a likelihood as to whether the boundary is a material boundary, wherein the series of spatio-spectral tests are selected from a group of tests for measuring characteristics of the boundary including a fine similarity of illumination change along the boundary, a gross similarity of illumination change, local continuity in illumination conditions, a gross match analysis of a BIDR model prediction, overall qualities of a bi-illuminant, a degree of variation of color values, an indication of the quantity of information relevant to the nature of the boundary change, a spatial quality of changes in appearance, linearity predicted by the BIDR model and a combination of the measured characteristics from the group, and classify the boundary as caused by a material change, as a function of the feature calculations.

17. A computer system which comprises:

a CPU; and a memory storing an image file including pixels forming an image;

the CPU arranged and configured to execute a routine to identify a boundary in the image, select representative pixels of the image relevant to the boundary, perform a feature calculation on the representative pixels, the feature calculation comprising performance of a series of spatio-spectral

tests on the pixels, to generate data indicating a likelihood as to whether the boundary is a material boundary, wherein the series of spatio-spectral tests are selected from a group of tests for measuring characteristics of the boundary including a fine similarity of illumination change along the boundary, a gross similarity of illumination change, local continuity in illumination conditions, a gross match analysis of a BIDR model prediction, overall qualities of a bi-illuminant, a degree of variation of color values, an indication of the quantity of information relevant to the nature of the boundary change, a spatial quality of changes in appearance, linearity predicted by the BIDR model and a combination of the measured characteristics from the group, and classify the boundary as caused by a material change, as a function of the feature calculation.