Dongcheng Hu

Rapid and brief communication: Camera calibration with one-dimensional objects moving under gravity

Camera calibration using one-dimensional (1D) rigid objects is arresting the attentions of researchers since the easy-to-construct geometrical structure of the apparatuses. In this paper, we extend the motion patterns applicable for calibration with the motion of 1D objects. We show that a 1D object with three or more markers, rotating around one marker which is moving in a plane, provides constraint equations on camera intrinsic parameters. A stick moving under gravity without other forces acting on performs such a motion. Simulated tests show the feasibility and numerical robustness of this method.

Constraints on general motions for camera calibration with one-dimensional objects

This paper focuses on two problems in camera calibration with one-dimensional (1D) objects: (a) to find out the general motion patterns well suited for solving the calibration problem, and (b) to improve the robustness and accuracy of the method. Firstly, a sufficient and necessary condition for the solvability of 1D calibration with general motions is proved. Then the special motion of tossing a 1D object is provided as an example to illustrate the correctness and feasibility of this condition. After that some practical issues on obtaining the solution are inspected. By avoiding singularities, the precision and robustness of the method are improved: the relative mean errors are reduced to less than 5% at the noise level of one pixel which surpasses the state-of-the-art methods of the same category.

Semi-automatic initialization of gradient vector flow snakes

Setting of initial contours is a critical problem of active contour models. Although gradient vector flow (GVF) has simplified them greatly compared to the traditional snake, additional work is still necessary. We propose a specific setting method based on vector field analysis. Some critical points in the force field are used to estimate curve deformation. These points seem like sources whose neighbor vectors emanate from them, therefore they can be regarded as inflated centers of deformable contours in GVF snakes. The proposed method detects these points and sets initial contours to contain all the ones in the object area and none of the others. It can prevent the contours from stabilizing on false boundaries in homogeneous regions or real boundaries of other objects. Several experimental results demonstrate that this method is robust and effective.

Pedestrian detection in nighttime driving

This paper presents an approach for pedestrian detection in the nighttime driving with a normal camera. Bright objects in the video are extracted with an adaptive thresholding segmentation algorithm. Then, the size, position, and shape of each object are analyzed to judge whether it is a pedestrian. A tracking module is used to verify the result at last. Experimental results show that the proposed method can detect 71.26% pedestrians.

Active surface model-based adaptive thresholding algorithm by repulsive external force

Inspired by the idea that threshold surface always inter- sects the image surface at high gradient points, an active surface- based adaptive thresholding algorithm is proposed to get the bina- rized result. In this model, the external force is designed to be repulsive from the image surface, thus at the equilibrium state the active surface tends to cover the supporting points of high gradient with smooth property, as well as be away from the image surface locally, which makes the obtained threshold surface properly sepa- rate the foreground and background. The description of the algo- rithm is in a simple and reasonable energy functional form, and only two parameters need to be tuned, which gives more convenience to the operation. Analysis and comparison for the experimental results reveal that it cannot only give the proper thresholding result but also restrain the occurrence of the ghost phenomenon.

A general approach to blind image super-resolution using a PDE framework

Blind super-resolution (BSR) is one of the challenges in the super-resolution image reconstruction area. In this paper, we propose a general approach, which is based on a partial differential equation (PDE) framework, to incorporate the image registration into the point spread function (PSF) estimation process and reconstruct an HR image simultaneously. Since the reconstruction problem is ill-posed, anisotropic diffusion techniques are employed as a regularization term to preserve discontinuities in the HR image estimation. Furthermore, a generalized version of the eigenvector-based alternating minimization (EVAM) constraint, which was proposed for a multichannel framework recently, is developed as another regularization term for the estimations of the PSFs. In this way, a novel blind super-resolution alternating minimization algorithm (BSR-AM) is developed to solve the general model. Experimental results are provided to demonstrate the performance of the proposed algorithm using simulated and real data. The proposed algorithm yields satisfying results, and quantitative error analysis and comparison with the MAP estimation method is illustrated.

Automatic seeded region growing based on gradient vector flow for color image segmentation

We propose a novel automatic seeded region growing method based on gradient vector flow (GVF) for color image segmentation. YCbCr color space is selected to avoid the high correlation of RGB color space. First, a GVF field is constructed from an edge map of the input image. Then a scaler force field is derived from it by minimizing an energy functional iteratively. From the scalar field, we can select a set of seeds and get an initial segmentation via a straightforward downstream process. Finally, a region adjacency graph–based region merging is applied to merge similar neighboring regions into true results. Experimental results demonstrate that this method is insensitive to noises and efficient to multiple objects segmentation in color images.

Recognition of perspectively distorted planar grids

In this paper, properties of planar (2D) grids are addressed in a mathematical point of view, and a novel algorithm for recognizing distorted grids with perspective transformations is presented. The proposed approach contains three parts: (a) recognizing parameters of affinely distorted grids by fitting Gaussian mixture models (GMMs) to grid spectrums, (b) rebuilding the grid structures via a generating iteration based on the acquired parameters, and (c) eliminating nonlinear effects caused by perspective transformations with the median of infinite lines from local structures (MILLS) method. All parts are precise and robust to local distortions, the absence of elements, and outliers. The accuracy and robustness are demonstrated by quantitative statistics in experiments on synthesis grids and real grid images.

Adaptive level set image segmentation using the Mumford and Shah functional

This paper proposes an adaptive 2-phase level set image segmentation algorithm to improve the original level set-based piecewise constant Mumford-Shah model. With the introduction of a multiplicative gain field, the model is adaptive to intensity inhomogeneity, thus tending to obtain the actual boundaries of the objects, as well as a property-seeking and -driven classification algorithm.

Image superresolution using fractal coding

We address the problem of image superresolution and present a novel approach to single-frame superresolution using fractal image coding. The proposed approach takes great advantage of the properties of fractals—resolution independence, similarity preservation, and nonlinear operation—which are suited for image superresolution, specifically for image restoration and magnification. The idea of our work is to estimate the fractal code of the original image from its degraded blurred and noised observation and decode it at a higher resolution, and all the strategies are performed in the fractal image coding frame- work. To achieve this, we employ an adaptive fractal coding scheme in the frequency domain, and further, we introduce an overlapping partition scheme to remove the blocky artifacts and improve the reconstruction quality. Experiments on simulated and real images show that the result- ing fractal-based superresolution method yields superior performance to conventional single-frame superresolution methods.