Recent years have seen a surge of interest in multilayer neural networks fueled by their successful applications in numerous image processing and computer vision tasks. In this article, we describe a C++ implementation of the stochastic gradient descent to train a multilayer neural network, where a fast and accurate acceleration of tanh(.) is achieved with linear interpolation. As an example of application, we present a neural network able to deliver state-of-the-art performance in image demosaicing.

After four years of development of the Image Processing On Line journal (IPOL), this article presents a first analysis and overview of its scientific and technical development. The main issues met and overcome from the beginning of the journal are described with a focus on the purpose of the journal to establish a state of the art on the main Image Processing topics. The evolution of the online demonstration is also presented with a first analysis of author/publisher criticism, which led to a proposal for a new modular architecture of its demo system.

OverviewReferencesOnline Demo The PDE-Retinex Model The Algorithm

The integral image representation is a remarkable idea that permits to evaluate the sum ofimage values over rectangular regions of the image with four operations, regardless of the size ofthe region. It was first proposed under the name of summed area table in the computer graphicscommunity by Crow’84, in order to efficiently filter texture maps. It was later popularized in thecomputer vision community by Viola & Jones’04 with its use in their real-time object detectionframework. In this article we describe the integral image algorithm and study its application inthe context of block matching. We investigate tradeoffs and the limits of the performance gainwith respect to exhaustive block matching.

The integral image representation is a remarkable idea that permits to evaluate the sum of image values over rectangular regions of the image with four operations, regardless of the size of the region. It was first proposed under the name of summed area table in the computer graphics community by Crow'84, in order to efficiently filter texture maps. It was later popularized in the computer vision community by Viola & Jones'04 with its use in their real-time object detection framework. In this article we describe the integral image algorithm and study its application in the context of block matching. We investigate tradeoffs and the limits of the performance gain with respect to exhaustive block matching.

Presented at the UPF-DTIC Research Seminar 2014-01-09, Barcelona, Spain Software is nowadays essential for research as we use it to solve all sort of problems. Unlike commercial software, research software is not meant to be released as it's usually tailored to a concrete experimental setup. Yet, nobody doubts its importance as in many cases it embodies the scientific process itself. In the first part of my talk I'm going to share our experiences with IPOL: an on-line journal for image processing that follows the reproducible research model. The distinguishing characteristic of IPOL articles is that each article includes an on-line demo that permits to test the algorithm with new data and parameters, and an archive that stores all original test data used in the demo. This provides great insight into the algorithm's inner workings and limitations, facilitating the experimentation and leading to a stricter verification of the algorithms. In the second part of my talk I'm going to present a stereo pipeline for Pleiades satellite imagery (S2P). This is ongoing work being developed at ENS Cachan as part of a PhD thesis. The S2P pipeline automatically generates digital elevation models from quasi-simultaneous tristereo images acquired with the Pleiades satellite. The particularity of the S2P pipeline is that it permits to use conventional stereo tools developed for the pinhole camera geometry. Although the acquisition system does not fit the pinhole camera model, we show that the errors due to its assumption are negligible for the intended image sizes. Another compelling characteristic of this pipeline is that its code will soon be made available and that it can be tested as an on-line IPOL demo. -------------------------------------- Some slides courtesy of C. De Franchis, N. Limare and E. Meinhardt-Llopis: http://dx.doi.org/10.6084/m9.figshare.155649 http://dx.doi.org/10.6084/m9.figshare.855451

The integral image representation is a remarkable idea that permits to evaluate the sum of image values over rectangular regions of the image with four operations, regardless of the size of the region. It was first proposed under the name of summed area table in the computer graphics community by Crow’84, in order to efficiently filter texture maps. It was later popularized in the computer vision community by Viola & Jones’04 with its use in their real-time object detection framework. In this article we describe the integral image algorithm and study its application in the context of block matching. We investigate tradeoffs and the limits of the performance gain with respect to exhaustive block matching. Source Code The source code, the code documentation, and the online demo are accessible at the IPOL web part of this article1.

and poster for a presentation given at the SMAI 2011 Congress, May 23-27, Guidel, France.

Cette these est basee sur une etude des problemes de reproductibilite rencontres dans la recherche en traitement d'image. Nous avons concu, cree et developpe un journal scientifique, Image Processing On Line (IPOL), dans lequel les articles sont publies avec une implementation complete des algorithmes decrits, validee par les rapporteurs. Un service web de demonstration des algorithmes est joint aux articles, permettant de les tester sur donnees libres et de consulter l'historique des experiences precedentes. Nous proposons egalement une politique de droits d'auteur et licences, adaptee aux manuscrits et aux logiciels issus de la recherche, et des regles visant a guider les rapporteurs dans leur evaluation du logiciel. Le projet scientifique que constitue IPOL nous apparait tres benefique a la recherche en traitement d'image. L'examen detaille des implementations et les tests intensifs via le service web de demonstration ont permis de publier des articles de meilleure qualite. La frequentation d'IPOL montre que ce journal est utile au-dela de la communaute de ses auteurs, qui sont globalement satisfaits de leur experience et apprecient les avantages en terme de comprehension des algorithmes, de qualite des logiciels produits, de diffusion des travaux et d'opportunites de collaboration. Disposant de definitions claires des objets et methodes, et d'implementations validees, il devient possible de construire des chaines complexes et fiables de traitement des images.

With the journal Image Processing On Line (IPOL), we propose to promote software to the status of regular research material and subject it to the same treatment as research papers: it must be reviewed, it must be reusable and verifiable by the research community, it must follow style and quality guidelines. In IPOL, algorithms are published with their implementation, codes are peer-reviewed, and a web-based test interface is attached to each of these articles. This results in more software released by the researchers, a better software quality achieved with the review process, and a large collection of test data gathered for each article. IPOL has been active since 2010, and has already published thirty articles.

In 1964 Edwin H. Land formulated the Retinex theory, the first attempt to simulate and explain how the human visual system perceives color. Unfortunately, the Retinex Land-McCann original algorithm is both complex and not fully specified. Indeed, this algorithm computes at each pixel an average of a very large set of paths on the image. For this reason, Retinex has received several interpretations and implementations which, among other aims, attempt to tune down its excessive complexity. But, Morel et al. have shown that the original Retinex algorithm can be formalized as a (discrete) partial differential equation. This article describes the PDE-Retinex, a fast implementation of the Land-McCann original theory using only two DFT’s.

In this paper we present the simplest possible color balance algorithm. The assumption underlying this algorithm is that the highest values of R, G, B observed in the image must correspond to white, and the lowest values to obscurity. The algorithm simply stretches, as much as it can, the values of the three channels Red, Green, Blue (R, G, B), so that they occupy the maximal possible range [0, 255] by applying an affine transform ax+b to each channel. Since many images contain a few aberrant pixels that already occupy the 0 and 255 values, the proposed method saturates a small percentage of the pixels with the highest values to 255 and a small percentage of the pixels with the lowest values to 0, before applying the affine transform.

Image Processing On Line (IPOL) publishes image processing and image analysis algorithms, described in accurate literary form, coupled with code. It allows scientists to check directly the published algorithms on line by providing a web execution interface on any uploaded image.This installation acts the universality of image science. It permits to transcend the artificial segmentation of the research community in groups using this or that image software, or working on dedicated incompatible image formats. It promotes reproducible research, and the establishment of a state of the art verifiable by all, and on any image.This paper describes the technical challenges raised by the foundation of this new kind of journal and its scientific evaluation issues. It finally analyzes the first publications, to demonstrate its potential impact on the development of image science.