Search

National audience; Nous proposons un nouveau modèle d'inpainting vidéo pour la restauration de films, qui combine la reconstruction de la structure par une méthode de diffusion et la reconstruction de la texture par une méthode de recopie de patchs. Les énergies proposées pour chacune de ces deux méthodes sont minimisées alternativement, afin de préser-ver la structure globale de l'image tout en affinant sa texture. Alors que la reconstruction de la structure est effec-tuée conjointement à l'estimation du mouvement par flux optique via plusieurs approches proximales, la reconstruction de la texture est traitée par une approche variation-nelle non locale (NL-means). Les résultats sur différentes séquences d'images de la base de données Middlebury et de la Cinémathèque de Toulouse montrent une améliora-tion dans la qualité des reconstructions. Mots Clef Inpainting vidéo, flux optique, méthodes variationnelles. Abstract We propose a new video inpainting model for movies restoration application. This model combines structural reconstruction with a diffusion-based method and textural reconstruction with a patch-based method. Both proposed energies (one for each method) are alternatively minimized in order to preserve the overall structure adding textu-ral refinement. While the structural reconstruction is obtained jointly with optical flow computation thanks to several proximal approaches, the textural reconstruction is processed by a variational non-local approach (NL-means). Results on different frames from the Middlebury database and from the Cinémathèque de Toulouse show quality improvement in the reconstructions.

International audience; Colorization of gray-scale images relies on prior color information. Exemplar-based methods use a color image as source of such information. Then the colors of the source image are transferred to the gray-scale target image. In the literature, this transfer is mainly guided by texture descriptors. Face images usually contain few texture so that the common approaches frequently fail. In this work, we propose a new method taking the geometric structure of the images rather their texture into account such that it is more reliable for faces. Our approach is based on image morphing and relies on the YUV color space. First, a correspondence mapping between the luminance Y channel of the color source image and the gray-scale target image is computed. This mapping is based on the time discrete metamorphosis model suggested by Berkels, Effland and Rumpf. We provide a new finite difference approach for the numerical computation of the mapping. Then, the chrominance U,V channels of the source image are transferred via this correspondence map to the target image. A possible postprocessing step by a variational model is developed to further improve the results. To keep the contrast special attention is paid to make the postprocessing unbiased. Our numerical experiments show that our morphing based approach clearly outperforms state-of-the-art methods.

International audience; – De nombreuses techniques de débrumage, dehazing en anglais, sont présentées dans la littérature. Leur objectif est d'améliorer le rendu des photographies, notamment des paysages, prises lors de condition météorologique défavorables. Afin d'éviter d'inverser le modèle de Koschmieder pour la formation de la brume, nous utilisons un modèle variationnel de rehaussement de contraste, pour lequel nous proposons un algorithme itératif. Dans ce papier, nous montrons que l'implémentation d'un algorithme rapide de rehaussement de contraste passe par la recherche de polynômes bornés. La méthode retenue permet d'obtenir un algorithme efficace, naturellement stable, avec des résultats réalistes. Abstract – Many dehazing methods are proposed in the literature with the aim of improving the rendering of pictures, such as landscapes. In order to avoid the inversion of Koschmieder's model for the hazed images, a variational model for image enhancement is used, solved by an iterative algorithm. In this paper, we show that the design of a fast algorithm for image enhancement needs bounded polynomials. The retained method performs an efficient algorithm, naturally stable, with realistic results.

International audience; This paper addresses the topic of image colorization that consists in converting a gray-scale image into a color one. In the literature, there exist two main types of approaches to tackle this problem. The first one is the manual methods where the color information is given by ome scribbles drawn by the user on the image. The interest of these approaches comes from the interactions with the user that can put any color he wants. Nevertheless, when the scene is complex many cribbles must be drawn and the interactive process becomes tedious and time-consuming. The second category of approaches is the emplar-based methods that require a color image as input. Once the example image is given, the colorization is generally fully automatic. A limitation of these methods is that the example image needs to contain all the desired colors in the final result. In this paper, we propose a new framework that unifies these two categories of approaches into a joint variational model. Our approach is able to take into account information coming from any colorization method among these two categories. Experiments and comparisons emonstrate that the proposed approach provides competitive colorization results compared to state-of-the-art methods.

