LightField-CameraArrayCalibration

Tools for calibrating a high-density array of cameras for light field imaging. Some of this code is an adaptation of Ashley Stewart's Planar-LF-Calibration.

Usage

A calibration pipeline using these functions should follow this:

createTransformMatrixLF_Center/Checkerboard->rectifyImagesViaTransforms->cropRectified->image2stuv

->analyzeEPI

Functions

analyzeEPI(st_uv)

Creates visuals of an EPI from the central light field sub-image and its frequency content. Can be used to validate the homography matrices applied to the light field sub-images. Choosing the V coordinate is important to get an EPI with enough depth content. Feature detection could be applied to find appropriate rows of pixels that display depth content.

Inputs:

• st_uv: A 4-D grayscale or 5-D RGB light field in two-plane parameterization form.

tforms = createTransformMatrixLF_Center(originalDir)

Finds the homography matrices for each light field sub-image such that they are warped to the perspective of the central light field sub-image. Uses SURF features to match points. Results vary based on the geometry of the objects in the scene.

Inputs:

• originalDir: directory of light field sub-images, numerically ordered such that the lowest number is the bottom left light field sub-image and a zig-zag naming pattern is followed.

Outputs:

• tforms: The homography matrices as a projective2d object

tforms = createTransformMatrixLF_Checkerboard(checkerboardDir)

Finds the homography matrices for each light field sub-image such that they are warped to the perspective of the central light field sub-image. Uses a checkerboard as a reference plane for calibration. This results in a better calibration than using SURF features, but requires calibrating the light field to a specific reference plane.

Inputs:

• checkerboardDir: directory of light field sub-images, numerically ordered such that the lowest number is the bottom left light field sub-image and a zig-zag naming pattern is followed. The light field should be an image of a rectangular checkerboard where one side has an even number of squares and the other is odd.

Outputs:

• tforms: The homography matrices as a projective2d object

cropRectified(rectifiedDir, croppedDir)

Crops the rectified images to a desired size, using the central light field sub-image as a reference.

Inputs:

• rectifiedDir: directory of rectified light field sub-images, numerically ordered such that the lowest number is the bottom left light field sub-image and a zig-zag naming pattern is followed.
• croppedDir: directory to write cropped and rectified light field sub-images.

[st_uv,st_uv_rgb] = image2stuv(imageDir)

Utility function to create 4-D grayscale and 5-D RGB light fields in the two-plane parameterization form from a directory of images.

Inputs:

• imageDir: Directory of light field sub-images, numerically ordered such that the lowest number is the bottom left light field sub-image and a zig-zag naming pattern is followed.

Outputs:

• st_uv: 4-D grayscale light field in two-plane parameterization form
• st_uv_rgb: 5-D RGB light field in two-plane parameterization form

rectifyImagesViaTransforms(tforms, originalDir, rectifiedDir)

Applies homography matrices found in createTransformMatrixLF_Center or createTransformMatrixLF_Checkerboard to a set of light field sub-images.

Inputs:

• tforms: A projective2d object of homography matrices
• originalDir: Directory of light field sub-images, numerically ordered such that the lowest number is the bottom left light field sub-image and a zig-zag naming pattern is followed.
• rectifiedDir: directory to write rectified light field sub-images.