Date: March 22, 2017 (Wednesday)
Time: 3PM to 4PM
Venue ESB 244
Speaker: Vijay Rengarajan A P (EE11D035)
Guides: A.N. Rajagopalan and R. Aravind
DC Members :
Dr. Krishna Jagannathan
Dr. Ramalingam C.S.
Dr. Chandrasekhar .C (CSE)
Dr. Sukhendu Das (CSE)
The rule of perspectivity that ‘straight-lines-must-remain-straight’ is easily inflected in CMOS cameras by distortions introduced by motion. Due to the row-wise exposure mechanism known as rolling shutter (RS), lines can be rendered as curves, and shapes which we consider as correct can be rendered differently in images. We solve the problem of correcting such perceived distortions arising from handheld cameras due to RS effect from a single image. We follow two approaches: the first one is an algorithmic approach based on the scene geometry with special relevance to urban scenes, and the second one is based on learning the mapping between the distorted image and camera motion for any class of scenes.
In the geometric approach, we first develop a procedure to extract prominent curves from the RS image since this is essential for deciphering the varying row-wise motion. We pose an optimization problem with line desirability costs based on straightness, angle, and length, to resolve the geometric ambiguities while estimating the camera motion based on a rotation-only model assuming known camera intrinsic matrix. Finally, we correct the RS image based on the estimated camera trajectory using inverse mapping. We compare our single image method against existing video and nonblind RS correction methods that typically require multiple images.
In the learning-based approach, we propose a convolutional neural network (CNN) architecture that automatically learns essential scene features from a single RS image to estimate the row-wise camera motion and undo RS distortions back to the time of first-row exposure. We employ long rectangular kernels to specifically learn the effects produced by the row-wise exposure. Experiments reveal that our proposed architecture performs better than the conventional CNN employing square kernels. Our single-image correction method fares well even operating in a frame-by-frame manner against video-based methods and performs better than scene-specific correction schemes even under challenging situations.
All are cordially invited.