Zuofeng Zhou

Contourlet-based image denoising algorithm using adaptive windows

Contourlet is a new effective signal representa-tion tool in many image applications. In this paper, a contourlet-based image denoising algorithm using adaptive windows which utilizes both the captured directional information by the contourlet transform and the intrinsic geometric structure information of the image is proposed. The adaptive window in each of the contourlet subband is first fixed by autocorrelation function of contourlet coefficients energy distribution, and then the local Wiener filtering is used to denoise the noisy image. Experiments show that the proposed algorithm achieves better performance than current subsampled contourletbased image denoising algorithms.

Multifocus Noisy Image Fusion Algorithm Using the Contourlet Transform

Multifocus image fusion deals with a stack of images that were acquired with a different focus point to get an image with all the objects in the scene full focused. Considered the noise effect, in this paper, we proposed a multifocus noisy image fusion algorithm using the contourlet transform. Utilizing the captured directional information by the contourlet transform, the directional windows are used to determine the fusion weight. The experiment results demonstrate the higher performance of the proposed multifocus noisy image fusion algorithm compared to the wavelet-based multifocus noisy image fusion method

Multi-Frame Super-Resolution Reconstruction Algorithm Based on Diffusion Tensor Regularization Term

This paper presents a multi-frame super-resolution (SR) reconstruction algorithm based on diffusion tensor regularization term. Firstly, L1-norm structure is used as data fidelity term, anisotropic diffusion equation with directional smooth characteristics is introduced as a prior knowledge to optimize reconstruction result. Secondly, combined with shock filter, SR reconstruction energy functional is established. Finally, Euler-Lagrange equation based on nonlinear diffusion model is exported. Simulation results validate that the proposed algorithm enhances image edges and suppresses noise effectively, which proves better robustness.

Image denoising algorithm via spatially adaptive bilateral filtering

The classical bilateral filtering algorithm is a non-linear and non-iterative image denoising method in spatial domain which utilizes the spatial information and the intensity information between a point and its neighbors to smooth the noisy images while preserving edges well. To further improve the image denoising performance, a spatially adaptive bilateral filtering image deonoising algorithm with low computational complexity is proposed. The proposed algorithm takes advantage of the local statistics characteristic of the image signal to better preserve the edges or textures while suppressing the noise. Experiment results show that the proposed image denoising algorithm achieves better performance than the classical bilateral filtering image denoising method.

Construction of 2-D directional filter bank by cascading checkerboard-shaped filter pair and CMFB

In this paper, we propose a new approach to construct a 2-dimensional (2-D) directional filter bank (DFB) by cascading a 2-D nonseparable checkerboard-shaped filter pair and 2-D separable cosine modulated filter bank (CMFB). Similar to diagonal subbands in 2-D separable wavelets, most of the subbands in 2-D separable CMFBs, tensor products of two 1-D CMFBs, are poor in directional selectivity due to the fact that the frequency supports of most of the subband filters are concentrated along two different directions. To improve the directional selectivity, we propose a new DFB to realize the subband decomposition. First, a checkerboard-shaped filter pair is used to decompose an input image into two images containing different directional information of the original image. Next, a 2-D separable CMFB is applied to each of the two images for directional decomposition. The new DFB is easy in design and has merits: low redundancy ratio and fine directional-frequency tiling. As its application, the BLS-GSM algorithm for image denoising is extended to use the new DFBs. Experimental results show that the proposed DFB achieves better denoising performance than the methods using other DFBs for images of abundant textures.

A novel real-time method for high dynamic range image tone mapping

Tone-mapping technique which can convert high dynamic range (HDR) to low dynamic range (LDR) images provides accurately visualization of HDR images on standard LDR displays. Most of the existing tone-mapping method could not realize real time processing while preserving good visualization. Utilizing an adaptive three-section lookup table, this paper proposes an effective, high quality, real time technique to convert 12-bit images to 8-bit image which can preserve abundant details and high contrast simultaneously. Experiment results show that this method can improve the weak signals of the image greatly, and the low luminance details can be observed distinctly on an 8-bit monitor.

Non-contact heart rate measurements based on skin detection

Non-contact measurements of the cardiac pulse of humanity is always a challenging problem. Almost all the existing research work concentrates on heart rate measurements based on video recordings of the human face, in which the subject is restricted rigidly. In this paper, we propose a novel approach based on skin detection method. The skin regions in color video recordings are generalized to anywhere of the human body, not limited to human face. The skin detector and a chrominance-based technique are firstly used to detect the photo-plethysmo-graphy (PPG) signal. Then, we use an empirical parameter selection method to get the final heart rate estimation. The experiment results show the effectiveness of the proposed method.

Target measurement in elliptical orbit using generalized voxel coloring

Measurement of 3D scene from image sequences is necessary for many computer vision applications. In this paper, we identify the volumetric scene reconstruction as a critical issue for target measurement and use the statistical method to estimate the size of the reconstructed target. The proposed approach unifies the view volume in volumetric scene reconstruction while we only know the location information of the camera relative to the target in elliptical orbit. Experimental results on elliptical orbit with 200 meters long axis and 100 meters short axis illustrate that the average error is about 44 millimeters, which meets the accuracy requirement of general measurement.

High precision measurement and 3D reconstruction of non-cooperative spacecraft based on binocular vision

In this paper, we propose an automatic method for high precision measurement and 3D reconstruction of non-cooperative spacecraft based on binocular vision. The Zhengyou Zhang’s calibration method was implemented to calibrate the camera’s internal and external parameters; the 8-point algorithm was adopted to compute the fundamental and essential matrix between the cameras; we got the relative position of binocular camera by the method of the SVD of essential matrix; a stereo matching algorithm depending on the Semi-Global Matching was adopted for disparity map. Subsequently, the cloud information of world points was calculated through least square method. In our experiment, a complex outdoor scene was used. We made a satellite model which has the same size of the real one. To get the accurate position and angle of the spacecraft, the ellipse marks on the spacecraft was exacted effectively under three constraints. The experimental results show that the spacecraft model can be reconstructed accurately by our method. The method contributes the error rate of 1% for the test length and 3% for the test angle in 1 meter.

Research for road detection technology in low illumination remote sensing image

In this paper, a road detection algorithm from the low illumination remote sensing images is proposed. First, the top-hat transform is used to enhance the edge information in low illumination images. Next, a road detection method based on parallel lines is proposed to detect the parallel characteristics of the two edges of the road. The experiment results show that the proposed algorithm can detect the road information effectively and precisely.