Similarity-based denoising of point-sampled surfaces

A non-local denoising (NLD) algorithm for point-sampled surfaces (PSSs) is presented based on similarities, including geometry intensity and features of sample points. By using the trilateral filtering operator, the differential signal of each sample point is determined and called "geometry intensity". Based on covariance analysis, a regular grid of geometry intensity of a sample point is constructed, and the geometry-intensity similarity of two points is measured according to their grids. Based on mean shift clustering, the PSSs are clustered in terms of the local geometry-features similarity. The smoothed geometry intensity, i.e., offset distance, of the sample point is estimated according to the two similarities. Using the resulting intensity, the noise component from PSSs is finally removed by adjusting the position of each sample point along its own normal direction. Ex- perimental results demonstrate that the algorithm is robust and can produce a more accurate denoising result while having better feature preservation.

Trimming self-intersections in swept volume solid modeling

Swept volume solid modeling has been applied to many areas such as NC machining simulation and verification, robot workspace analysis, collision detection, and CAD. But self-intersections continue to be a challenging problem in the boundary representation of swept volume solids. A novel algorithm is presented in this paper to trim self-intersection regions in swept volume solids modeling. This trimming algorithm consists of two major steps: (1) roughly detecting self-intersection regions by checking intersections or overlapping of the envelop profiles; (2) splitting the whole envelop surfaces of the swept volume solid into separate non-self-intersecting patches to trim global self-intersections, and to trim local self-intersections, dividing local self-intersecting regions into patches and replacing self-intersecting patches with non-self-intersecting ones. Examples show that our algorithm is efficient and robust.

Mesh sharpening via normal filtering

We present a robust mesh sharpening approach to reconstructing sharp features from blended or chamfered features, even with noise and aliasing errors. Feature regions were first recognized via normal variation according to the user's input, and then normal filtering was applied to faces of feature regions. Finally, the vertices of the feature region were gradually updated based on new face normals using a least-squares error criterion. Experimental results demonstrate that the method is effective and robust in sharpening meshes.

Image-based 3D model retrieval using manifold learning

We propose a new framework for image-based three-dimensional(3D) model retrieval. We first model the query image as a Euclidean point. Then we model all projected views of a 3D model as a symmetric positive definite(SPD) matrix, which is a point on a Riemannian manifold. Thus, the image-based 3D model retrieval is reduced to a problem of Euclid-to-Riemann metric learning. To solve this heterogeneous matching problem, we map the Euclidean space and SPD Riemannian manifold to the same high-dimensional Hilbert space, thus shrinking the great gap between them. Finally, we design an optimization algorithm to learn a metric in this Hilbert space using a kernel trick. Any new image descriptors, such as the features from deep learning, can be easily embedded in our framework. Experimental results show the advantages of our approach over the state-of-the-art methods for image-based 3D model retrieval.

Detection of engineering vehicles in high-resolution monitoring images

This paper presents a novel formulation for detecting objects with articulated rigid bodies from highresolution monitoring images, particularly engineering vehicles. There are many pixels in high-resolution monitoring images, and most of them represent the background. Our method first detects ob ject patches from monitoring images using a coarse detection process. In this phase, we build a descriptor based on histograms of oriented gradient, which contain color frequency information. Then we use a linear support vector machine to rapidly detect many image patches that may contain ob ject parts, with a low false negative rate and a high false positive rate. In the second phase, we apply a refinement classification to determine the patches that actually contain ob jects. In this stage, we increase the size of the image patches so that they include the complete ob ject using models of the ob ject parts.Then an accelerated and improved salient mask is used to improve the performance of the dense scale-invariant feature transform descriptor. The detection process returns the absolute position of positive ob jects in the original images. We have applied our methods to three datasets to demonstrate their effectiveness.