Generating dolly zoom effect using light field image data

Abstract

A sensor is configured to acquire a light field by imaging a scene. A processor is configured to determine four-dimensional (4D) coordinates of points in a light field and generate dollied coordinates from the 4D coordinates based on a dolly transform and a dolly parameter. The processor is also configured to project rays associated with the dollied coordinates from the light field onto corresponding points in an output raster. In some cases, the processor applies an aperture function to filter the rays in the coordinate system of the dollied coordinates. The aperture function has a first value in a first region of an aperture plane and the aperture value has a second value in a second region of the aperture plane. Rays passing through the first region are accepted by the aperture function and rays passing through the second region are rejected.

Description

BACKGROUND

A light field camera, also known as a plenoptic camera, captures information about the light field emanating from a scene; that is, the intensity of light in a scene, and also the direction that the light rays are traveling in space. This contrasts with a conventional camera, which records only light intensity. One type of light field camera uses an array of micro-lenses placed in front of an otherwise conventional image sensor to sense intensity, color, and directional information. Multi-camera arrays are another type of light field camera.

BRIEF DESCRIPTION OF THE DRAWINGS

The present disclosure may be better understood, and its numerous features and advantages made apparent to those skilled in the art by referencing the accompanying drawings. The use of the same reference symbols in different drawings indicates similar or identical items

FIG. 1 illustrates an example of a dolly zoom effect according to some embodiments.

FIG. 2 illustrates an example of dolly-zoom as a coordinate transformation according to some embodiments.

FIG. 3 is an illustration of a dolly-zoom as a transformation of the light field according to some embodiments.

FIG. 4 illustrates a “dolly twirl” analogous to the dolly zoom of FIG. 2 according to some embodiments.

FIGS. 5 and 6 show examples of opposite extremes of a dolly-zoom action according to some embodiments.

FIGS. 7 and 8 show examples of opposite extremes of a dolly-twirl action according to some embodiments.

FIG. 9 is a block diagram of an architecture of a light field capture device such as a camera according to some embodiments.

FIG. 10 is a block diagram of an architecture of a light field capture system according to some embodiments.

DETAILED DESCRIPTION

Light fields can be manipulated to computationally reproduce various elements of image capture. One example is refocusing, which is equivalent to a virtual motion of the sensor plane. As described herein, virtual motion of the main lens plane is also possible, resulting in a dolly zoom effect in which a field-of-view is modified while a camel location moves towards or away from a subject in a manner that maintains a size of the subject in the frame. The dolly zoom effect was first used to portray James Stewart's fear of heights in the film Vertigo and is also well known for its use portraying Roy Scheider's anxiety in the film Jaws. Existing techniques to create a dolly zoom effect rely on the use of a depth map during 2D reconstruction. This requirement of an accurate depth map limits the robustness and applicability of the dolly-zoom effect.

Some embodiments of the light field are represented as a four-dimensional (4D) function that can be formed of two-dimensional (2D) images of a portion of a scene. The images represent views of the portions of the scene from different perspectives and frames are rendered from the point of view of the camera by sampling portions of the 2D images. For example, the coordinates (u, v, s, t) in the 4D function that represents the light field can be defined so that (u, v) represent coordinates within one of the camera images in the light field and (s, t) are coordinates of pixels within the camera image. Other definitions of the four coordinates of the 4D function that represents the light field can also be used.

According to various embodiments, a dolly zoom effect can be applied directly to the light field as a 4D transformation of some or all of the coordinates (u, v, s, t) that are used to define locations in the 4D light field. Using the techniques described herein, explicit depth calculations are not required, allowing the approach to be robustly applied to any light field. Furthermore, the techniques described herein do not require 2D reconstruction, thus allowing the full 4D light field to be dollied, for example for applications such as light field displays.

The described techniques can be implemented as a 4D transformation of the light field, and can easily be combined with other transformations, such as refocus, to create various effects.

Basic Dolly Zoom Equation

A simple version of the dolly zoom equation can be derived analogously to the refocus equation. For this derivation we reduce the light field to 2D without loss of generality.

FIG. 1 illustrates an example 100 of a dolly zoom effect according to some embodiments. The s coordinate denotes position on the sensor 105, while the u coordinate denotes position on the aperture 110. Three rays 111, 112, 113 (collectively referred to herein as “the rays 111-113”) are shown to pass through a location on the aperture 110 (not necessarily at the origin of the coordinate system used in the plane of the aperture 110) and hit the sensor 105.

