Yair Weiss

Colorization using optimization

Colorization is a computer-assisted process of adding color to a monochrome image or movie. The process typically involves segmenting images into regions and tracking these regions across image sequences. Neither of these tasks can be performed reliably in practice; consequently, colorization requires considerable user intervention and remains a tedious, time-consuming, and expensive task.In this paper we present a simple colorization method that requires neither precise image segmentation, nor accurate region tracking. Our method is based on a simple premise; neighboring pixels in space-time that have similar intensities should have similar colors. We formalize this premise using a quadratic cost function and obtain an optimization problem that can be solved efficiently using standard techniques. In our approach an artist only needs to annotate the image with a few color scribbles, and the indicated colors are automatically propagated in both space and time to produce a fully colorized image or sequence. We demonstrate that high quality colorizations of stills and movie clips may be obtained from a relatively modest amount of user input.

Segmentation using eigenvectors: a unifying view

Automatic grouping and segmentation of images remains a challenging problem in computer vision. Recently, a number of authors have demonstrated good performance on this task using methods that are based on eigenvectors of the affinity matrix. These approaches are extremely attractive in that they are based on simple eigendecomposition algorithms whose stability is well understood. Nevertheless, the use of eigendecompositions in the context of segmentation is far from well understood. In this paper we give a unified treatment of these algorithms, and show the close connections between them while highlighting their distinguishing features. We then prove results on eigenvectors of block matrices that allow us to analyze the performance of these algorithms in simple grouping settings. Finally, we use our analysis to motivate a variation on the existing methods that combines aspects from different eigenvector segmentation algorithms. We illustrate our analysis with results on real and synthetic images.

Understanding and evaluating blind deconvolution algorithms

Blind deconvolution is the recovery of a sharp version of a blurred image when the blur kernel is unknown. Recent algorithms have afforded dramatic progress, yet many aspects of the problem remain challenging and hard to understand. The goal of this paper is to analyze and evaluate recent blind deconvolution algorithms both theoretically and experimentally. We explain the previously reported failure of the naive MAP approach by demonstrating that it mostly favors no-blur explanations. On the other hand we show that since the kernel size is often smaller than the image size a MAP estimation of the kernel alone can be well constrained and accurately recover the true blur. The plethora of recent deconvolution techniques makes an experimental evaluation on ground-truth data important. We have collected blur data with ground truth and compared recent algorithms under equal settings. Additionally, our data demonstrates that the shift-invariant blur assumption made by most algorithms is often violated.

Deriving intrinsic images from image sequences

Intrinsic images are a useful midlevel description of scenes proposed by H.G. Barrow and J.M. Tenenbaum (1978). An image is de-composed into two images: a reflectance image and an illumination image. Finding such a decomposition remains a difficult problem in computer vision. We focus on a slightly, easier problem: given a sequence of T images where the reflectance is constant and the illumination changes, can we recover T illumination images and a single reflectance image? We show that this problem is still imposed and suggest approaching it as a maximum-likelihood estimation problem. Following recent work on the statistics of natural images, we use a prior that assumes that illumination images will give rise to sparse filter outputs. We show that this leads to a simple, novel algorithm for recovering reflectance images. We illustrate the algorithms performance on real and synthetic image sequences.

From learning models of natural image patches to whole image restoration

Learning good image priors is of utmost importance for the study of vision, computer vision and image processing applications. Learning priors and optimizing over whole images can lead to tremendous computational challenges. In contrast, when we work with small image patches, it is possible to learn priors and perform patch restoration very efficiently. This raises three questions - do priors that give high likelihood to the data also lead to good performance in restoration? Can we use such patch based priors to restore a full image? Can we learn better patch priors? In this work we answer these questions. We compare the likelihood of several patch models and show that priors that give high likelihood to data perform better in patch restoration. Motivated by this result, we propose a generic framework which allows for whole image restoration using any patch based prior for which a MAP (or approximate MAP) estimate can be calculated. We show how to derive an appropriate cost function, how to optimize it and how to use it to restore whole images. Finally, we present a generic, surprisingly simple Gaussian Mixture prior, learned from a set of natural images. When used with the proposed framework, this Gaussian Mixture Model outperforms all other generic prior methods for image denoising, deblurring and inpainting.

