Richard Lepage

Knowledge-Based Image Understanding Systems

The development of software that would be to image understanding systems what expert system shells are to expert systems has been the subject of considerable enquiry over the last ten years: this paper reviews pertinent publications and tries to present a coherent view of the field. After a survey of the advantages of explicit knowledge representation in image understanding, we tackle the subject under two main headings. We first expose the nature of the knowledge that the various authors have represented for image understanding. To this effect, we have elaborated a knowledge taxonomy consisting of seven modules, ranging in specificity from task domain knowledge to generic knowledge about the use of software systems. We then examine how researchers have represented these various kinds of knowledge. Most of the representations known to artificial intelligence were pressed into service, and a discussion of their relative merits is presented.

Characterizing human shape variation using 3D anthropometric data

Characterizing the variations of the human body shape is fundamentally important in many applications ranging from animation to product design. 3D scanning technology makes it possible to digitize the complete surfaces of a large number of human bodies, providing much richer information about the body shape than traditional anthropometric measurements. This technology opens up opportunities to extract new measurements for quantifying the body shape. In this paper, we present a new method for extracting the main modes of variations of the human shape from a 3D anthropometric database. Previous approaches rely on anatomical landmarks. Using a volumetric representation, we show that human shape analysis can be performed despite the lack of such information. We first introduce a technique for repairing the 3D models from the original scans. Principal component analysis analysis is then applied to the volumetric description of a set of human models to extract dominant components of shape variability for a target population. We demonstrate a good reconstruction of the original models from a reduced number of components. Finally, we provide tools for visualizing the main modes of human shape variation.

CAD-based range sensor placement for optimum 3D data acquisition

The use of laser range sensor allows a very significant improvement in acquisition speed but does not equal the accuracy obtained with a coordinate measuring machine. In order to obtain a quality control close to that obtained in metrology we suggest improving the accuracy of the depth measurements by positioning the sensors head according to a strategy for optimum 3D data acquisition. We propose such a strategy to automatically produce a sensing plan for completely and accurately acquiring the geometry of a surface or of the whole piece whenever possible. The system requires the exact position and orientation of the part and its CAD model in IGES format. There is no limitation regarding the shape of the part to be digitized. An auto-synchronized range sensor developed at the NRCC was used, and for this sensor, the accuracy of the 3D measured points is a function of the distance and of the incident angle relative to the surface. Our strategy guarantees that the viewpoints found meet the best accuracy conditions in the scanning process.

A CAD-based 3D data acquisition strategy for inspection

Abstract.The use of a laser range sensor in the 3D digitalization process allows significant improvement in acquisition speed and in 3D measurement point density. However, if we want to use these 3D data in applications that require data with a high degree of accuracy like inspection tasks, it is mandatory that the 3D points be acquired under the best conditions of accuracy. During 3D capture of a part, several sources of error can alter the measured values. Thus we must find and model the most important parameters affecting the accuracy of the range sensor. This error model, along with the CAD model of the part, is used to produce a sensing plan to completely and accurately acquire the geometry of the part. The sensing plan is comprised of the set of viewpoints that defines the exact position and orientation of the camera relative to the part. There is no limitation with regard to the shape of the part to be digitalized. An autosynchronized range sensor fixed on a coordinate measuring machine was used. For this sensor, the accuracy of the 3D measured points is a function of the distance and of the angle of incidence relative to the surface. The strategy proposed to find the acquisition plan guarantees that the viewpoints meet the best accuracy conditions in the scanning process, solving the occlusion problems. It was found that the 3D data acquired by using the proposed strategy are around 30% more accurate than the 3D data obtained in a standard acquisition.

Extracting main modes of human body shape variation from 3D anthropometric data

Characterizing the variations of the human body shape is fundamentally important to many applications ranging from animation to product design. 3D scanning technology makes it possible to digitize the complete surfaces of a large number of human bodies, providing much richer information about the body shape than the traditional anthropometric measurements. This technology opens up opportunities to extract new measurements for quantifying the body shape. Using the data from the first large scale 3D anthropometric survey, the CAESAR project, we demonstrate that the human body shape can be represented by a small number of principal components. Principal component analysis extracts orthogonal basis vectors, called eigenpersons, from the space of body shapes. The shape of any individual person can then be expressed by the linear combination of the basis vectors. We demonstrate that some of these components correspond to the commonly used body measurements like height and weight and others indicate new ways of charactering body shape variations. We develop tools to visualize the changes of the body shape along the main components. These tools help understand the meaningful components of the human body shape.

