Post on 17-Jul-2020
transcript
A Method for Refocusing Photos using Depth from DefocusBenjamin Wiberg
Department of Electrical Engineering, Stanford University
Motivation Image Processing Pipeline
Related Work Experimental Results
• Depth information from single viewpoint • Extract image that is sharp everywhere • Adjust blur “look” after taking photo • Uses normal camera, more accessible than light field
Capture stack of frames from same viewpoint while varying focal depth “through” scene
Focal stackFocal stackFocal stack
Sharp image
Gaussian blur with increasing
sigmas (σ)
Select pixel from frame with highest sharpness
Select pixel from frame with highest sharpness
Select pixel from frame with highest sharpness
Uniformly blurred stack
Uniformly blurred stack
Uniformly blurred stack (σ)
Focal stackFocal stackFocal stack
Align with homography and RANSAC
Align with homography and RANSAC
Align with homography and RANSAC
First frame (near focus)
Last frame (far focus)
Select sigma where diff is
smallest
Select sigma where diff is
smallest
Blur map for near/far focus frames
Blur maps for near/far focus frames
Combine: near + far
Depth map
Refocus by spatially vary blur
with depth
Refocused image
10 images 7 images
dept
h
dept
h
refo
cuse
d
refo
cuse
d
Most DfD methods estimate physical depth by estimating aperture, focal length and frame focal depths. • Depth from Focus with Your Mobile Phone, S. Suwajanakorn, C. Hernandez, S. Seitz
• An investigation of methods for determining depth from focus, J. Ens, P. Lawrence
Here we approximate depth by blurriness directly to reduce computation time, which is good enough for refocusing purposes. The major drawback is that the depth map does not contain physical distances to objects in the scene.