基于改进Census变换的图像立体匹配算法研究

针对传统的 Census 区域匹配算法过分依赖窗口中心像素信息,导致算法受到噪声干扰时匹配精度降低的问题,提出一种基于改进 Census 变换的匹配算法.采用局部像素反差值为中心像素选择的评判标准,对传统的 Census 变换进行改进,增强了窗口像素信息的利用,提高了算法对像素值突变的适应性,使算法有更好的鲁棒性;代价聚合阶段采用引导图滤波算法并结合多尺度聚合模型,增强平坦区域像素间的区分度;采用 win-take-all 算法选取最优视差值,完成视差计算;采用区域投票策略和中值滤波算法完成视差精化.利用该改进算法对 Middlebury 平台提供的标准图像进行实验,实验结果表明该算法较传统Census 算法有较好的抗噪能力和立体匹配精度.     

Accurate particle position measurement from images

The moment method is an image analysis technique for subpixel estimation of particle positions. The total error in the calculated particle position includes effects of pixel locking and random noise in each pixel. Pixel locking, also known as peak locking, is an artifact where calculated particle positions are concentrated at certain locations relative to pixel edges. We report simulations to gain an understanding of the sources of error and their dependence on parameters the experimenter can control. We suggest an algorithm, and we find optimal parameters an experimenter can use to minimize total error and pixel locking. For a dusty plasma experiment, we find that a subpixel accuracy of 0.017 pixel or better can be attained. These results are also useful for improving particle position measurement and particle tracking velocimetry using video microscopy in fields including colloids, biology, and fluid mechanics.     

Potential use of image analysis techniques in surface texture studies

This paper discusses a possible new approach to surface texture measurement based on the computer analysis of pictorial surface images. The method involves the evaluation of image textural features which are then compared with prototype standards derived from surfaces with known roughness properties. Preliminary results show that the technique is capable of distinguishing between different manufacturing methods and roughness values     

Detection of pinhole defects on chips and wafers using DCT enhancement in computer vision systems

This paper presents a global approach for the automatic inspection of tiny pinhole defects in randomly textured surfaces of surface barrier layer (SBL) chips. By means of a discrete cosine transform (DCT)-based image restoration scheme, the proposed method is independent of textural features and thus not confined by the limitations of feature extraction based methods. Through properly decomposing the frequency matrix of an image in the DCT domain and selecting the best radius of the sector filter for the high-pass filtering operation, we effectively attenuate the global random texture pattern and accentuate only tiny pinhole defects in the restored image. We also develop two accumulative sum detection procedures that automatically determine the best high-pass filtering parameters based on the abrupt changes of the frequency coefficients in the decomposed matrix. Experimental results show that the proposed method outperforms the traditional approach in reducing the Type I error by 70–80% and in decreasing the deviation of the defect areas by 95%. Moreover, the proposed method can be applied to various types of passive components in large-batch production because no precise positioning of the target chip or template matching is required.     

交叉双边滤波和视觉权重信息的图像融合

图像融合技术是图像分析领域重点研究内容之一,为了更好地保留原图像中的细节信息,提高融合图像的对比度,提出了基于视觉权重图的多尺度图像融合方法。首先,利用可变参数的交叉双边滤波器对两幅待融合图像进行多尺度分解;然后,在每个分解层分别计算相应的视觉权重图,并针对不同分解层赋予不同的权重值;最后,综合这些结果生成融合图像。由于对原始图像的分解没有采用下采样和上采样操作,因此不会损失图像中的信息,且克服了传统像素级融合方法中融合图像模糊、对噪声敏感等不足。通过4种定量分析实验表明,在多种模式的图像融合应用中,本方法优于其他5种对比方法,融合时间小于0.2 s。融合后图像细节信息、对比度得到增强,同时降低处理时间。     

Saving the photons: mapping X-rays by position-tagged spectrometry

Since the early years of X-ray spectrometry in electron microscopes, mapping the locations of chemical elements has been important. The X-rays needed in large numbers for this are rare, owing to poor production efficiency compared with electron signals, and at risk of loss by many mechanisms such as missing the limited solid angle of the detector, absorption before reaching the detector and pulse pile-up conventional digital mapping hardware reduces the information contained in the X-ray spectrum at each pixel to the itegrated counts from a few regions of interest.
The acquisition technique of position-tagged spectrometry eliminates the conflict between the desire to see full frame X-ray images quickly versus the analytical advantages of having complete spectra for each pixel. As the beam is scanned rapidly relative to traditional X-ray mapping, photons are counted in a virtual 3-D multichannel analyser on disk, preserving both spatial and spectral information. Along with the sophisticated post-processing allowed by storing an entire spectrum per pixel, a unique degree of dynamic interaction with the developing data is made possible by integrating many short scans instead of using a single long dwell time at each pixel.     