This work proposes a novel variational frame- work for exemplar-based image colorization in RGB while most of existing methods use a luminance-chro- minance space. Using directly these three RGB chan- nels, our model does not need any pre or post-pro- cessing step. We design a primal-dual like algorithm that solves the proposed variational approach. The constraint when colorizing a gray-scale image is to maintain its luminance. In our RGB model this constraint is added in the variational model. Its preservation requires introducing a novel projection scheme that takes care of the hue value (H) of the HSI color-space. The behavior of our algorithm is illustrated on numerous examples, and a qualitative comparison is made with state-of-the-art methods.

International audience; – Cet article propose une approche originale pour construire une métrique optimale. La méthode décrite calcule une combinaison linéaire de métriques, afin de créer une nouvelle distance capable de mieux comparer des patchs. L'utilisateur fournit un ensemble d'images d'apprentissage, ainsi que les métriques de base. Le cadre théorique et un algorithme efficace sont décrits pour résoudre ce problème. Des expériences numériques en colorisation d'images, sont réalisées pour démontrer l'efficacité de cette nouvelle approche. Abstract – This paper proposes a novel approach to compute an optimal metric. This approach performs a linear combination of distances, in order to compute a new metric which can better compare patches. The user provides a set of training images, as well as the basic distances. A theoretical framework and an efficient algorithm are described to solve this problem. Numerical experiments in image colorization are performed in order to demonstrate the interest of this new approach.

International audience; Detection of defects is the first step in the restoration of oldmovies. The specificity of our work is to learn the expertiseof a film restorer through a pair of sequences consistingof a film with defects, and this same film after semi-manualrestoration with the help of a specialized software. In orderto detect defects with a minimum of human interaction, andto reduce the time spent on restoration, we feed a U-Netneural network with a sequence of defective images as in-put, in order to detect abnormally high spatio-temporal va-riations in pixel intensity. The output of the network beingthe defect mask, we create comparative masks using thedifferences between the defective and restored versions of the film (those used during restoration not being directlyaccessible). Our network manages to automatically detectreal defects more accurately than manual selections, oreven some defects missed by the restoration expert.; La détection des défauts constitue la première étape de la restauration de vieux films. La spécificité de notre travail est d'apprendre l'expertise d'un restaurateur de films au travers d'une paire de séquences constituée d'un film comportant des défauts, et de ce même film après restauration semi-manuelle aidée d'un logiciel spécialisé. Pour détecter les défauts avec un minimum d'interaction humaine, et réduire le temps consacré à la restauration, nous alimentons un réseau de neurones de type U-Net avec une séquence d'images défectueuses en entrée, afin de détecter les variations spatio-temporelles anormalement élevées de l'intensité des pixels. La sortie du réseau étant le masque des défauts, nous créons des masques comparatifs en utilisant les différences entre les versions défectueuse et restaurée du film (ceux utilisés lors de la restauration n'étant pas directement accessibles). Notre réseau réussit à détecter automatiquement les défauts réels plus précisément que les sélections manuelles, voire certains défauts oubliés par l'expert en restauration. Mots Clef Détection de défauts, vieux films, apprentissage profond.

International audience; This paper focuses on the denoising of chrominance channels of color images. We propose a variational framework involving TV reg-ularization that modifies the chrominance channel while preserving the input luminance of the image. The main issue of such a problem is to ensure that the denoised chrominance together with the original luminance belong to the RGB space after color format conversion. Standard methods of the literature simply truncate the converted RGB values, which lead to a change of hue in the denoised image. In order to tackle this issue, a " RGB compatible " chrominance range is defined on each pixel with respect to the input luminance. An algorithm to compute the orthogonal projection onto such a set is then introduced. Next, we propose to extend the CLEAR debiasing technique to avoid the loss of colourfulness produced by TV regularization. The benefits of our approach with respect to state-of-the-art methods are illustrated on several experiments.

