Paul L. Rosin

Registration of 3D Point Clouds and Meshes: A Survey from Rigid to Nonrigid

Three-dimensional surface registration transforms multiple three-dimensional data sets into the same coordinate system so as to align overlapping components of these sets. Recent surveys have covered different aspects of either rigid or nonrigid registration, but seldom discuss them as a whole. Our study serves two purposes: 1) To give a comprehensive survey of both types of registration, focusing on three-dimensional point clouds and meshes and 2) to provide a better understanding of registration from the perspective of data fitting. Registration is closely related to data fitting in which it comprises three core interwoven components: model selection, correspondences and constraints, and optimization. Study of these components 1) provides a basis for comparison of the novelties of different techniques, 2) reveals the similarity of rigid and nonrigid registration in terms of problem representations, and 3) shows how overfitting arises in nonrigid registration and the reasons for increasing interest in intrinsic techniques. We further summarize some practical issues of registration which include initializations and evaluations, and discuss some of our own observations, insights and foreseeable research trends.

Measuring shape: ellipticity, rectangularity, and triangularity

Abstract. Object classification often operates by making decisions based on the values of several shape properties measured from an image of the object. This paper describes several algorithms (both old and new) for calculating ellipticity, rectangularity, and triangularity shape descriptors. The methods are evaluated by testing on both synthetic and real data.

Segmentation of edges into lines and arcs

Abstract A long standing problem in computer vision is the extraction of meaningful features from images. This paper describes a method of segmenting curves in images into a combination of circular arcs and straight lines. This uses a recursive algorithm that first analyses lists of connected edge points and finds a polygonal description, and then analyses this description fitting arcs to groups of connected lines. The result is a description of image edges consisting of circular arcs and lines. The algorithm uses no thresholding. Instead the best option is chosen at each decision stage.

Multi view image surveillance and tracking

The paper presents a set of methods for multi view image tracking using a set of calibrated cameras. We demonstrate how effective the approach is for resolving occlusions and tracking objects between overlapping and non-overlapping camera views. Moving objects are initially detected using background subtraction. Temporal alignment is then performed between each video sequence in order to compensate for the different processing rates of each camera. A Kalman filter is used to track each object in 3D world coordinates and 2D image coordinates. Information is shared between the 2D/3D trackers of each camera view in order to improve the performance of object tracking and trajectory prediction. The system is shown to be robust in resolving dynamic and static object occlusions. Results are presented from a variety of outdoor surveillance video sequences.

Fast and Effective Feature-Preserving Mesh Denoising

We present a simple and fast mesh denoising method, which can remove noise effectively while preserving mesh features such as sharp edges and corners. The method consists of two stages. First, noisy face normals are filtered iteratively by weighted averaging of neighboring face normals. Second, vertex positions are iteratively updated to agree with the denoised face normals. The weight function used during normal filtering is much simpler than that used in previous similar approaches, being simply a trimmed quadratic. This makes the algorithm both fast and simple to implement. Vertex position updating is based on the integration of surface normals using a least-squares error criterion. Like previous algorithms, we solve the least-squares problem by gradient descent; whereas previous methods needed user input to determine the iteration step size, we determine it automatically. In addition, we prove the convergence of the vertex position updating approach. Analysis and experiments show the advantages of our proposed method over various earlier surface denoising methods.

A simple method for detecting salient regions

A simple method for detecting salient regions in images is proposed. It requires only edge detection, threshold decomposition, the distance transform, and thresholding. Moreover, it avoids the need for setting any parameter values. Experiments show that the resulting regions are relatively coarse, but overall the method is surprisingly effective, and has the benefit of easy implementation. Quantitative tests were carried out on Liu et al.s dataset of 5000 images. Although the ratings of our simple method were not as good as their approach which involved an extensive training stage, they were comparable to several other popular methods from the literature. Further tests on Kootstra and Schomakers dataset of 99 images also showed promising results.

