Interactive digital matting, the process of extracting a foreground object from an image based on limited user input, is important task in and video editing. From computer vision perspective, this extremely challenging because it massively ill-posed -- at each pixel we must estimate background colors, as well opacity ("alpha matte") single color measurement. Current approaches either restrict estimation to small part image, estimating colors nearby pixels where they are known, or perform...

Colorization is a computer-assisted process of adding color to monochrome image or movie. The typically involves segmenting images into regions and tracking these across sequences. Neither tasks can be performed reliably in practice; consequently, colorization requires considerable user intervention remains tedious, time-consuming, expensive task.In this paper we present simple method that neither precise segmentation, nor accurate region tracking. Our based on premise; neighboring pixels...

Blind deconvolution is the recovery of a sharp version blurred image when blur kernel unknown. Recent algorithms have afforded dramatic progress, yet many aspects problem remain challenging and hard to understand. The goal this paper analyze evaluate recent blind both theoretically experimentally. We explain previously reported failure naive MAP approach by demonstrating that it mostly favors no-blur explanations. On other hand we show since size often smaller than estimation alone can be...

A conventional camera captures blurred versions of scene information away from the plane focus. Camera systems have been proposed that allow for recording all-focus images, or extracting depth, but to record both simultaneously has required more extensive hardware and reduced spatial resolution. We propose a simple modification allows simultaneous recovery (a) high resolution image (b) depth adequate semi-automatic extraction layered representation image.

A conventional camera captures blurred versions of scene information away from the plane focus. Camera systems have been proposed that allow for recording all-focus images, or extracting depth, but to record both simultaneously has required more extensive hardware and reduced spatial resolution. We propose a simple modification allows simultaneous recovery (a) high resolution image (b) depth adequate semi-automatic extraction layered representation image. Our is insert patterned occluder...

In blind deconvolution one aims to estimate from an input blurred image y a sharp x and unknown blur kernel k. Recent research shows that key success is consider the overall shape of posterior distribution p(x, k\y) not only its mode. This leads distinction between MAP <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">x, k</sub> strategies which mode pair x, k often lead undesired results, xmlns:xlink="http://www.w3.org/1999/xlink">k</sub> select...

Interactive digital matting, the process of extracting a foreground object from an image based on limited user input, is important task in and video editing. From computer vision perspective, this extremely challenging because it massively ill-posed - at each pixel we must estimate background colors, as well opacity ("alpha matte") single color measurement. Current approaches either restrict estimation to small part image, estimating colors nearby pixels where they are known, or perform...

Colorization is a computer-assisted process of adding color to monochrome image or movie. The typically involves segmenting images into regions and tracking these across sequences. Neither tasks can be performed reliably in practice; consequently, colorization requires considerable user intervention remains tedious, time-consuming, expensive task.In this paper we present simple method that neither precise segmentation, nor accurate region tracking. Our based on premise; neighboring pixels...

Blind deconvolution is the recovery of a sharp version blurred image when blur kernel unknown. Recent algorithms have afforded dramatic progress, yet many aspects problem remain challenging and hard to understand. The goal this paper analyze evaluate recent blind both theoretically experimentally. We explain previously reported failure naive MAP approach by demonstrating that it mostly favors no-blur explanations. show that, using reasonable priors, simulations estimation latent guaranteed...

When we take a picture through transparent glass, the image obtain is often linear superposition of two images: The scene beyond glass plus reflected by glass. Decomposing single input into images massively ill-posed problem: In absence additional knowledge about being viewed, there are an infinite number valid decompositions. this paper, focus on easier user assisted separation in which interactively labels small gradients as belonging to one layers. Even given part gradients, problem still...

We present spectral matting: a new approach to natural image matting that automatically computes basis set of fuzzy components from the smallest eigenvectors suitably defined Laplacian matrix. Thus, our extends segmentation techniques, whose goal is extract hard segments, extraction soft components. These may then be used as building blocks easily construct semantically meaningful foreground mattes, either in an unsupervised fashion, or based on small amount user input.

The goal of natural image denoising is to estimate a clean version given noisy image, utilizing prior knowledge on the statistics images. problem has been studied intensively with considerable progress made in recent years. However, it seems that algorithms are starting converge and improve over previous ones by only fractional dB values. It thus important understand how much more can we still what inherent limits imposed actual data. challenge evaluating such constructing proper models long...

Over the past decade, single image Super-Resolution (SR) research has focused on developing sophisticated priors, leading to significant advances. Estimating and incorporating blur model, that relates high-res low-res images, received much less attention, however. In particular, reconstruction constraint, namely blurred down sampled output should approximately equal input image, been either ignored or applied with default fixed models. this work, we examine relative importance of prior...

Given a collection of images (matrices) representing "class" objects we present method for extracting the commonalities image space directly from matrix representations (rather than vectorized representation which one would normally do in PCA approach, example). The general idea is to consider matrices as tensor and look an approximation its tensor-rank. tensor-rank designed such that SVD decomposition emerges special case where all input are repeatition single matrix. We evaluate coding...

Object motion during camera exposure often leads to noticeable blurring artifacts. Proper elimination of this blur is challenging because the kernel unknown, varies over image as a function object velocity, and destroys high frequencies. In case motions along 1D direction (e.g. horizontal) we show that these challenges can be addressed using moves exposure. Through analysis space-time integration, parabolic integration (corresponding constant sensor acceleration) invariant velocity. Thus,...

Various applications such as mosaicing and object insertion require stitching of image parts. The quality is measured visually by the similarity stitched to each input images, visibility seam between images. In order define get best possible stitching, we introduce several formal cost functions for evaluation quality. these images are defined in gradient domain, minimizing disturbing edges along seam. A good will optimize functions, overcoming both photometric inconsistencies geometric...

Acquiring and representing the 4D space of rays in world (the light field) is important for many computer vision graphics applications. Yet, field acquisition costly due to their high dimensionality. Existing approaches either capture explicitly, or involve an error-sensitive depth estimation process. This paper argues that fundamental difference between different rendering techniques a prior assumptions on field. We use previously reported dimensionality gap spectrum propose new prior. The...

Translucent materials are ubiquitous, and simulating their appearance requires accurate physical parameters. However, physically-accurate parameters for scattering difficult to acquire. We introduce an optimization framework measuring bulk properties of homogeneous (phase function, coefficient, absorption coefficient) that is more accurate, applicable a broad range materials. The combines stochastic gradient descent with Monte Carlo rendering material dictionary invert the radiative transfer...

Depth of field (DOF), the range scene depths that appear sharp in a photograph, poses fundamental tradeoff photography---wide apertures are important to reduce imaging noise, but they also increase defocus blur. Recent advances computational modify acquisition process extend DOF through deconvolution. Because deconvolution quality is tight function frequency power spectrum kernel, designs with high spectra desirable. In this paper we study how design effective extended-DOF systems, and show...