Auto-focusing techniques based on statistical blur estimation and associated systems and methods
View Patent ↗This is generally directed to auto-focusing techniques based on statistical blur estimation. An image can be captured at two or more candidate lens positions. The amount of blur of each image can then be determined, and the image containing the least amount of blur can be chosen as the “in-focus” image. In some embodiments, the amount of blur of an image can be determined by identifying how “Gaussian” an image is. Characteristics that are more Gaussian in nature can indicate that the image is more blurry. The Gaussianity of an image can be determined by estimating a generalized Gaussian shape parameter for that image. A smaller shape parameter can indicate the image is less Gaussian in nature. The shape parameter can be estimated in any suitable manner such as, for example, through a 2-d discrete wavelet transform, through a 1-d discrete wavelet transform, or through any other suitable manner.
1. A method of auto-focusing an imaging system, the method comprising:
capturing a plurality of images, wherein each image of the plurality of images is captured at a different candidate lens position of the imaging system;
estimating a Gaussianity value of each image of the plurality of images, wherein the Gaussianity value is estimated based on observed statistics of at least one high frequency subband of each image and a generalized Gaussian ratio function;
identifying a particular image of the plurality of images that comprises a smallest Gaussianity value; and
moving a lens of the imaging system to a particular candidate lens position, wherein the particular image was captured at the particular candidate lens position.
2. The method of claim 1 , wherein the estimating the Gaussianity value comprises:
applying a first stage of a discrete wavelet transform (“DWT”) to each image of the plurality of images, wherein the application of the first stage of the DWT produces the at least one high frequency subband for each image; and
calculating, for each image, the observed statistics of the at least one high frequency subband for each image.
3. The method of claim 2 , wherein the DWT comprises one of a Cohen-Daubechies-Feauveau 9/7 wavelet kernel and a spline 5/3 wavelet kernel.
4. The method of claim 2 , wherein applying a first stage of a DWT to each image of the plurality of image comprises:
applying the first stage of the DWT to a Region of Interest of each image of the plurality of images.
5. The method of claim 1 , wherein determining a Gaussianity value comprises:
determining a shape parameter value for each image of the plurality of images.
6. The method of claim 5 , wherein identifying the particular image of the plurality of images comprises:
identifying the particular image of the plurality of images that comprises a shape parameter value farthest from a value of two.
7. The method of claim 5 , wherein identifying the particular image of the plurality of images comprises:
identifying the particular image of the plurality of images that comprises a shape parameter value that is least in value.
8. The method of claim 5 , wherein determining a shape parameter value comprises:
estimating the shape parameter value by an n th order Method of Moments, wherein n corresponds to an order of a moment.
9. An auto-focusing system, the system comprising:
a lens operable to capture an image, wherein the image comprises a plurality of pixels grouped into a plurality of pixel sets;
a lens driver operable to move the lens to a plurality of lens positions, wherein the image is captured at a certain lens position of the plurality of lens positions; and
a processor operable to:
sequentially process each pixel set by applying, on each luma pixel location of the plurality of pixel sets, a one-dimensional discrete wavelet transform (“1-d DWT”) to produce a luma component coefficient for each luma pixel location;
sequentially process each pixel set by performing, on each green pixel location of a raw Bayer array of the image, the 1-d DWT on each diagonal to produce a first diagonal coefficient and a second diagonal coefficient for each green pixel location;
determine a shape parameter value for each pixel set based on each pixel set's generated luma component coefficients, first diagonal coefficients, and second diagonal coefficients; and
determine a final shape parameter value for the image based on the shape parameter values of each pixel set; and wherein:
the auto-focusing system is operable to similarly generate a final shape parameter value for each of a plurality of images, wherein each image of the plurality of images is captured at a different lens position.
10. The system of claim 9 , wherein the processor is further operable to:
identify a particular image of the plurality of images that is associated with a smallest final shape parameter value; and
direct the lens driver to move the lens to a particular lens position at which the particular image was captured.
11. The system of claim 9 , wherein the processor is further operable to:
update, at each luma pixel location of a pixel set, a sum of luma component coefficients, a sum of luma component coefficients squared, a sum of an absolute value of luma component coefficients; and
determine, at the end of each pixel set, a variance and a first absolute moment for each pixel set based on that pixel set's updated sum of luma component coefficients, updated sum of luma component coefficients squared, and updated sum of the absolute value of luma component coefficients.
