by Anika Radiya-Dixit

On Friday, April 10, while campus was abuzz with Blue Devil Days, a series of programs for newly admitted students, a group of digital image buffs gathered in the Levine Science Research Center to learn about the latest research on image and video de-blurring from Duke electrical and computer engineering professor Guillermo Sapiro. Professor Sapiro specializes in image and signal analysis in the department of Computer and Electrical Engineering in Duke’s Pratt School of Engineering. Working alongside Duke postdoctoral researcher Mauricio Delbracio, Sapiro has been researching methods to remove image blur due to camera shake.

Sapiro’s proposed algorithm is called burst photography, which achieves “state-of-the-art results an order of magnitude faster, with simplicity for on-board implementation on camera phones.” As shown in the image below, this technique combines multiple images, where each has a random camera shake and therefore each image in the burst is blurred slightly differently.

Professor Sapiro explains the basic principle of burst photography.

Professor Sapiro explains the basic principle of burst photography.

To de-blur the image, Sapiro’s algorithm then aligns the images together using a gyroscope and combines them in the Fourier domain. The final result essentially takes the best parts of each slightly-blurred image — such as the ones below — and gives sharpened images a greater weight when averaging blurred images in the burst.

Set of images with varying degrees of linear blur.

Set of images with varying degrees of linear blur.

This technique also produces phenomenal effects in video sharpening by collapsing multiple blurred frames into a single sharpened picture:

Contrast between sample frame of original video (left) with FBA sharpened video (right).

Contrast between sample frame of original video (left) with FBA sharpened video (right).

One impressive feature of burst photography is that it allows the user to obtain a mixed-exposure image by taking multiple images at various levels of exposure, as can be seen in parts (a) and (b) in the figure below, and then combining these images to produce a splendid picture (c) with captivating special effects.

Result of FBA algorithm on combining images with various levels of exposure.

Result of FBA algorithm on combining images with various levels of exposure.

If you are interested in video and image processing, email Professor Sapiro or check out his lab.