Graphics Reference
In-Depth Information
Fig. 13.5
Light collected over multiple frames is fused after alignment of video frames to com-
pensate for platform motion
object is smeared across the image. Processing steps to deblur the image, e.g., by
deconvolution, is near impossible because it is an ill-posed problem with uncertainty
in both object and platform motion. In a computational imaging framework, a coded
exposure approach would vary the period of time when the shutter is open during a
chosen exposure time. This “flutter shutter” approach changes the ways in which the
light is captured on each pixel, and preserves the high-frequency spatial details. To
that end, deblurring through deconvolution becomes a solvable solution, and has been
demonstrated for several challenging cases of motion-blur including large motion,
textured backgrounds, and partial occlusions [
8
].
Another approach for motion deblurring relies on alignment and fusion of the
coded exposure images. That is, by removing camera motion through video stabi-
lization of images, light can be “collected” appropriately by summing the proper
pixels associated with a particular part of the scene and object, as shown in Fig.
13.5
.
Light from the “same location” will be fused or summed across temporal frames,
and as such, the described motion compensated system would generate video at a
lower rate based on a higher rate input source (e.g., output at 60Hz based on an input
at 240Hz frame rate). With this approach, the deblurring problem is once again a
solvable problem, and the video frames contain light intensity information across
multiple frames.
In this approach, images captured with low exposure would be less blurry but
would be noisy because the sensor does not have time to capture sufficient amount of
light. Dark current noise from each sensor pixel would dominate the signal produced
from captured light, similar to low-light conditions where images are speckled with
random salt-and-pepper noise. Images captured with high exposure would be less
noisy but would be susceptible to blur because light is smeared across the image.
Combining both low and high exposure image would produce both a low blur and
low noise, with high dynamic range.
