Image-Based Rendering from a Single Image

نویسندگان

  • Samuel Boivin
  • André Gagalowicz
چکیده

In this report, we present a new method to recover an approximation of the bidirectional re ectance distribution function (BRDF) of the surfaces present in a real scene. This is done from a single photograph and a 3D geometric model of the scene. The result is a full model of the re ectance properties of all surfaces, which can be rendered under novel illumination conditions, with, for example, viewpoint modi cation and the addition of new synthetic objects. Our technique produces a re ectance model using a small number of parameters. Using a new hierarchical algorithm assumptions are done for each surface about their BRDF, which becomes more and more complex during the rerendering iterations. We treat each type of BRDF separately, applying a speci c algorithm to take into account its own characteristics and its parameters. The number of parameters varies with the complexity of the model. When an approximation of the BRDF of all surfaces has been found, high quality rerenderings are produced using a global illumination software. We present several synthetic images that are compared to the original ones, as well as some possible applications in augmented reality. Key-words: Computer Graphics, Computer Vision, Illumination, Image-Based Rendering, Radiosity, Ray Tracing, Re ectance Recovery, Rendering, Inverse Rendering, Rerendering, BRDF, Augmented Reality Rendu Inverse à partir d'une Seule Image Réelle Résumé : Dans ce rapport de recherche, nous présentons une nouvelle méthode pour la reconstruction d'une approximation de la fonction de distribution de ré exion bidirectionnelle (BRDF). Les données nécessaires à une telle reconstruction sont décrites par une seule photographie de la scène et un modèle géométrique 3D de cette scène. Le résultat produit est un modèle complet des propriétés de ré exion de toutes les surfaces, incluant la possibilité de recalculer une nouvelle image synthétique, soit sous les conditions initiales, soit sous de nouvelles conditions (nouvel éclairement, nouveau point de vue, ajout/ablation d'objets, etc.). Notre technique emploie un modèle de BRDF simple, basé sur cinq paramètres au maximum, pour calculer les ré ectances des objets, qui deviennent de plus en plus complexes au fur et à mesure des itérations de rendu inverse. Chaque type de surface extraite de l'image réelle est donc traitée séparément, en appliquant un algorithme spéci que pour leur calculer des propriétés de ré ectance. Le nombre de paramètres de ré ectance varie avec la complexité de la BRDF choisie pour une surface. Lorsqu'une approximation de la BRDF de chaque surface de la scène a été trouvée, il est possible de calculer une nouvelle image synthétique en utilisant un logiciel d'illumination globale. Nous présentons ici plusieurs images synthétiques qui sont directement comparées aux images réelles, ainsi que des possibilités d'applications dans le domaine de la réalite augmentée. Mots-clés : Vision par Ordinateur, Synthèse d'Images, Illumination, Rendu à base d'Images Réelles, Radiosité, Lancer de Rayons, Recouvrement de Ré ectances, Rendu Réaliste, Rendu Inverse, BRDF, Réalité Augmentée Image-Based Rendering 3

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Segmentation Assisted Object Distinction for Direct Volume Rendering

Ray Casting is a direct volume rendering technique for visualizing 3D arrays of sampled data. It has vital applications in medical and biological imaging. Nevertheless, it is inherently open to cluttered classification results. It suffers from overlapping transfer function values and lacks a sufficiently powerful voxel parsing mechanism for object distinction. In this work, we are proposing an ...

متن کامل

Object-Oriented Method for Automatic Extraction of Road from High Resolution Satellite Images

As the information carried in a high spatial resolution image is not represented by single pixels but by meaningful image objects, which include the association of multiple pixels and their mutual relations, the object based method has become one of the most commonly used strategies for the processing of high resolution imagery. This processing comprises two fundamental and critical steps towar...

متن کامل

Single Image Dehazing Algorithm Based on Dark Channel Prior and Inverse Image

The sky regions of foggy image processed by all the existing conventional dehazing methods are degraded by color distortion and severe noise. This paper proposes an improved algorithm which combines dark channel prior and inverse image. We first invert the foggy image, and then estimate the transmission of the inverse image. At last, compared with the non-inversed transmission, the larger value...

متن کامل

Fusion of Panchromatic and Multispectral Images Using Non Subsampled Contourlet Transform and FFT Based Spectral Histogram (RESEARCH NOTE)

Image fusion is a method for obtaining a highly informative image by merging the relative information of an object obtained from two or more image sources of the same scene. The satellite cameras give a single band panchromatic (PAN) image with high spatial information and multispectral (MS) image with more spectral information. The problem exists today is either PAN or MS image is available fr...

متن کامل

A Deep Model for Super-resolution Enhancement from a Single Image

This study presents a method to reconstruct a high-resolution image using a deep convolution neural network. We propose a deep model, entitled Deep Block Super Resolution (DBSR), by fusing the output features of a deep convolutional network and a shallow convolutional network. In this way, our model benefits from high frequency and low frequency features extracted from deep and shallow networks...

متن کامل

Digital surface model extraction with high details using single high resolution satellite image and SRTM global DEM based on deep learning

The digital surface model (DSM) is an important product in the field of photogrammetry and remote sensing and has variety of applications in this field. Existed techniques require more than one image for DSM extraction and in this paper it is tried to investigate and analyze the probability of DSM extraction from a single satellite image. In this regard, an algorithm based on deep convolutional...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:

دوره   شماره 

صفحات  -

تاریخ انتشار 2001