CVG Projects

Coded Photography

Computer Vision Group, Universit├Ąt Bern, Switzerland

Coding Defocus Blur

Real cameras have a depth of field beyond which the image appears blurry. We can exploit this defocus blur to determine the 3D structure of a scene. Our research in depth from defocus information is built upon two questions. First, we investigate which aperture shapes are most suitable in different depth estimation setups. Second, we look for algorithms that allow for depth estimation from a single input image.

Optimized Aperture Shapes

Conventional digital cameras are built to simulate the human eye. With this intention the apertures of these cameras are approximatly circular. But are these aperture shapes actually best when we want to estimate depth from defocus information?
We designed an intuitive criterion to evaluate aperture shapes for depth estimation. Among the multitude of thinkable apertures, we can find aperture shapes that optimize this criterion efficiently. And, indeed, these apertures show a better performance in depth estimation. We evaluated this criterion for several depth estimation setups, i.e., for depth from different focus settings, for depth from coded aperture pairs, and for depth from a single coded aperture image.


Depth from Defocus all masks as zip
Depth from Coded Aperture Pairs all masks as zip
Depth from a Single Coded Aperture Image all masks as zip
3 x 3 11 x 11 21 x 21

Single Image Depth Estimation

Modifying the shape of the defocus blur, the ambiguity between low frequency, in-focus texture and blurred texture is mitigated. Thus depth information can be obtained from a single image. We introduce algorithms that are fast and efficient. We can achieve this degree of efficiency, as our algorithms do not rely on explicit deblurring of the input image for depth estimation. Instead, we make use of the modified image properties directly.


Single Image Depth Estimation: MATLAB Code


A. Sellent and P. Favaro:
"Coded Aperture Flow",
in Proceedings of the German Conference on Pattern Recognition, pp. 1 - 11, September 2014.
A. Sellent and P. Favaro:
"Which Side of the Focal Plane are You on?",
in IEEE Conference on Computational Photography (ICCP), 2014.
A. Sellent and P. Favaro:
"Optimized Aperture Shapes for Depth Estimation",
Pattern Recognition Letters, to appear.
A. Sellent and P. Favaro:
"Optimising Aperture Shapes for Depth Estimation",
Workshop on Vision, Modeling and Visualization (VMV), 2013
M. Martinello and P. Favaro :
"Depth estimation from a video sequence with moving and deformable objects",
IET Conf. on Image Processing, pp.1-6
M. Martinello and P. Favaro :
"Single Image Blind Deconvolution with Higher-Order Texture Statistics",
Video Processing and Computational Video, 2011, Dagstuhl Proceedings, pp. 124-151
M. Martinello and P. Favaro:
"Fragmented Aperture Imaging for Motion and Defocus Deblurring",
IEEE International Conference on Image Processing (ICIP), 2011, pp. 3413-3416
M. Martinello, T. E. Bishop and P. Favaro:
"A Bayesian Approach to Shape From Coded Aperture",
IEEE International Conference on Image Processing (ICIP), 2010, pp.3521-3524