Method and system for automatic white balancing
View Patent ↗An automatic white balancing method and system. A modification of “color in perspective” white balancing technique is implemented. Rather than a default projection plane used by the typical white balancing technique, the projection plane for a RGB space is automatically selected to allows the projected reference gamuts in the projection plane to be as disjoined as possible, thereby making them as uncorrelated as possible. In turn, automatic white balancing is performed using the color temperature associated with the projected reference gamut most correlated with the projected gamut of an image sensor.
1. A method for white balancing an image captured by an image sensor, said method comprising the steps of:
a) forming n reference clusters of points in a 3-dimensional color space, wherein said n reference clusters represent said image sensor's response respectively to n color temperatures;
b) finding within said color space a projection plane on which to project said n reference clusters to form respectively n projected reference clusters that are approximately maximally disjoined;
c) forming a projected target cluster by projecting onto said projection plane a cluster points in said color space, said target cluster representing said image sensor's response to said captured image; and
d) white balancing said captured image with the color temperature associated with one of said n projected clusters that among said n projected clusters has the highest correlation with said projected target cluster.
2. The method of claim 1 , wherein said color space is a RGB space.
3. The method of claim 1 , wherein each of said n reference clusters comprises m subsets of points, wherein a subset of points for each of said n reference cluster in said color space represents the response of said image sensor to a color temperature reflected from one of m surfaces.
4. The method of claim 1 , wherein said step b) comprising the steps of:
b1) generating in said color space a first vector that is the difference between a vector m(L) and a vector m(M), wherein said m(L) is the mean of one of said n reference clusters having low-range color temperature, and said m(M) is the mean of one of said n reference clusters having mid-range color temperature;
b2) generating in said color space a second vector that is the difference between said vector m(M) and a vector m(H), wherein said m(H) is the mean of one of said n reference clusters having high-range color temperature; and
b3) forming said projection plane as the span of said first vector and said second vector.
5. The method of claim 1 , wherein said step b) comprising the steps of:
b1) given n means respectively of said n reference clusters, obtaining a plane X that contain as many of said n means as possible by using regression analysis; and
b2) forming said projection plane by fine-tuning orientation of said plane X to allow said n projected reference clusters to be as uncorrelated as possible.
6. A method for white balancing an image captured by an image sensor, said method comprising the steps of:
a) for j=1 to n, forming a reference cluster j of points in a 3-dimensional color space by recording for k=1 to m a subset of points (j,k) in said 3-dimensional color space;
b) forming n projected reference clusters of points by projecting said n reference clusters onto a projection plane whose orientation in said color space is adapted to minimize correlation among said n projected reference clusters; and
c) storing said n projected reference clusters in a table adapted to be referenced by said image sensor in white balancing said capture image using the color temperature associated with one of said n projected reference clusters that among said n projected reference clusters has the highest correlation with a projected target cluster on said projection plane, said projected target cluster resulted from projecting onto said projection plane a color space representation of said image sensor's response to said captured image.
7. The method of claim 6 , wherein said color space is a RGB space.
8. The method of claim 6 , wherein said reference cluster j comprises m subsets of points (j,1), (j,2), . . . (j,n), wherein said subset of points (j,k) for said reference cluster j in said color space represents a response (j,k) of said image sensor to a reflection of a color temperature j reflecting from a surface k.
9. The method of claim 6 , wherein said step b) comprising the steps of:
b1) generating in said color space a first vector that is the difference between a vector m(L) and a vector m(M), wherein said m(L) is the mean of one of said n reference clusters having low-range color temperature, and said m(M) is the mean of one of said n reference clusters having mid-range color temperature;
b2) generating in said color space a second vector that is the difference between said vector m(M) and a vector m(H), wherein said m(H) is the mean of one of said n reference clusters having high-range color temperature; and
b3) forming said projection plane as the span of said first vector and said second vector.
