Microsoft — yes, Microsoft — has demonstrated a technique consisting of a combination of inexpensive gyroscopes and accelerometer attached to a DSLR camera to estimate a blur function from the camera’s acceleration and angular velocity during an exposure and then deblur the resultant image.
The most common commercial approach for reducing image blur today is image stabilization (IS). They are mechanical methods that dampen camera motion by either offsetting lens elements or move the image sensor. Image Deblurring (ID) is similar to current IS methods in that it also uses inertial sensors but, while IS tries to dampen motion by assuming that the past motion predicts the future motion and reduces blur, ID records the actual camera motion and removes the blur from the image. ID can dampen 2D motion, camera roll, up to six degrees of motion.
The results are pretty convincing and amazing. See for yourself:
Read the article and view the images at: Microsoft.
- Research Paper[pdf 2.96MB]
- Slide Show [zip 23MB]