Road Extraction From Very High Resolution Remote Sensing Optical Images Based on Texture Analysis and Beamlet Transform

Road extraction from very high resolution sensors is a very popular topic in panchromatic and multispectral remote sensing image analysis. Despite the vast number of methods proposed in the literature to deal with this problem, in practice, most are quite limited and do not account for geometric and radiometric variability. Our aim is to propose a novel road extraction approach able to efficiently extract roads and reduce computation time using texture analysis and multiscale reasoning based on the beamlet transform. The proposed methodology consists of two stages: 1) road edge candidate selection and 2) multiscale reasoning with the beamlet transform. In the first step, mathematical morphology is applied to distinguish rectilinear structures, and road edge candidates are identified using the Canny edge detector. In the second phase, multiscale reasoning using the beamlet transform allows local and global information to be combined. Global information is introduced to distinguish main road axes at coarser scales, and local segments in finer scales, which are aggregated to reconstruct the road network. Rules based on the spatial relationships between segments belonging to different levels of resolution are also introduced at this stage. The experiments are performed based on the images acquired from the city of Port-au-Prince in Haiti during the earthquake of January 2010. The results demonstrate the accuracy and efficiency of our algorithm.

Food Grading/Sorting Based on Color Appearance trough Machine Vision: the Case of Fresh Cranberries

Food quality depends mostly upon external appearance and in particular upon color. The purpose of this paper is to show a non destructive method for fresh cranberries color evaluation. A fresh cranberry color scale is built-up by machine vision system. Each cranberries batch to be evaluated is first photographed. The color which appears at the surface of the cranberry is not uniform and half of the surface is hidden to the camera, so a few pictures of the fruit samples are taken out, mixing the container after each shot. The individual cranberries are segmented from the background and the mean cranberry color is computed for each cranberry. The statistical validity of this multiple replicate method is first examined by an analysis of variance-one-way. The cranberries color is converted in the CIELAB color space, which is close to human perception. Then a cranberry color scale is built-up by performing first a principal component analysis on the dataset, and then by a robust data fitting method

Learning and extracting edges from images by a modified Hopfield neural network

This paper introduces a modified unsupervised Hopfield network that can learn the underlying process in an edge detection task from grey level images. After the learning phase, the network performance is tested to ensure that it can detect only the significant edges and that it can generalise other images.

Inspection of 3D parts using high accuracy range data

The use of a laser range sensor in the 3D part digitalization process for inspection tasks allows very significant improvement in acquisition speed and in 3D measurement points density but does not equal the accuracy obtained with a coordinate measuring machine (CMM). Inspection consists in verifying the accuracy of a part related to a given set of tolerances. It is thus necessary that the 3D measurements be accurate. In the 3D capture of a part, several sources of error can alter the measured values. So, we have to find and model the most influent parameters affecting the accuracy of the range sensor in the digitalization process. This model is used to produce a sensing plan to acquire completely and accurately the geometry of a part. The sensing plan is composed of the set of viewpoints which defines the exact position and orientation of the camera relative to the part. The 3D cloud obtained from the sensing plan is registered with the CAD model of the part and then segmented according to the different surfaces. Segmentation results are used to check tolerances of the part. By using the noise model, we introduce a dispersion value for each 3D point acquired according to the sensing plan. This value of dispersion is shown as a weight factor in the inspection results.

Fast and Efficient Evaluation of Building Damage From Very High Resolution Optical Satellite Images

In this paper, we present a novel combination of object features to both match buildings from predisaster images to shapes in a postdisaster image and assess damage on those buildings. These features include scale profile ratios extracted from a tree of shapes representation of the original image as well as texture features. A supervised classifier is used to classify building damage into three representative classes tied to the European Macroseismic Scale (EMS-98). The method is compared to visual inspection results as well as other automated methods. Results clearly show the benefits of our method for fast crisis mapping applications with few human inputs required.