A virtual aperture 120 is dollied a distance d relative to the plane of the aperture 110. From simple geometry it is apparent that the locations where the rays 111-113 intersect the virtual aperture 120 change proportionally to the slope of the rays 111-113 and d. Thus, the location where one of the rays 111-113 intersects the virtual aperture 120 is:u′=u+k·s·d

where k is a normalization constant such that k*s is the slope of the ray.

In some embodiments, the values of k and d our combined into a single constant γ.

Thus, the light field transformation can be expressed as:

$\left[\begin{array}{c}{s}^{\prime }\\ u^{\prime }\end{array}\right]=\left[\begin{array}{cc}1& 0\\ \gamma & 1\end{array}\right]\left[\begin{array}{c}s\\ u\end{array}\right]$

or for the full 4D transform the light field is represented as:

$\left[\begin{array}{c}{s}^{\prime }\\ t^{\prime }\\ u^{\prime }\\ v^{\prime }\end{array}\right]=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ \gamma & 0& 1& 0\\ 0& \gamma & 0& 1\end{array}\right]\left[\begin{array}{c}s\\ t\\ u\\ v\end{array}\right]$

Mathematically this transformation skews the light field along the uv plane. Intuitively, this skewing of the light field along the uv plane can be thought of as introducing a perspective shift that is proportional to the st location. An object's motion with a perspective shift is proportional to its depth, which can be interpreted as a depth dependent scaling. However, there is no need for the depth to be explicitly calculated.

Thinking of this dolly action as a depth-dependent scaling reveals that, for objects at the focus position (lambda 0), no scaling occurs. This can be thought of as a counter zoom being applied in order to keep apparent size the same for objects at the focus distance. Thus this transform simultaneously dollies and zooms the light field image.

Example Implementation

In at least one embodiment, a dolly zoom effect is implemented using the light field projection method of 2D reconstruction as follows:

• • 1. A 4D coordinate (s, t, u, v) is accessed or calculated for each point in the light field.
  • 2. Given a dolly parameter γ, the dollied coordinate (s′, t′, u′, v′) is determined using the above equation, i.e. s′=s, t′=t, u′=u+γs, v′=v+γt
  • 3. An aperture function α(u, v) is accessed or generated. In the illustrated embodiment, the aperture function has a value of 1 where the aperture accepts rays and 0 where the aperture rejects rays. However, in other embodiments, the aperture function is implemented using other values or methods of representation.
    • a. Some embodiments of a circular aperture with radius r have a value of the aperture function α(u, v) equal to 1 when u²+v²<r² and 0 otherwise. The natural fully open aperture would have r=0.5
    • b. The transformation of u′ and v′ effectively shifts the center of the aperture. A somewhat smaller aperture than full open can be used, so as to avoid the aperture function going out of bounds of the actual data, as shown in FIG. 2.
    • 4. If the aperture function α(u′, v′) accepts the ray, then the light field ray is projected to point (x, y)=(s′, t′) on the output raster.

FIG. 2 illustrates an example of dolly-zoom as a coordinate transformation according to some embodiments. From left to right, there is shown: the original light field 200, the light field with negative dolly-zoom 205, and the light field with positive dolly-zoom 210. The center (uv) location of each st cell changes.

In the example of FIG. 2, the above-described method is illustrated using a 3×3 light field. Each cell represents an (s, t) coordinate, while the coordinate within each cell represents (u, v). The central point 215 (only one indicated by a reference numeral in the interest of clarity) within each cell is the center of the (u, v) coordinate system. The gray areas represent valid data within the light field. The circle 220 (only one indicated by a reference numeral in the interest of clarity) represents the aperture function that is used in the above algorithm, which as noted above is smaller than the full size of the captured data.

The illustrations are examples only, and do not imply any particular method of light field capture. However, in some embodiments, a plenoptic camera with a 3×3 microlens array is used to capture the light field. In such a case, each microlens would be the (s, t) location, and within each microlens is a (u, v) coordinate system centered on the red x.

