Johannes D'Haeyer

Gaussian filtering of images: A regularization approach

Abstract Approximate Gaussian filtering of equidistant data can be obtained by regularizing the data with Tikhonovs second order stabilizing functionals. The correspondence between the resulting cubic spline functions and Gaussian functions was first shown by Poggio. The impulse response arising from cubic spline approximation is however not positive everywhere. As an alternative, approximating cubic splines under tension are considered in this paper and a fast implementation is proposed that requires the same amount of calculations for all values of the spread. The performance (approximation error versus computational efficiency) is compared with that of single- and multi-stage FIR Gaussian filter approximations.

Characterizing the performance of automatic road detection using error propagation

Abstract A methodology is introduced to predict the performance of automatic road detection using image examples of typical road types. In contrast to previous work on road detection, the focus is on characterizing the detection performance to achieve reliable performance measures of the detection. It is studied how noise, like road markings, shadows, trees and buildings, influences the detection of road. This noise is modeled using second-order statistics and its effects are calculated using error propagation on the detection equations. The method predicts the performance in terms of detection rate and gives the optimal parameter set that is needed for this detection. Experiments have been conducted on a set of images of typical roads in very high-resolution satellite images.

Relevance criteria for data mining using error-tolerant graph matching

In this paper we present a graph based approach for mining geospatial data. The system uses error-tolerant graph matching to find correspondences between the detected image information and the geospatial vector data. Spatial relations between objects are used to find a reliable object-to-object mapping. Graph matching is used as a flexible query mechanism to answer the spatial query. A condition based on the expected graph error has been presented which allows to determine the bounds of error tolerance and in this way characterizes the relevancy of a query solution. We show that the number of null labels is an important measure to determine relevancy. To be able to correctly interpret the matching results in terms of relevancy the derived bounds of error tolerance are essential.

Graph‐based change detection in geographic information using VHR satellite images

In this paper we examine a system based on computer vision for automated detection of change and anomalies in GIS road networks using very high resolution satellite images. The system consists of a low‐level feature detection process, which extracts road junctions, and a high‐level matching process, which uses graph matching to find correspondences between the detected image information and the road vector data. The matching process is based on continuous relaxation labelling. It is driven by spatial relations between the objects and takes into account different errors that can occur. The result is an object‐to‐object mapping between image and vector dataset. The mapping result can be used to calculate a rubbersheeting transformation which is able to compensate for local distortions. A measure of change is defined based on the number of null assignments. We show how combined with a condition to characterize acceptable errors, this measure is useful and reliable to characterize inconsistencies between image and vector data.

Context driven matching in structural pattern recognition

Publisher Summary This chapter examines the problem of structural pattern recognition using graph structures. To speed up the correspondence problem, the chapter proposes a histogram technique which characterizes the context of a primitive within a pattern and allows indexing in the model database with polynomial complexity. The chapter presents a new iterative matching technique based on a histogram of structural context information. Experiments show a good noise suppressing ability while retaining adequate recognition results with minimal false alarms. Because scene primitives are structurally mapped onto the model, orientation and scale can be hypothesized from a match, whereas the model can be used to direct a search for missing information, thereby improving or ignoring the match. This will be the subject for further work.

Reliable flaw classifiers for machine vision-based quality control

The paper presents a hybrid approach to the problem of classifier construction for machine vision based inspection systems. The method allows the user to integrate different types of classifiers and exploit different sources of information such as sample data and expert knowledge. Of particular interest is the problem of classification reliability in the case of small training sets. A novel forward fuzzy decision tree induction method is proposed to handle different types of uncertainty. The performance of the method is compared experimentally with other classifiers using artificial and machine vision data.

Robust statistics for automated quality assessment of road network data based on VHR images

A method is explored to assess the quality of road network data based on image information in a reliable and accurate way. In the field of geography, an accuracy assessment method, called buffer-overlay-statistics, is known to assess the spatial quality of a line data set by using another line data set of higher spatial accuracy. Here, the method is adapted to assess the quality of a line data set based on image information rather than vector data. The average displacement accuracy measure is redefined, such that it is able to take into account line detection errors (fragmentation and noise). Experiments are conducted on artificial data showing how road extraction out of very high resolution satellite images can be used to asses the spatial accuracy of an existing road vector database.

Road detection statistics for automated quality control of GIS data

We examine the use of road detection in VHR satellite images to automate the process of quality assessment of digital road network data. An important aspect is the emphasis on accuracy and reliability of the system. Although road detection has been studied for more than a decade, it is often difficult to assess what performance can be expected over datasets other than the images published. We propose a system to train a road detector based on image examples of typical roads. The system calculates the optimal detection parameters and estimates the performance over the dataset in terms of detection rate and degree of fragmentation. The methodology relies on error propagation and image statistics, and is generic in nature. By showing image examples, variations due to shadow, activity on the road, weather conditions etc. can be taken into account when estimating the expected performance

Automatic induction of relational models

The problem of shape recognition is studied through the use of relational models based on the hypergraph representation and the context similarity measure. Formal definitions are introduced and graph properties are calculated important to the matching process. A conflict is shown to exist between the interclass distance and the semantical distance between the vertices within a model. The representation is extended with the notion of vertex neighborhood, which increases the semantical distance and makes the processing of complex scenes feasible.

Modelling spatial correlation and image statistics for improved tracking of human gestures

In this paper, we examine sensor specific distributions of local image operators (edge and line detectors), which describe the appearance of people in video sequences. The distributions are used to describe a probabilistic articulated motion model to track the gestures of a person in terms of arms and body movement, which is solved using a particle filter. We focus on modeling the statistics of one sensor and examine the influence of image noise and scale, and the spatial accuracy that is obtainable. Additionally spatial correlation between pixels is modeled in the appearance model. We show that by neglecting the correlation high detection probabilities are quickly overestimated, which can often lead to false positives. Using the weighted geometric mean of pixel information leads to much improved results.