10. The method of claim 6 , wherein said step b) comprising the steps of:
b1) given n means respectively of said n reference clusters, obtaining a plane X that contain as many of said n means as possible by using regression analysis; and
b2) forming said projection plane by fine-tuning orientation of said plane X to allow said n projected reference clusters to be as uncorrelated as possible.
11. A method for image white balancing, said method comprising the steps of:
in response to receiving an image, forming a cluster in a 3-dimensional color space by plotting the pixel values of said image in said color space;
forming a projected cluster by projecting said cluster onto a projection plane, said projection plane determined using mathematical method on n mean cluster points of respectively n 3-dimensional reference clusters in said color space, wherein on said projection plane n 2-dimensional projected clusters respectively of said n 3D reference clusters are maximally uncorrelated;
determining correlation between said projected cluster with each of n projected reference clusters; and
white balancing said image using the color temperature associated a projected reference cluster most correlated with said projected cluster.
12. The method of claim 11 , wherein said color space is a RGB space.
13. The method of claim 11 , wherein each of said n reference clusters comprises m subsets of points, wherein a subset of points for each of said n reference cluster in said color space represents the response of said image sensor to a color temperature reflected from one of m surfaces.
14. The method of claim 11 , wherein said step b) comprising the steps of:
b1) generating in said color space a first vector that is the difference between a vector m(L) and a vector m(M), wherein said m(L) is the mean of one of said n reference clusters having low-range color temperature, and said m(M) is the mean of one of said n reference clusters having mid-range color temperature;
b2) generating in said color space a second vector that is the difference between said vector m(M) and a vector m(H), wherein said m(H) is the mean of one of said n reference clusters having high-range color temperature; and
b3) forming said projection plane as the span of said first vector and said second vector.
15. The method of claim 11 , wherein said step b) comprising the steps of:
b1) given n means respectively of said n reference clusters, obtaining a plane X that contain as many of said n means as possible by using regression analysis; and
b2) forming said projection plane by fine-tuning orientation of said plane X to allow said n projected reference clusters to be as uncorrelated as possible.
16. An image sensor chip comprising:
a memory storing a lookup table that associates n color temperatures to n projected reference clusters on a projection plane within a 3-dimensional color space, said n projected reference clusters being projections respectively of n reference clusters in said color space wherein each reference cluster represents the response of said image sensor chip to a color temperature reflected from m surfaces, said projection plane having an orientation that minimizes correlation among said n projected reference clusters; and
a sensor array coupled to said lookup table, wherein the representation of a captured image in said color space is projected onto said projection plane to result in a projected sensor response that is correlated with said n projected reference clusters stored in said lookup table, wherein said capture image is white balanced using the color temperature of the projected reference cluster that among said n projected reference clusters is most correlated with said projected sensor response.
17. The image sensor chip of claim 16 , wherein said color space is a RGB space.
18. The image sensor chip of claim 16 , wherein each of said n reference clusters comprises m subsets of points, wherein a subset of points for each of said n reference cluster in said color space represents the response of said image sensor to a color temperature reflected from one of m surfaces.
19. The image sensor chip of claim 16 , wherein said projection plane is obtained using the steps comprising:
a) generating in said color space a first vector that is the difference between a vector m(L) and a vector m(M), wherein said m(L) is the mean of one of said n reference clusters having low-range color temperature, and said m(M) is the mean of one of said n reference clusters having mid-range color temperature;
b) generating in said color space a second vector that is the difference between said vector m(M) and a vector m(H), wherein said m(H) is the mean of one of said n reference clusters having high-range color temperature; and
c) forming said projection plane as the span of said first vector and said second vector.
20. The image sensor chip of claim 16 , wherein said projection plane is obtained using the steps comprising:
a) given n means respectively of said n reference clusters, obtaining a plane X that contain as many of said n means as possible by using regression analysis; and
b) forming said projection plane by fine-tuning orientation of said plane X to allow said n projected reference clusters to be as uncorrelated as possible.