FIG. 2 illustrates an intuitive interpretation that dolly-zoom produces a spatially varying perspective shift. The virtual aperture, a.k.a. the aperture function, undergoes a perspective shift that is proportional to its (s, t) location, thus producing depth-dependent scaling in the 2D reconstructed image. This can be visualized as the scaling of the grid formed by the central points 215. As mentioned above, since the virtual aperture is being shifted, the virtual aperture is oriented or sized so that it overlaps with valid data. This also means a smaller virtual aperture will allow for more dolly-zoom.

FIG. 3 is an illustration of a dolly-zoom as a transformation of the light field according to some embodiments. From left to right, there is shown: the original light field 300, the light field with negative dolly-zoom 305, and the light field with positive dolly-zoom 310. The coordinate system is the same throughout, but the underlying light field is transformed. The central point 315 (only one indicated by a reference numeral in the interest of clarity) within each cell is the center of the (u, v) coordinate system. The gray areas represent valid data within the light field. The circle 320 (only one indicated by a reference numeral in the interest of clarity) represents the aperture function that could be used in the above algorithm, which as noted above is smaller than the full size of the captured data.

FIG. 3 thus illustrates the mathematical equivalent of performing a transformation on the underlying light field data (perhaps using some sort of interpolation) instead of shifting the virtual apertures.

None of the above is to imply that an aperture function is needed or that 2D reconstruction is necessary. The use of the aperture function in the above figures is merely to illustrate the effect the dolly-zoom would have in the example 2D reconstruction algorithm. Other 2D reconstruction methods may be used that do not utilize an aperture, and 2D reconstruction is not even necessary if method is applied as a 4D transform of the light field. The 4D light field itself may be modified, either by changing the coordinate mapping as in FIG. 2, or by transformation of the underlying data as in FIG. 3, and then stored for future processing.

FIGS. 2 and 3 illustrate two different approaches that are useful in different situations. The first approach (FIG. 2) is useful in cases where light field data is irregularly gridded, or in other cases where explicit coordinates of each light field point are known and/or preferred. The modified light field coordinates are then fed to downstream processing, such as for example projection for 2D reconstruction. The second approach (FIG. 3) is useful in cases of where light field data is regularly gridded, or in other cases where the light field coordinates are implicitly known. The underlying data is modified and passed onto downstream processing. This can be useful in the case of a light field display, for example. In the second case, the data may be modified such that the data ends up outside of the original implicit grid, as shown in FIG. 3. This can be compensated for by padding the original data before transformation.

Composing Multiple Transforms

For illustrative purposes, and without loss of generality, the technique of combining multiple transforms is described in terms of a 2D light field.

Since dolly zoom is mathematically a linear transform, it can be combined with other linear transforms by matrix multiplication. For example, refocus and dolly zoom operations can be combined by using one of two transforms:

$\left[\begin{array}{c}{s}^{\prime }\\ u^{\prime }\end{array}\right]=\left[\begin{array}{cc}1& \lambda \\ 0& 1\end{array}\right]\left[\begin{array}{cc}1& 0\\ \gamma & 1\end{array}\right]\left[\begin{array}{c}s\\ u\end{array}\right]=\left[\begin{array}{cc}1+\lambda \gamma & \lambda \\ \gamma & 1\end{array}\right]\left[\begin{array}{c}s\\ u\end{array}\right]\left[\begin{array}{c}{s}^{\prime }\\ u^{\prime }\end{array}\right]=\left[\begin{array}{cc}1& 0\\ \gamma & 1\end{array}\right]\left[\begin{array}{cc}1& \lambda \\ 0& 1\end{array}\right]\left[\begin{array}{c}s\\ u\end{array}\right]=\left[\begin{array}{cc}1& \lambda \\ \gamma & 1+\lambda \gamma \end{array}\right]\left[\begin{array}{c}s\\ u\end{array}\right]$

The two transformations are not equivalent because composing transformations is not a commutative operation. Depending on the effect one wishes to achieve, the order of composition must be carefully selected.

In order to implement this in the previous example algorithm, step 2 can be modified in one of two ways.

