Research

S. McCloskey. Temporally Coded Flash Illumination for Motion De-blurring. IEEE International Conference on Computer Vision (ICCV), 2011.
PDF © 2011 by IEEE.

S. McCloskey, K. Muldoon, and S. Venkatesha. Motion Invariance and Custom Blur from Lens Motion. 3rd IEEE International Conference on Computational Photography (ICCP), 2011.
PDF © 2011 by IEEE.

W. Xu and S. McCloskey. 2D Barcode Localization and Motion Deblurring Using a Flutter Shutter Camera. Workshop on Applications of Computer Vision (WACV), 2011.
PDF © 2011 by IEEE. Poster

S. McCloskey, W. Au, and J. Jelinek. Iris Capture from Moving Subjects Using a Fluttering Shutter Proceedings of the Fourth IEEE Conference on Biometrics Theory, Applications, and Systems (BTAS), 2010.
PDF © 2010 by IEEE.

S. McCloskey. Velocity-Dependent Shutter Sequences for Motion Deblurring. Proceedings of the European Conference on Computer Vision (ECCV), 2010.
PDF © 2010 by Springer-Verlag. Poster

Y. Ding, S. McCloskey, and J. Yu. Analysis of Motion Blur with a Flutter Shutter Camera for Non-Linear Motion. Proceedings of the European Conference on Computer Vision (ECCV), 2010 (oral).
PDF © 2010 by Springer-Verlag

S. McCloskey, M. Langer, and K. Siddiqi. Removing Partial Occlusion from Blurred Thin Occluders. Proceedings of the 20th International Conference on Pattern Recognition (ICPR), 2010.
PDF © 2010 by IEEE. Poster

A. Kembhavi, B. Siddiquie, R. Miezianko, S. McCloskey, L. Davis. Scene it or not? Incremental Multiple Kernel Learning for Object Detection. Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2009.
PDF © 2009 by IEEE

S. McCloskey and M. Langer. Planar Orientation from Blur Gradients in a Single Image. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2009.
PDF © 2009 by IEEE. Poster

S. McCloskey. Confidence Weighting for Sensor Fingerprinting Workshop on Vision of the Unseen (with CVPR), 2008.
PDF © 2008 by IEEE

S. McCloskey, M. Langer, and K. Siddiqi. Evolving Measurement Regions for Depth from Defocus. Proceedings of the 8th Asian Conference on Computer Vision (ACCV), 2007.
PDF © 2007 by Springer-Verlag

S. McCloskey, M. Langer, and K. Siddiqi. Automatic Removal of Partial Occlusion Blur. Proceedings of the 8th Asian Conference on Computer Vision (ACCV), 2007.
PDF © 2007 by Springer-Verlag

S. McCloskey, M. Langer, and K. Siddiqi. Seeing Around Occluded Objects Proceedings of the 18th International Conference on Pattern Recognition (ICPR), 2006.
PDF © 2006 by IEEE

S. McCloskey, M. Langer, and K. Siddiqi. The Reverse Projection Correlation Principle for Depth from Defocus. Proceedings of the 3rd International Symposium on 3D Data Processing, Visualization and Transmission (3D PVT), 2006.
PDF © 2006 by IEEE

The main theme of my PhD research has to do with a reverse projection model for image formation and the consequences for blur. Reverse projection provides the answer to the following question: What region of a scene is recorded by a particular point (or pixel) on the sensor?

In the case of a pinhole aperture, each point records light that travels along the chief ray, which connects the sensor point with the lens' optical center. In the more general case of a finite aperture, each point on the sensor will record light reflected or emitted from a double cone in scene space. That cone has its base at the lens aperture and its apex at the point where the chief ray intersects the plane of focus.

In the framework of reverse projection, I've experimented with two problems: 3D scene reconstruction by depth from defocus and the removal of unwanted image intensity due to partial occlusions.

Depth from defocus (DFD) is a classical problem in Computer Vision, where we attempt to reconstruct the 3D structure of a scene by measuring the change in optical blur between two images. Having developed a reverse projection camera model, we illustrate the fact that - when out of focus - neighboring pixels record light from overlapping regions of a scene. This overlap results in an increased correlation between the intensities of neighboring pixels, which we measure and relate to scene depth.

Test Images:

More recently, we have developed an algorithm that evolves a measurement region in order to allow for the recovery of higher-quality depth maps in regions of changing depth. This relaxes the equifocal assumption, which requires that all points over which blue is measured have constant depth.

Partial occlusions arise in natural images when an object is much closer to the camera then the plane of focus. This would occur when trying to take a picture through a fence, or when the camera strap or a photographer's finger accidentally falls with in the camera's field of view. Using a reverse projection model, we determine the fraction of a pixel's intensity that is due to the occluding object. This allows for the removal of the unwanted intensity under common conditions.

Test Images:

In order to remove the effect, it is necessary to find the contour of complete occlusion, and to estimate the width of the partially-occluded region. Manual specification of these characteristics is time consuming and error prone, so we have developed an automated method to estimate these quantities from a single image. Some results (input/output pairs):