Method of chromatic classification of pixels and method of adaptive enhancement of a color image

• US 20080144946A1
• Filed: 12/17/2007
• Published: 06/19/2008
• Est. Priority Date: 12/19/2006
• Status: Active Grant

• Current Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Original Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Inventors: Naccari, Filippo, Castorina, Alfio, Bruna, Arcangelo Ranieri
• Granted Patent: US 8,374,425 B2
• Time in Patent Office: Days
• US Class Current: 382/225
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06V 10/28 Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
• CPC Class Codes: G06V 10/56 relating to colour
• CPC Class Codes: H04N 1/62 Retouching, i.e. modification of isolated colours only or in isolated picture areas only
• CPC Class Codes: H04N 1/628 Memory colours, e.g. skin or sky

Abstract

A method of processing digital images by transforming a set of pixels from a three-dimensional space to a normalized two-dimensional space, determining a membership class and membership class level of each pixel in the set of pixels, and selectively modifying colors of pixels in the set of pixels based on the determined membership classes and membership class levels.

Claims

1. A method of chromatic classification of pixels of a digital image comprising the steps of:

transforming a chromatic domain of definition of pixels from a three-dimensional space into a respective two-dimensional normalized space;

defining a reference model of classification by analyzing a plurality of images of distinct types of subjects, generating distinct collections of samples with respective coordinates in the normalized two-dimensional space for each of a number of chromatic classes corresponding to real subjects; and

evaluating for each image pixel to be classified a membership class and a membership level based on the reference model;

wherein said reference model is defined by a parametric membership function continuous on said normalized two-dimensional space for each of said classes, parameters of which are determined by processing the collection of two-dimensional samples of each of said classes.

2. The method of claim 1 wherein defining the reference model comprises calculating a mean position of clusters of samples on the normalized two-dimensional space, a variance in respect to a principal direction and to a direction orthogonal to it and a slope angle of said principal direction using a principal component analysis technique, thus obtaining an elliptical Gaussian model for each of said classes on said normalized two-dimensional space.

3. The method of claim 1 wherein said three-dimensional space is an RGB space and the respective normalized two-dimensional space is an rg space.

4. The method of claim 1 wherein for each class a relative membership value is obtained for a current pixel being processed, a maximum membership value being established in function of the respective membership class, and the membership level being determined in function of a maximum quantized value with a number of bits.

5. The method of claim 1, further comprising storing in an array values of said reference model in correspondence of pairs of values in the normalized two-dimensional space, and of determining the membership level of each pixel by reading from the array a value corresponding to the pair that represents said pixel in the normalized two-dimensional space.

6. The method of claim 2, comprising steps of:

verifying whether levels of the membership functions in correspondence of a neutral axis are smaller than a threshold; and

reducing proportionally the variances of membership functions for which the verification of the previous step is not satisfied, such that a respective value of the membership level of said neutral axis is smaller than said threshold.

7. The method of claim 6 wherein said threshold is 2%.

8. The method of claim 5 wherein the stored values in said two-dimensional array are obtained through steps of:

on clusters of samples of said images of distinct types of subjects, calculating a mean position, a variance in respect to a principal direction, a variance in respect to a direction orthogonal to the principal direction and a slope angle of the principal direction;

generating values according to an elliptical Gaussian model in the normalized two-dimensional space; and

undersampling and quantizing the values obtained at a previous point and storing them in the two-dimensional array.

9. The method of claim 1, further comprising:

generating a membership image of pixels of the digital image with the chromatic classes of the reference model;

defining a table of chromaticity target values for each chromatic class;

converting definition values of each pixel of an image to be processed from the chromatic definition domain to the respective normalized two-dimensional space;

calculating for each cluster of pixels belonging to a chromatic class a pair of statistic parameters;

calculating for each cluster a distance from a respective chromaticity target value;

determining for each pixel of the image belonging to a respective chromatic class, respective values in the chromatic definition domain corrected in function of coordinates of original values in the respective normalized two-dimensional space and of the distance of the relative cluster from a relative target color; and

carrying out a modulated correction of a color of the pixel using a weighted linear combination of the values of the pixel and of said triplet of target values.

10. The method of claim 9 wherein said weighted linear combination is carried out using weights determined in function of the membership levels such that an intensity of pixels of the enhanced image is equal to an intensity of the corresponding pixel of the image to be enhanced.

11. The method of claim 1 wherein the subject is an object.

12. A system, comprising:

means for transforming a chromatic domain representation of a pixel from a three-dimensional space to a two-dimensional normalized space;

means for determining a membership class of the pixel in a set of membership classes and a corresponding membership class level of the pixel in the determined membership class based on the transformed representation of the pixel; and

means for selectively modifying a color of the pixel based on the determined membership class and membership class level of the pixel.

13. The system of claim 12 wherein the means for determining is configured to determine a membership class level for the pixel in each class in the set of membership classes and the determined membership class is the membership class in the set of membership classes with a highest determined membership class level.

14. The system of claim 12 wherein each class in the set of membership classes is based on a parametric class-membership function that is continuous on the two-dimensional normalized space.

15. The system of claim 12 wherein the means for determining comprises a look-up table.

16. The system of claim 12 wherein the means for determining is configured to apply a Gaussian function to the transformed representation of the pixel.

17. The system of claim 12, further comprising an image pre-processor.

18. A chromatic classification device, comprising:

a domain transform module configured to transform pixels in a digital image from a three-dimensional space to a two-dimensional normalized space; and

a classification module coupled to the domain transform module and configured to output an indication of membership classes and membership class levels associated with respective pixels in the digital image.

19. The chromatic classification device of claim 18 wherein the classification module comprises a Gaussian function generator.

20. The chromatic classification device of claim 18 wherein the classification module is configured to generate the indication using a look-up table.

21. The chromatic classification device of claim 18 wherein the classification module is configured to output a first signal indicative of a membership class associated with respective pixels and a second signal indicative of a membership class level associated with respective pixels and the indication comprises the first signal and the second signal.

22. A computer-readable medium comprising program code that causes a computing device to implement a method that includes:

transforming a chromatic domain representation of a pixel from a three-dimensional space to a two-dimensional normalized space;

determining a membership class of the pixel in a set of membership classes and a corresponding membership class level of the pixel in the determined membership class based on the transformed representation of the pixel; and

selectively modifying a color of the pixel based on the determined membership class and membership class level of the pixel.

23. The computer-readable memory medium of claim 22 wherein the program code causes the processor to determine a membership class level for the pixel in each class in the set of membership classes and to determine the determined membership class by selecting a membership class in the set of membership classes with a highest determined membership class level.

24. A method of processing pixels in a digital image, the method comprising:

generating a membership image of pixels of the digital image based on chromatic classes of a reference model and membership levels of pixels of the digital image in the chromatic classes of the reference model;

defining a table of chromaticity target values for each chromatic class;

calculating for each cluster of pixels belonging to a chromatic class a pair of statistic parameters;

calculating for each cluster a distance from a respective chromaticity target value;

determining for each pixel of the image belonging to a respective chromatic class, respective values in the chromatic definition domain corrected in function of coordinates of original values in the respective normalized two-dimensional space and of the distance of the relative cluster from a relative target color; and

carrying out a modulated correction of a color of the pixel using a weighted linear combination of the values of the pixel and of said triplet of target values.

25. The method of claim 24 wherein said weighted linear combination is carried out using weights determined in function of the membership levels such that an intensity of pixels of the enhanced image is equal to an intensity of the corresponding pixel of the image to be enhanced.