Image Processing

Technical Papers

Image Processing

Tuesday, 11 August 3:45 PM - 5:35 PM | Los Angeles Convention Center, Room 150/151 Session Chair: Steve Lin, Microsoft Research Asia


Perceptually Based Downscaling of Images

This paper proposes a perceptually based method for downscaling images that provides a better apparent depiction of the input image. The downscaled images retain perceptually important features and details, and are spatio-temporally consistent. The algorithm has a simple, efficient, and parallelizable implementation, with running times similar to linear filtering.

Cengiz Oztireli
ETH Zürich

Markus Gross
ETH Zürich

Dehazing Using Color Lines

This paper proposes modified color lines prior for image dehazing. The new prior allows assessment of its validity across the image and avoids mispredictions where it fails to take place. Extensive evaluation of the new method demonstrates its greater accuracy.

Raanan Fattal
Hebrew University of Jerusalem

An L1 Image Transform for Edge-Preserving Smoothing and Scene-Level Intrinsic Decomposition

This paper introduces piecewise image flattening, which preserves salient edges and contours while eliminating insignificant details, producing a nearly piecewise constant image with sparse structures. Image flattening facilitates both edge-preserving smoothing and intrinsic image decomposition. The decomposed images have applications in surface re-texturing and object insertion into photographs.

Sai Bi
The University of Hong Kong

Xiaoguang Han
The University of Hong Kong

Yizhou Yu
The University of Hong Kong

Learning to Remove Soft Shadows

A data-driven method for removing soft shadows from RGB photographs. After the user selects a region using broad brush strokes, the algorithm automatically reconstructs the underlying matte and enables removal or modification of the shadow.

Maciej Gryka
University College London

Michael Terry
University of Waterloo

Gabriel J. Brostow
University College London

A Computational Approach for Obstruction-Free Photography

A unified computational approach for taking photos through reflecting or occluding visual obstructions, such as windows and fences. From a short image sequence, the algorithm leverages motion parallax to automatically remove these visual obstructions.

Tianfan Xue
Microsoft Corporation, Massachusetts Institute of Technology

Michael Rubinstein
Google Inc.

Liu Ce
Google Inc.

William Freeman
Massachusetts Institute of Technology