Colorization of gray-scale images relies on prior color information. Exemplar-based methods use a color image as source of such information. Then the colors of the source image are transferred to the gray-scale target image. In the literature, this transfer is mainly guided by texture descriptors. Face images usually contain few texture so that the common approaches frequently fail. In this paper, we propose a new method taking the geometric structure of the images rather their texture into account such that it is more reliable for faces. Our approach is based on image morphing and relies on the YUV color space. First, a correspondence mapping between the luminance Y channel of the color source image and the gray-scale target image is computed. This mapping is based on the time discrete metamorphosis model suggested by Berkels, Effland and Rumpf. We provide a new finite difference approach for the numerical computation of the mapping. Then, the chrominance U,V channels of the source image are transferred via this correspondence map to the target image. A possible postprocessing step by a variational model is developed to further improve the results. To keep the contrast special attention is paid to make the postprocessing unbiased. Our numerical experiments show that our morphing based approach clearly outperforms state-of-the-art methods.

International audience; This paper deals with the difficult problem of video colorization. Methods in the literature are generally based on spatio-temporal video blocks, or on frame to frame color propagation methods, each technique having its own advantages and drawbacks. In this paper , we present both a novel automatic frame-to-frame propagation approach and an interactive correction method within a variational framework. The proposed method propagates colors from an initial colorized frame to the whole grayscale video sequence. The automatic propagation results may be visually unsuitable in some cases. To overcome this limitation, a spatio-temporal functional with a user-guided correction is introduced. Two primal-dual algorithms are designed to solve the proposed variational models. Numerical results show the efficiency and the potentiality of the proposed approach in comparison with state-of-the-art methods.

International audience; This paper deals with the problem of image colorization. A model including total variation regularization is proposed. Our approach colorizes directly the three RGB channels, while most existing methods were only focusing on the two chrominance channels. By using the three channels, our approach is able to better preserve color consistency. Our model is non convex, but we propose an efficient primal-dual like algorithm to compute a local minimizer. Numerical examples illustrate the good behavior of our algorithm with respect to state-of-the-art methods.

International audience; This paper provides a new method to colorize gray-scale images. While the computation of the luminance channel is directly performed by a linear transformation, the colorization process is an ill-posed problem that requires some priors. In the literature two classes of approach exist. The first class includes manual methods that need the user to manually add colors on the image to colorize. The second class includes exemplar-based approaches where a color image, with a similar semantic content, is provided as input to the method. These two types of priors have their own advantages and drawbacks. In this paper, a new variational framework for exemplar-based colorization is proposed. A nonlocal approach is used to find relevant color in the source image in order to suggest colors on the gray-scale image. The spatial coherency of the result as well as the final color selection is provided by a nonconvex variational framework based on a total variation. An efficient primal-dual algorithm is provided, and a proof of its convergence is proposed. In this work, we also extend the proposed exemplar-based approach to combine both exemplar-based and manual methods. It provides a single framework that unifies advantages of both approaches. Finally, experiments and comparisons with state-of-the-art methods illustrate the efficiency of our proposal. 1. Introduction. The colorization of a gray-scale image consists of adding color information to it. It is useful in the entertainment industry to make old productions more attractive. The reverse operation is based on perceptual assumptions and is today an active research area [28], [13], [37]. Colorization can also be used to add information in order to help further analysis of the image by a user (e.g., sensor fusion [43]). It can also be used for art restoration ; see, e.g., [17] or [41]. It is an old subject that began with the ability of screens and devices to display colors. A seminal approach consists in mapping each level of gray into a color-space [18]. Nevertheless, all colors cannot be recovered without an additional prior. In the existing approaches, priors can be added in two ways: with a direct addition of color on

International audience; Color image enhancement is a challenging task in digital imaging with many applications. This paper contributes to image enhancement methods. We propose a new variational model for color improvement in the RGB space based on a desired target intensity image. Our model improves the visual quality of the color image while it preserves the range and takes the hue of the original, badly exposed image into account without amplifying its color artifacts. To approximate the hue of the original image we use the fact that affine transforms are hue preserving. To cope with the noise in the color channels we design a particular coupled TV regularization term. Since the target intensity of the image is unaltered our model respects important image structures. Numerical results demonstrate the very good performance of our method.