Dolly First:
2.1   First modify the (u, v) coordinates as u′ = u + γs, v′ = v + γt
2.2   Then modify the (s, t) coordinates as s′ = s + λu′, t′ = t + λv′
2.2.1 Alternatively, modify the (s, t) coordinates at the same time as
      2.1, as per the above equations
      s′ = s(1 + λγ) + λu, t′ = t(1 + λγ) + λv
Refocus First:
2.1   First modify the (s, t) coordinates as s′ = s + λu, t′ = t + λv
2.2   Then modify the (u, v) coordinates as u′ = u + γs′, v′ = v + γt′
2.2.1 Alternatively, modify the (u, v) coordinates at the same time as
      2.1, as per the above equations
      u′ = u(1 + λγ) + γs, v′ = v(1 + λγ) + γt

One might be tempted to try to refocus and dolly simultaneously by using the following transform:

$\left[\begin{array}{c}{s}^{\prime }\\ u^{\prime }\end{array}\right]=\left[\begin{array}{cc}1& \lambda \\ \gamma & 1\end{array}\right]\left[\begin{array}{c}s\\ u\end{array}\right]$

However this is not a valid coordinate transform. In particular it is possible to select λ and γ such that the matrix is singular and thus not invertible.

More complicated compositions can be used as well. For example:

$\left[\begin{array}{c}{s}^{\prime }\\ u^{\prime }\end{array}\right]=\left[\begin{array}{cc}1& {\lambda }_2\\ 0& 1\end{array}\right]\left[\begin{array}{cc}1& 0\\ \gamma & 1\end{array}\right]\left[\begin{array}{cc}1& {\lambda }_1\\ 0& 1\end{array}\right]\left[\begin{array}{c}s\\ u\end{array}\right]$

One use of this composition is to execute a dolly zoom where the constant-size depth is at a different depth than lambda zero. In this case, the first refocus effectively changes the zero lambda depth, the dolly zoom occurs, and then the second refocus undoes the first refocus. The second refocus is not simply the inverse of the first refocus (since the dolly action remaps the refocus depths), but this could be a good first approximation. Exact relations may be derived so to have the refocuses exactly cancel out.

Certain compositions of transforms may be expressed as other transforms as well. In some embodiments, the above composition is utilized with the values:

${\lambda }_1={\lambda }_2=-\tan \left(\frac{\theta }{2}\right)$ $\gamma =\sin \left(\theta \right)$

In that case, the transformation is equivalent to rotating the light field in the epipolar plane

$\left[\begin{array}{c}{s}^{\prime }\\ u^{\prime }\end{array}\right]=\left[\begin{array}{cc}\cos \theta & \sin \theta \\ -\sin \theta & \cos \theta \end{array}\right]\left[\begin{array}{c}s\\ u\end{array}\right]$

Rotation by shearing is described, for example, in A. W. Paeth, A Fast Algorithm for General Raster Rotation, Computer Graphics Laboratory, Department of Computer Science, University of Waterloo, 1986.

Spatially Varying Transform

Neither the refocus parameter λ nor the dolly zoom parameter γ is required to be constant. Some embodiments of the refocus parameter λ or the dolly zoom parameter γ vary as a function of (s, t, u, v), or even (s′, t′, u′, v′). For example, the dolly zoom parameter γ can vary as a function of (u, v) and the refocus parameter λ can vary as a function of (s, t). If γ(u,v) is a planar function, this mimics the physical action of tilting the lens. This is analogous to how sensor tilt can be mimicked by having λ(s, t) be a planar function.

One artistic effect is called “Lens Whacking”, which is achieved by shooting with the lens free floating from the body. This means that focus and tilt effects are achieved by the relative placement of the lens to the sensor. By utilizing the ability to virtually produce tilt in both the lens plane and aperture plane, lens whacking effects can be performed computationally in a light field.

Depth-Based Transformations

In at least one embodiment, the dolly zoom effect represents a depth-dependent scaling of the image. Thus, many 2D transforms can be generalized to be depth dependent. One possible way for linear transforms is to use the equation

$\left[\begin{array}{c}{s}^{\prime }\\ t^{\prime }\\ u^{\prime }\\ v^{\prime }\end{array}\right]=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ t_{11}-1& t_{12}& 1& 0\\ t_{21}& t_{22}-1& 0& 1\end{array}\right]\left[\begin{array}{c}s\\ t\\ u\\ v\end{array}\right]=\left[\begin{array}{cc}I& 0\\ T-I& I\end{array}\right]\left[\begin{array}{c}s\\ t\\ u\\ v\end{array}\right]$

Where the linear transform is described by

$T=\left[\begin{array}{cc}{t}_{11}& t_{12}\\ t_{21}& t_{22}\end{array}\right]$

In this simple case where T is a scaling matrix,

$T=\left[\begin{array}{cc}s& 0\\ 0& s\end{array}\right]\Rightarrow T-I=\left[\begin{array}{cc}s-1& 0\\ 0& s-1\end{array}\right]$

It is apparent that this reduces to the dolly zoom equation in the case in which γ=s−1.