Correctness of Belief Propagation in Gaussian Graphical Models of Arbitrary Topology

Graphical models, such as Bayesian networks and Markov random fields, represent statistical dependencies of variables by a graph. Local belief propagation rules of the sort proposed by Pearl (1988) are guaranteed to converge to the correct posterior probabilities in singly connected graphs. Recently, good performance has been obtained by using these same rules on graphs with loops, a method we refer to as loopy belief propagation. Perhaps the most dramatic instance is the near Shannon-limit performance of Turbo codes, whose decoding algorithm is equivalent to loopy propagation. Except for the case of graphs with a single loop, there has been little theoretical understanding of loopy propagation. Here we analyze belief propagation in networks with arbitrary topologies when the nodes in the graph describe jointly gaussian random variables. We give an analytical formula relating the true posterior probabilities with those calculated using loopy propagation. We give sufficient conditions for convergence and show that when belief propagation converges, it gives the correct posterior means for all graph topologies, not just networks with a single loop. These results motivate using the powerful belief propagation algorithm in a broader class of networks and help clarify the empirical performance results.

Correctness of Local Probability Propagation in Graphical Models with Loops

Graphical models, such as Bayesian networks and Markov networks, represent joint distributions over a set of variables by means of a graph. When the graph is singly connected, local propagation rules of the sort proposed by Pearl (1988) are guaranteed to converge to the correct posterior probabilities. Recently a number of researchers have empirically demonstrated good performance of these same local propagation schemes on graphs with loops, but a theoretical understanding of this performance has yet to be achieved. For graphical models with a single loop, we derive an analytical relationship between the probabilities computed using local propagation and the correct marginals. Using this relationship we show a category of graphical models with loops for which local propagation gives rise to provably optimal maximum a posteriori assignments (although the computed marginals will be incorrect). We also show how nodes can use local information in the messages they receive in order to correct their computed marginals. We discuss how these results can be extended to graphical models with multiple loops and show simulation results suggesting that some properties of propagation on single-loop graphs may hold for a larger class of graphs. Specifically we discuss the implication of our results for understanding a class of recently proposed error-correcting codes known as turbo codes.

Seamless Image Stitching in the Gradient Domain

Image stitching is used to combine several individual images having some overlap into a composite image. The quality of image stitching is measured by the similarity of the stitched image to each of the input images, and by the visibility of the seam between the stitched images.

Learning object detection from a small number of examples: the importance of good features

Face detection systems have recently achieved high detection rates and real-time performance. However, these methods usually rely on a huge training database (around 5,000 positive examples for good performance). While such huge databases may be feasible for building a system that detects a single object, it is obviously problematic for scenarios where multiple objects (or multiple views of a single object) need to be detected. Indeed, even for multi-viewface detection the performance of existing systems is far from satisfactory. In this work we focus on the problem of learning to detect objects from a small training database. We show that performance depends crucially on the features that are used to represent the objects. Specifically, we show that using local edge orientation histograms (EOH) as features can significantly improve performance compared to the standard linear features used in existing systems. For frontal faces, local orientation histograms enable state of the art performance using only a few hundred training examples. For profile view faces, local orientation histograms enable learning a system that seems to outperform the state of the art in real-time systems even with a small number of training examples.

Efficient marginal likelihood optimization in blind deconvolution

In blind deconvolution one aims to estimate from an input blurred image y a sharp image x and an unknown blur kernel k. Recent research shows that a key to success is to consider the overall shape of the posterior distribution p(x, k\y) and not only its mode. This leads to a distinction between MAPx, k strategies which estimate the mode pair x, k and often lead to undesired results, and MAPk strategies which select the best k while marginalizing over all possible x images. The MAPk principle is significantly more robust than the MAPx, k one, yet, it involves a challenging marginalization over latent images. As a result, MAPk techniques are considered complicated, and have not been widely exploited. This paper derives a simple approximated MAPk algorithm which involves only a modest modification of common MAPx, k algorithms. We show that MAPk can, in fact, be optimized easily, with no additional computational complexity.