Convex clustering for chromatic content modeling
View Patent ↗A system and method are provided for modeling a chromatic object, such as an image. For a set of colors of a chromatic object that are expressed as color values in a perceptual color space, the method includes optimizing a convex objective function which is a log likelihood function of a combination of weighted kernels centered on each color in the set over each of the other colors in the set. A number N c of weighted kernels in the optimized function which each have a weight which is at least greater than 0 is identified. The chromatic object is modeled with a mixture model in which the complexity of the model is based on the identified number N c .
1. A method for modeling a chromatic object comprising:
for a set of colors of a chromatic object that are expressed as color values in a perceptual color space, optimizing a convex objective function which is a log likelihood function of a combination of weighted kernels each kernel being centered on a respective one of the colors in the set and being based on distances to each of the other colors in the set;
identifying a number of weighted kernels in the optimized function which each have a weight which is at least greater than 0; and
modeling the chromatic object with a mixture model in which the complexity of the model is based on the identified number.
2. The method of claim 1 , wherein the mixture model comprises a Gaussian Mixture model and wherein the identified number corresponds to the number of Gaussians in the mixture model and the mean of each Gaussian function corresponds to a color.
3. The method of claim 1 , wherein the chromatic object comprises a photographic image.
4. The method of claim 1 , wherein the perceptual color space comprises a space in which one of its dimensions is a luminance dimension.
5. The method of claim 4 , wherein the colors are expressed in terms of the non-luminance dimensions.
6. The method of claim 4 , wherein the perceptual color space is a Lab color space.
7. The method of claim 1 , wherein the kernels are Mercer kernels.
8. The method of claim 1 , wherein each kernel is a function of the Euclidian distance between the color on which the kernel is centered and each of the other colors.
9. The method claim 8 , wherein the kernels are Gaussian functions.
10. The method of claim 1 , wherein the optimization is performed with an expectation maximization algorithm or pairwise optimization.
11. The method of claim 1 , wherein the optimization comprises pruning weighted kernels from the convex objective function which fail to satisfy a threshold condition.
12. The method of claim 11 , wherein the pruning comprises setting a weight of one of the kernels to 0 for which there exists at least one other color m in the set of colors for which γ nn ≦γ nm where γ nn is the estimated assignment of the color on which the kernel is centered to that color and γ nm is the estimated assignment of the color on which the kernel is centered to the other color of the set of colors.
13. The method of claim 1 , wherein the convex objective function is a sum of log likelihood values, wherein each log likelihood value is a convex combination of the weighted kernels.
14. The method of claim 1 , wherein the convex objective function is expressed as:
L=Σ m=1 N log(Σ n=1 N w n k n ( x m )) (EQ1)
where N represents a number of the colors in the set;
x m represents a color in the set on which a kernel k n is centered; and
w n represents a weight associated with kernel k n .
15. The method of claim 14 , wherein in the convex objective function the weights w n are subject to the following set of constraints:
0 ≦w n ≦1 (EQ2);
and
Σ n=1 N w n =1 (EQ3).
16. The method of claim 1 , wherein the modeling of the chromatic object with a mixture model in which the complexity of the model is based on the identified number comprises generating a mixture model comprising N c weighted functions, where N c is the number of weighted kernels in the optimized objective function.
17. The method of claim 1 , wherein the identified number is less than a number of the colors in the set of colors.
18. The method of claim 1 , further comprising outputting at least one of the mixture model and information derived therefrom.
19. The method of claim 1 , further comprising transforming a set of colors of a chromatic object that are not expressed as color values in a perceptual color space to the set of colors of the chromatic object that are expressed as color values in the perceptual color space.
20. The method of claim 1 , wherein at least one of the optimizing, identifying, and modeling is performed with a computer processor.
21. A method of color transfer comprising:
modeling a first chromatic object with colors of a mixture model generated according to the method of claim 1 ;
generating pairs of mapped colors, each pair comprising a respective one of a set of colors forming the mixture model of the first chromatic object and a respective one of a set of colors in a mixture model of a second chromatic object;
for each pair of the mapped colors, computing a linear transform between the colors;
assigning colors of the first chromatic object to the set of colors forming the mixture model of the first chromatic object; and
modifying colors of the first chromatic object based on the assignments and the linear transforms.
22. The method of color transfer of claim 21 , further comprising printing the first chromatic object with the modified colors.
23. A computer program product comprising a non-transitory recording medium encoding instructions, which when executed on a computer, cause the computer to perform the method of claim 1 .
24. A system comprising non-transitory memory which stores instructions for performing the method of claim 1 and a processor, in communication with the memory, for executing the instructions.
25. A system for modeling a chromatic object comprising:
a color space transformation component which for a set of colors of a chromatic object that are not expressed as color values in a perceptual color space, transforms the set of colors to color values in a perceptual color space;
a convex clustering component which applies a convex objective function which is a log likelihood function of a combination of weighted kernels, each kernel being centered on a respective one of the colors in the set and being based on distances to each of the other colors in the set, to identify a number of weighted kernels in the optimized function which have a weight which is at least greater than 0; and
a modeling component which generates a chromatic model of the chromatic object using the identified number as its complexity.
26. The system of claim 25 , further comprising a color transfer component for transferring color from a second chromatic object, based on the chromatic model of the first chromatic object and a chromatic model of the second chromatic object.
27. A method for modeling an image comprising:
for a representative set of colors of the image, expressing each of the colors as color values in a perceptual color space;
with an algorithm optimizing, through a plurality of iterations, a convex objective function which has a single optimum value, the optimization of the convex optimizing function outputting a set of N c non-zero weighted kernel functions; and
modeling the chromatic object with a mixture model comprising exactly N c weighted functions.