Another potentially interesting effect is rotation. In this case

$T=\left[\begin{array}{cc}\cos \theta & \sin \theta \\ -\sin \theta & \cos \theta \end{array}\right]\Rightarrow T-I=\left[\begin{array}{cc}\cos \theta -1& \sin \theta \\ -\sin \theta & \cos \theta -1\end{array}\right]$

Using a small angle approximation this can be further simplified to

$\left[\begin{array}{cc}\cos \theta -1& \sin \theta \\ -\sin \theta & \cos \theta -1\end{array}\right]\approx \left[\begin{array}{cc}0& \theta \\ -\theta & 0\end{array}\right]$

Thus, the following transform produces a rotation in the image proportional to depth (“dolly twirl”):

$\left[\begin{array}{c}{s}^{\prime }\\ t^{\prime }\\ u^{\prime }\\ v^{\prime }\end{array}\right]=\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& \theta & 1& 0\\ -\theta & 0& 0& 1\end{array}\right]\left[\begin{array}{c}s\\ t\\ u\\ v\end{array}\right]$

FIG. 4 illustrates a “dolly twirl” analogous to the dolly zoom of FIG. 2 according to some embodiments. From left to right there is shown: the original light field 400, the light field with negative theta 405, and the light field with positive theta 410. The central point 415 (only one indicated by a reference numeral in the interest of clarity) within each cell is the center of the (u, v) coordinate system. The gray areas represent valid data within the light field. The circle 420 (only one indicated by a reference numeral in the interest of clarity) represents the aperture function that could be used in the above algorithm, which as noted above is smaller than the full size of the captured data. The center (uv) location 415 of each st cell changes.

This technique can be further generalized to higher order transforms. For any given 2D transform, the identity part can be removed, and the remainder applied to the uv coordinates. This can be used to create effects that are otherwise not possible to produce using traditional camera setups.

For example, lens distortion (barrel and pincushion) is modeled as a quadratic transformation, with barrel and pincushion having opposite effects. If this distortion is applied proportionally to depth, the result is an image that has pincushion distortion for near objects and barrel distortion for far objects, or vice versa. Other distortions, such as perspective distortion, can be applied in this depth dependent manner as well.

Image Examples

FIGS. 5 and 6 show examples 500, 600 of opposite extremes of a dolly-zoom action according to some embodiments.

FIGS. 7 and 8 show examples 700, 800 of opposite extremes of a dolly-twirl action according to some embodiments.

Image Data Acquisition Devices

FIG. 9 is a block diagram of an architecture 900 of a light field capture device such as a camera according to some embodiments. The light field capture device is used to implement some embodiments of the techniques described herein. In the illustrated embodiment, the light field capture device includes a light field image data acquisition device 905 that is configured to capture images using optics 910, an image sensor 915, which is implemented using a plurality of individual sensors for capturing pixels, and a microlens array 920. Some embodiments of the optics 910 include an aperture 925 for allowing a selectable amount of light into the light field capture device and a main lens 930 for focusing light toward the microlens array 920. Some embodiments of the microlens array 920 are disposed or incorporated in the optical path of the light field capture device so as to facilitate acquisition, capture, sampling of, recording, or obtaining light field image data via the sensor 915.

The light field image data acquisition device 905 includes the user interface 935 for allowing a user to provide input for controlling the operation of the light field capture device for capturing, acquiring, storing, or processing image data. Control circuitry 940 is used to facilitate acquisition, sampling, recording, or obtaining light field image data. For example, the control circuitry 940 can manage or control (automatically or in response to user input) the acquisition time and, rated acquisition, sampling, capturing, recording, or obtaining light field image data. Memory 945 is used to store image data, such as output from the image sensor 915. The memory 945 is implemented as external or internal memory, which can be provided as a separate device or location relative to the light field capture device. For example, the light field capture device can store raw light field image data output by the image sensor 915, or a representation thereof, such as a compressed image data file.

