Systems and methods for synthesizing high resolution images using images captured by an array of independently controllable imagers

1. A non-transitory machine readable medium containing processor instructions, where execution of the instructions by a processor causes the processor to perform a process that comprises:

acquiring a plurality of low resolution (LR) images from different viewpoints, where the plurality of LR images capture a scene in which depths of points in the scene vary and each of the plurality of LR images differs from other LR images due to scene dependent geometric displacements due to parallax;

determining scene dependent parallax information with respect to the LR images based upon disparity relative to a reference point of view resulting from the depths of points in the captured scene; and

fusing at least portions of the LR images to form a high resolution (HR) image by:

populating a high resolution (HR) grid corresponding to pixel locations of the HR image with LR image samples from the LR images using the scene dependent parallax information;

interpolating samples at grid position of the HR grid using neighboring available samples from the LR image samples placed on the HR quid; and

instructing a display device to display the HR image.

2. The non-transitory machine readable medium of claim 1 , wherein the process further comprises identifying and removing outlier LR image samples by identifying pixels that possess low levels of correspondence with neighboring pixels.

3. The non-transitory machine readable medium of claim 1 , wherein the process further comprises using a nearest neighbor process based upon the placed LR samples to determine pixel values on the HR grid.

4. The non-transitory machine readable medium of claim 3 , wherein the nearest neighbor process comprises locally, and adaptively filtering available samples in a neighborhood of each grid position of the HR grid to determine a value of an image sample at that position.

5. The non-transitory machine readable medium of claim 1 , wherein the process further comprises performing a filtering process to remove outlier LR image samples and interpolate values at each HR grid position.

6. The non-transitory machine readable medium of claim 5 , wherein the filtering process is implemented using a locally adaptive interpolation process.

7. The non-transitory machine readable medium of claim 1 , wherein the process further comprises identifying outliers by constructing a measure based upon filter weights calculated using an LR image sample and its list of neighbors on the HR grid.

8. The non-transitory machine readable medium of claim 7 , wherein the filter is an edge preserving filter.

9. The non-transitory machine readable medium of claim 8 , wherein the edge preserving filter is configured to filter LR image samples based upon both range and domain.

10. The non-transitory machine readable medium of claim 9 , wherein the filter applies a Gaussian smoothing that comprises weighting filter coefficients based upon corresponding relative pixel intensities.

11. The non-transitory machine readable medium of claim 10 , wherein pixels that are different photometrically from a central pixel are weighted less.

12. The non-transitory machine readable medium of claim 1 , wherein the process further comprises obtaining a measure of a consistency of a given LR image sample with its neighbors utilizing filter weights that are computed based upon the given LR image sample and its neighboring samples on the HR grid.

13. The non-transitory machine readable medium of claim 12 , wherein the weights are summed and the sum normalized by a number of neighboring LR image samples to produce a value indicative of a similarity between a given LR image sample point and its neighbors.

14. The non-transitory machine readable medium of claim 13 , wherein the process further comprises determining that the given LR image sample is an outlier when the normalized sum of the weights is below a threshold.

15. The non-transitory machine readable medium of claim 14 , wherein the filter weights utilize a range kernel that is a function of a standard deviation of a noise estimate at a position and a spatial kernel that is a predetermined Gaussian that decays with distance.

16. The non-transitory machine readable medium of claim 1 , wherein the process further comprises locally, and adaptively filtering available samples in a neighborhood of each grid position of the HR grid to determine a value of an HR image sample at that position of the HR grid.

17. The non-transitory machine readable medium of claim 1 , wherein acquiring the plurality of low resolution (LR) images from different viewpoints comprises using a camera array comprising a first camera with a first field of view and a first magnification and a second camera with a different second field of view and a different second magnification.

18. The non-transitory machine readable medium of claim 17 , wherein the camera array comprises a third camera with a third field of view and a third magnification, wherein the first field of view is at a maximum field of view, the second field of view is at a first fraction of the maximum field of view and the third field of view is at a different second fraction of the maximum field of view.

19. A method of synthesizing an image, the method comprising:

acquiring a plurality of low resolution (LR) images from different viewpoints, where the plurality of LR images capture a scene in which depths of points in the scene vary and each of the plurality of LR images differs from other LR images due to scene dependent geometric displacements due to parallax;

determining scene dependent parallax information with respect to the LR images based upon disparity relative to a reference point of view resulting from the depths of points in the captured scene; and

fusing at least portions of the LR images to form a high resolution (HR) image by:

populating a high resolution (HR) grid corresponding to pixel locations of the HR image with LR image samples from the LR images using the scene dependent parallax information;

interpolating samples at grid position of the HR grid using neighboring available samples from the LR image samples placed on the HR grid; and

instructing a display device to display the HR image.

20. An array camera, comprising:

a plurality of cameras with different viewpoints;

at least one processor; and

memory containing an image processing pipeline software application;

wherein the image processing pipeline software directs the at least one processor to:

acquire a plurality of low resolution (LR) images from the different viewpoints, where the plurality of LR images capture a scene in which depths of points in the scene vary and each of the plurality of LR images differs from other LR images due to scene dependent geometric displacements due to parallax;

determine scene dependent parallax information with respect to the LR images based upon disparity relative to a reference point of view resulting from the depths of points in the captured scene; and

fuse at least portions of the LR images to form a high resolution (HR) image by:

populate a high resolution (HR) grid corresponding to pixel locations of the HR image with LR image samples from the LR images using the scene dependent parallax information;

interpolate samples at grid position of the HR grid using neighboring available samples from the LR image samples placed on the HR grid; and

instructing a display device to display the HR image.