Monitoring landslides from optical remotely sensed imagery: the case history of Tessina landslide, Italy

Collecting information on landslide occurrence and activity over wide areas is a crucial task for landslide hazard assessment. Field techniques, despite being very precise, are usually not sufficient to achieve this goal, since they mostly provide pointbased measurements. Mainly because of its synoptic view and its capability for repetitive observations, optical (visible-infrared) remotely sensed imagery acquired at different dates and at high spatial resolution can be considered as an effective complementary tool for field techniques to derive such information. An image-processing method to map and monitor landslide activity using multitemporal optical imagery is proposed. The method entails automatic change detection of suitably pre-processed (geometrically registered and radiometrically normalised) sequential images, followed by thresholding into landslide-related change pixels. Subsequent filtering based on the degree of rectangularity of regions can also be considered to eliminate pixel clusters corresponding to man-made land use changes. The application of this method is illustrated in the complex Tessina landslide in the Eastern Italian Alps. It has focused on discriminating the effects of a major reactivation that occurred in 1992, hence inferring the dynamics of the landslide at that time. Although the method has been devised for optical remote sensing imagery in general, in the absence of high-resolution satellite imagery covering that period, digital images derived by scanning existing aerial photograph diapositives at 1-m pixel size have been used. The method is able to classify image pixels according to landslide activity conditions. D 2003 Elsevier Science B.V. All rights reserved.

Fast mesh segmentation using random walks

3D mesh models are now widely available for use in various applications. The demand for automatic model analysis and understanding is ever increasing. Mesh segmentation is an important step towards model understanding, and acts as a useful tool for different mesh processing applications, e.g. reverse engineering and modeling by example. We extend a random walk method used previously for image segmentation to give algorithms for both interactive and automatic mesh segmentation. This method is extremely efficient, and scales almost linearly with increasing number of faces. For models of moderate size, interactive performance is achieved with commodity PCs. It is easy-to-implement, robust to noise in the mesh, and yields results suitable for downstream applications for both graphical and engineering models.

A Hu moment invariant as a shape circularity measure

In this paper we propose a new circularity measure which defines the degree to which a shape differs from a perfect circle. The new measure is easy to compute and, being area based, is robust-e.g., with respect to noise or narrow intrusions. Also, it satisfies the following desirable properties:*it ranges over (0,1] and gives the measured circularity equal to 1 if and only if the measured shape is a circle; *it is invariant with respect to translations, rotations and scaling. Compared with the most standard circularity measure, which considers the relation between the shape area and the shape perimeter, the new measure performs better in the case of shapes with boundary defects (which lead to a large increase in perimeter) and in the case of compound shapes. In contrast to the standard circularity measure, the new measure depends on the mutual position of the components inside a compound shape. Also, the new measure performs consistently in the case of shapes with very small (i.e., close to zero) measured circularity. It turns out that such a property enables the new measure to measure the linearity of shapes. In addition, we propose a generalisation of the new measure so that shape circularity can be computed while controlling the impact of the relative position of points inside the shape. An additional advantage of the generalised measure is that it can be used for detecting small irregularities in nearly circular shapes damaged by noise or during an extraction process in a particular image processing task.

Robust segmentation of intraretinal layers in the normal human fovea using a novel statistical model based on texture and shape analysis

A novel statistical model based on texture and shape for fully automatic intraretinal layer segmentation of normal retinal tomograms obtained by a commercial 800nm optical coherence tomography (OCT) system is developed. While existing algorithms often fail dramatically due to strong speckle noise, non-optimal imaging conditions, shadows and other artefacts, the novel algorithms accuracy only slowly deteriorates when progressively increasing segmentation task difficulty. Evaluation against a large set of manual segmentations shows unprecedented robustness, even in the presence of additional strong speckle noise, with dynamic range tested down to 12dB, enabling segmentation of almost all intraretinal layers in cases previously inaccessible to the existing algorithms. For the first time, an error measure is computed from a large, representative manually segmented data set (466 B-scans from 17 eyes, segmented twice by different operators) and compared to the automatic segmentation with a difference of only 2.6% against the inter-observer variability.