A N Rajagopalan

Convolutional Neural Networks (CNN) have recently advanced the state-of-the-art in generalized motion deblurring. Literature suggests that restoration of high-resolution blurred images requires a design with a large receptive field, which existing networks achieve by increasing the number of generic convolution layers, kernel-size, or the scales at which the image is processed. However, increasing the network capacity in this form comes with the burden of increased model size and lower speed. To resolve this, we propose a novel architecture composed of dynamic convolutional modules, namely feature transformation (FT) and spatial attention (SA). An FT module addresses the camera shifts responsible for the global blur in the input image, while a SA module addresses spatially varying blur due to dynamic objects and depth changes. Qualitative and quantitative comparisons on deblurring benchmarks demonstrate that our network outperforms prior art across factors of accuracy, compactness, and speed, enabling real-time deblurring.

Capturing multiple images using the burst mode of handheld cameras can be a boon to obtain a high resolution (HR) image by exploiting the subpixel motion among the captured images arising from handshake. However, the caveat with mobile phone cameras is that they produce rolling shutter (RS) distortions that must be accounted for in the super-resolution process. We propose a method in which we obtain an RS-free HR image using HR camera trajectory estimated by leveraging the intra- and inter-frame continuity of the camera motion. Experimental evaluations demonstrate that our approach can effectively recover a super-resolved image free from RS artifacts.

The Learning with Errors (LWE) problem has been extensively studied in cryptography due to its strong hardness guarantees, efficiency and expressiveness in constructing advanced cryptographic primitives. In this work, we show that using polar codes in conjunction with LWE-based encryption yields several advantages. To begin, we demonstrate the obvious improvements in the efficiency or rate of information transmission in the LWE-based scheme by leveraging polar coding (with no change in the cryptographic security guarantee). Next, we integrate wiretap polar coding with LWE-based encryption to ensure provable semantic security over a wiretap channel in addition to cryptographic security based on the hardness of LWE. To the best of our knowledge this is the first wiretap code to have cryptographic security guarantees as well. Finally, we study the security of the private key used in LWE-based encryption with wiretap polar coding, and propose a key refresh method using random bits used in wiretap coding. Under a known-plaintext attack, we show that non-vanishing information-theoretic secrecy can be achieved for the key. We believe our approach is at least as interesting as our final results: our work combines cryptography and coding theory in a novel 'non blackbox-way' which may be relevant to other scenarios as well.

Existing image registration methods that work in the presence of motion blur assume the scene to be fronto-parallel. In this work, we extend the state-of-the-art by cueing motion blur itself to infer plane inclination. This is achieved by matching extremities of blur kernels computed at different locations in the image. Because these extremities correspond to the same homography, we show that it is possible to find the orientation. Following this, we propose a registration method that reorients the source image to the target plane within a reblur-difference framework to detect the actual changes.

In this paper, we propose a video matting method with simultaneous noise reduction based on the Unscented Kalman filter (UKF). This recursive approach extracts the alpha mattes and denoised foregrounds from noisy videos, in a unified framework. No assumptions are made about the type of motion of the camera or of the foreground object in the video. Moreover, user-specified trimaps are required only once every ten frames. In order to accurately extract information at the borders between the foreground and the background, we include a discontinuity-adaptive Markov random field (MRF) prior. It incorporates spatio-temporal information from the current and previous frame during estimation of the alpha matte as well as the foreground. Results are given on videos with real film-grain noise. © 2010 IEEE.

The presence of motion blur is unavoidable in hand-held cameras, especially in low-light conditions. In this paper, we address the inverse rendering problem of estimating the latent image, scene depth and camera motion from a set of differently blurred images of the scene. Our framework can account for depth variations, non-uniform motion blur as well as mis-alignments in the captured observations. We initially describe an iterative algorithm to estimate ego motion in 3D scenes by suitably harnessing the point spread functions across the blurred images at different spatial locations. This is followed by recovery of latent image and scene depth by alternate minimization.

In this paper, we extend traditional Shape from focus (SFF) to the underwater scenario. Specifically, we show how 3D shape of objects immersed in water can be extracted from images captured under an optical microscope. Traditional SFF employs telecentric optics to achieve geometric registration among the frames. We extend conventional SFF to underwater objects under the assumption of negligible scattering. We establish that the property of telecentricity holds for objects immersed in water provided the numerical aperture is small. By modeling geometrical distortions due to refraction effects on the water surface, we prove that the depth map obtained in the presence of water is a scaled version of the original depth map. We also reveal that this scale factor is directly related to the refractive index of water. We validate performance with real experiments.

The color appearance of an object can vary widely as a function of camera sensitivity and ambient illumination. In this paper, we discuss a methodology for seamless interfacing across imaging sensors and under varying illumination conditions for two very relevant problems in aerial imaging, namely, change detection and mosaicing. The proposed approach works by estimating surface reflectance which is an intrinsic property of the scene and is invariant to both camera and illumination. We advocate SIFT-based feature detection and matching in the reflectance domain followed by registration. We demonstrate that mosaicing and change detection when performed in the high-dimensional reflectance space yields better results as compared to operating in the 3-dimensional color space.

The rule of perspectivity that 'straight-lines-mustremain-straight' is easily inflected in CMOS cameras by distortions introduced by motion. Lines can be rendered as curves due to the row-wise exposure mechanism known as rolling shutter (RS). We solve the problem of correcting distortions arising from handheld cameras due to RS effect from a single image free from motion blur with special relevance to urban scenes. We 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 rectify the RS image based on the estimated camera trajectory using inverse mapping. We show rectification results for RS images captured using mobile phone cameras. We also compare our single image method against existing video and nonblind RS rectification methods that typically require multiple images.

Dense point clouds of naturally existing but damaged real world structures result in 3D models that may be visually unpleasant for applications such as a 'Virtual Tour'. This paper addresses the reconstruction of such damaged regions (or, 'holes') in digital models. Without constraining the complexity or size of the hole, a noniterative framework based on Tensor Voting (TV) is proposed to fill-in missing regions by using neighbourhood surface geometry and geometric prior derived from registered self-similar examples. The effectiveness of the reconstruction is demonstrated on holes with different complexities on synthetic and real data. © 2012 ICPR Org Committee.