12. The system of claim 11 , wherein the processor is further operable to:
determine a Ratio Function value for each pixel set based on each pixel set's variance and first absolute moment; and
determine the shape parameter value for each pixel set based on each pixel set's Ratio Function value.
13. The system of claim 12 , further comprising:
a memory operable to store a lookup table, wherein the lookup table comprises a plurality of Ratio Function values and a plurality of corresponding shape parameter values, and wherein:
the processor is further operable to determine the shape parameter value for each pixel set based on each pixel set's Ratio Function value and the lookup table.
14. The system of claim 13 , wherein the plurality of Ratio Function values are scaled by a first scaling factor and the plurality of shape parameter values are scaled by a second scaling factor, and wherein the first and second scaling factors convert the plurality of Ratio Function values and the plurality of shape parameter values into integer forms.
15. The system of claim 14 , wherein the first and second scaling factors are each a power of two.
16. The system of claim 12 , further comprising:
a memory operable to store a lookup table, wherein the lookup table comprises a plurality of knee points of a Ratio Function curve, and wherein:
the processor is further operable to determine, for each pixel set, the shape parameter value by:
identifying at least two knee points of the plurality of knee points based on the Ratio Function value of that pixel set; and
linearly interpolating the shape parameter value from that pixel set's identified at least two knee points.
17. The system of claim 16 , wherein the knee points are determined by calculating a first derivative of the Ratio Function curve.
18. The system of claim 12 , further comprising:
a memory operable to store coefficients of a parametric function, wherein the parametric function approximates a Ratio Function curve; and wherein:
the processor is operable to determine the shape parameter value for each pixel set based on each pixel set's Ratio Function value and based on the coefficients of the parametric function.
19. An auto-focusing system, the system comprising:
a lens operable to capture an image, wherein the image comprises a plurality of pixels grouped into a plurality of pixel sets;
a lens driver operable to move the lens to a plurality of lens positions, wherein the image is captured at a certain lens position of the plurality of lens positions;
a high-pass filter operable to sequentially process each pixel set by filtering each luma pixel location of the plurality of pixel sets, wherein the filtering generates a luma component coefficient for each luma pixel location; and
a processor operable to:
determine a variance and first absolute moment for each pixel set of the plurality of pixel sets based on each pixel set's generated luma component coefficients;
determine a Ratio Function value for each pixel set based on each pixel set's variance and first absolute moment;
determine a shape parameter value for each pixel set based on each pixel set's Ratio Function value; and
determine a final shape parameter value for the image based on the shape parameter value of each pixel set; and wherein:
the auto-focusing system is operable to similarly generate a plurality of final shape parameter values for a plurality of images, wherein each image of the plurality of images is captured at a different lens position.
20. The system of claim 19 , wherein the processor is further operable to:
identify a particular image of the plurality of images that is associated with a particular final shape parameter value of the plurality of final shape parameter values farthest from a value of two; and
direct the lens driver to move the lens to a particular lens position at which the particular image was captured.
21. The system of claim 19 , wherein the high-pass filter comprises at least three taps.
22. The system of claim 19 , further comprising:
a memory operable to store a lookup table, wherein the lookup table comprises a plurality of Ratio Function values and a plurality of corresponding shape parameter values, and wherein:
the processor is further operable to determine the shape parameter value for each pixel set based on each pixel set's Ratio Function value and the lookup table.
23. The system of claim 22 , wherein the plurality of Ratio Function values are scaled by a first scaling factor and the plurality of shape parameter values are scaled by a second scaling factor, and wherein the first and second scaling factors convert the plurality of Ratio Function values and the plurality of shape parameter values into integer forms.
24. The system of claim 23 , wherein the first and second scaling factors are each a power of two.
25. The system of claim 19 , further comprising:
a memory operable to store a lookup table, wherein the lookup table comprises a plurality of knee points of a Ratio Function curve, and wherein:
the processor is further operable to determine, for each pixel set, the shape parameter value by:
identifying at least two knee points of the plurality of knee points based on the Ratio Function value of that pixel set; and
linearly interpolating the shape parameter value from that pixel set's identified at least two knee points.
26. The system of claim 25 , wherein the knee points are determined by calculating a first derivative of the Ratio Function curve.
27. The system of claim 19 , further comprising:
a memory operable to store coefficients of a parametric function, wherein the parametric function approximates a Ratio Function curve; and wherein:
the processor is operable to determine the shape parameter value for each pixel set based on each pixel set's Ratio Function value and based on the coefficients of the parametric function.