Post-processing circuitry 950 in the light field image data acquisition device 905 is used to access or modify image data acquired by the image sensor 915. Some embodiments of the post-processing circuitry 950 are configured to create dolly zoom effects using the image data, as discussed herein. For example, the post-processing circuitry 950 can include one or more processors executing software stored in the memory 945 to access and modify image data stored in the memory 945 to produce dolly zoom effects in images captured by the image sensor 915.

FIG. 10 is a block diagram of an architecture 1000 of a light field capture system according to some embodiments. The light field capture device is used to implement some embodiments of the techniques described herein. Elements in FIG. 10 that are referenced using the same reference numerals as corresponding elements in FIG. 9 perform the same or similar functions as the corresponding elements in FIG. 9. The architecture 1000 shown in FIG. 10 differs from the architecture 900 shown in FIG. 9 because the light field image data acquisition device 905 is implemented separately from a postprocessing system 1005, which includes the memory 945, the post-processing circuitry 950, and the user interface 935. The light field image data acquisition device 905 provides image data 1010 to the postprocessing system 1005.

Refocus Analogies

The above-described techniques may, by analogy, be applied to light field refocus as well. For example, the refocus analogy of the dolly twirl effect would be:

$\left[\begin{array}{c}{s}^{\prime }\\ t^{\prime }\\ u^{\prime }\\ v^{\prime }\end{array}\right]=\left[\begin{array}{cccc}1& 0& 0& \phi \\ 0& 1& -\phi & 0\\ 0& 0& 1& 0\\ 0& 0& 0& 1\end{array}\right]\left[\begin{array}{c}s\\ t\\ u\\ v\end{array}\right]$

Claims

1. A method comprising: determining four-dimensional (4D) coordinates of points in a light field;generating dollied coordinates from the 4D coordinates based on a dolly transform and a dolly parameter;applying an aperture function to filter rays in the coordinate system of the dollied coordinates, wherein the aperture function has a first value in a first region of an aperture plane and the aperture value has a second value in a second region of the aperture plane such that rays passing through the first region are accepted by the aperture function and rays passing through the second region are rejected; andprojecting rays associated with the dollied coordinates from the light field onto corresponding points in an output raster.

2. The method of claim 1, wherein applying the aperture function to filter the rays in the coordinate system of the dollied coordinates comprises dollying a virtual aperture by a distance relative to the aperture plane.

3. The method of claim 2, wherein generating the dollied coordinates comprises applying a depth dependent dolly transformation to the 4D coordinates.

4. The method of claim 3, wherein dollying the virtual aperture produces a depth-dependent scaling of a two dimensional (2D) reconstructed image that is reconstructed from the light field.

5. The method of claim 1, further comprising: refocusing portions of the light field using a linear refocusing transform and a refocusing parameter.

6. The method of claim 1, wherein generating the dollied coordinates comprises generating the dollied coordinates by applying a quadratic dolly transform.

7. The method of claim 1, wherein generating the dollied coordinates comprises producing tilt in a lens plane and an aperture plane.

8. An apparatus comprising: a sensor configured to acquire a light field by imaging a scene; anda processor configured to: determine four-dimensional (4D) coordinates of points in a light field;generate dollied coordinates from the 4D coordinates based on a dolly transform and a dolly parameter;apply an aperture function to filter rays in the coordinate system of the dollied coordinates, wherein the aperture function has a first value in a first region of an aperture plane and the aperture value has a second value in a second region of the aperture plane such that rays passing through the first region are accepted by the aperture function and rays passing through the second region are rejected; andproject rays associated with the dollied coordinates from the light field onto corresponding points in an output raster.

9. The apparatus of claim 8, wherein the processor is configured to dolly a virtual aperture by a distance relative to the aperture plane.

10. The apparatus of claim 9, wherein the processor is configured to apply a depth dependent dolly transformation to the 4D coordinates.

11. The apparatus of claim 10, wherein dollying the virtual aperture produces a depth-dependent scaling of a two dimensional (2D) reconstructed image that is reconstructed from the light field.

12. The apparatus of claim 8, wherein the processor is configured to refocus portions of the light field using a linear refocusing transform and a refocusing parameter.

13. The apparatus of claim 8, wherein the processor is configured to generate the dollied coordinates by applying a quadratic dolly transform.

14. The apparatus of claim 8, wherein the processor is configured to tilt an image in a lens plane and an aperture plane applying the dolly transform.