结合假边缘提取和直方图分析的图像灰度变换

光学刻划字符图像的前景通常由高灰度像素区域和低灰度像素区域组成,导致使用传统的基于梯度的边缘提取方法如Canny算子不能准确地获得字符的轮廓,为此本文提出了一种利用假边缘信息结合直方图分析的图像灰度变换方法,变换后的图像前景仅由低灰度像素构成。首先对Canny边缘点进行特征分析,提取其假边缘点;然后进行图像直方图分析,确定灰度变换范围以及灰度对应关系,并对假边缘两侧的边界区域施加光滑度约束来确定灰度变换参数;最后以此变换参数对非背景区域的高灰度像素进行灰度变换。实验证明,灰度变换后字符笔画仅由低灰度像素构成,原边界处的灰度变化足够光滑,可以使用基于梯度的方法来提取完整而准确的字符轮廓。     

Microscopic retinal blood vessels detection and segmentation using support vector machine and K-nearest neighbors

The retina is the deepest layer of texture covering the rear of the eye, recorded by fundus images. Vessel detection and segmentation are useful in disease diagnosis. The retina's blood vessels could help diagnose maladies such as glaucoma, diabetic retinopathy, and blood pressure. A mix of supervised and unsupervised strategies exists for the detection and segmentation of blood vessels images. The tree structure of retinal blood vessels, their random area, and different thickness have caused vessel detection difficulties at machine learning calculations. Since the green band of retinal images conveys more information about the vessels, they are utilized for microscopic vessels detection. The current research proposes an administered calculation for segmentation of retinal vessels, where two upgrading stages depending on filtering and comparative histogram were applied after pre-processing and image quality improvement. At that point, statistical features of vessel tracking, maximum curvature and curvelet coefficient are extracted for each pixel. The extracted features are classified by support vector machine and the k-nearest neighbors. The morphological operators then enhance the classified image at the final stage to segment with higher accuracy. The dice coefficient is utilized for the evaluation of the proposed method. The proposed approach is concluded to be better than different strategies with a normal of 92%.     

A Comparison of Texture Feature Extraction Methods for Machine Condition Monitoring and Failure Analysis

The diagnosis of worn and damaged surfaces is an important issue in machine failure analysis and condition monitoring. Of many approaches used, image classification based on feature parameters has often proven to be particularly useful. Accurate classification can, however, be limited by the fact that feature parameters vary with scale and orientation. Hence, it is essential to determine which feature parameters are both scale and rotation invariant. This paper presents a performance evaluation of feature extraction methods currently used in pattern recognition. A comparison of six methods is conducted, in order to find the method that provides the most consistent results over a large range of image sizes and rotations. The methods analysed are: co-occurrence matrix, discrete wavelet transform, combination of wavelet and co-occurrence features, Gabor filter, circular Gaussian Markov random field and local binary patterns. For the comparison, four datasets of images with different scales and rotations are used, i.e. Brodatz textures, artificially generated isotropic fractal images and Talysurf images of sandblasted and abraded steel surfaces. The performance of each method is evaluated on the datasets using k-nearest neighbours and linear based normal densities classifiers. The results showed that the combined feature extraction method produced the most robust and accurate results for each of the datasets, and appears to be suitable for the classification of tribological surfaces.     

Diffusion as a model of formation and development of surface topography

Changes of surface topography in tribological systems are due to inelastic processes as plastic deformation, detaching of wear particles and their reintegration into the surfaces. Due to these processes, the material particles are transported either along a surface or from one tribological partner to the other. Both processes are due to random interactions between surface asperities and are stochastic processes. The stochastic mass transfer between the surfaces is interpreted and described in the paper as a random deposition, the transport along a surface as a `diffusion' processes with some effective diffusion coefficient. We consider the development of the surface topography due to the described two kinds of random processes. There exist some stationary (in statistical sense) random surface topography with a power spectrum (spectral density) typical for many real frictional surfaces. The parameters of the model can either be obtained from comparison with measured topography or extracted from simulations at a lower space scale. The proposed mass transport model further allows to determine the wear rate in the system.     

基于3-D Facet模型的亚体素边缘检测算法研究

在产品的虚拟测量中,需要获得产品内外表面的3D高精度轮廓信息。本文首先介绍了一种直接从CT切片序列获得表面点的方法,通过3DFacet模型计算方向导数并根据沿梯度方向的二阶导数零点获得边缘点的亚体级精确位置。该方法可以检测到法向垂直于切片的表面,并且由于考虑了待检测切片相邻层的信息,与仅考虑单层信息的2D亚像素边缘检测方法相比,它抗噪能力更强、精度更高。其次,本文提出了一种改进方法,通过缩减候选像素的预处理,在很大程度上提高了原始算法的处理速度。最后将改进算法应用到航空发动机叶片仿真模型的检测中,结果表明本算法检测精度可达0.1个像素左右,速度较原始算法提高约10倍。     

A Novel Approach Using a Two-Dimensional Wavelet Transform in Ball Grid Array (BGA) Substrate Conducting Path Inspections

This paper presents a novel approach for localising open and short boundary defect candidates on ball grid array (BGA) substrate conducting paths by using a 2D wavelet transform (2D WT). Once the potential defects are identified, traditional printed circuit board (PCB) inspection algorithms can focus on these candidates for further analysis to identify true open and short defects. The defect-detecting scope and the inspection effort are thereby significantly reduced. The binary BGA substrate image is processed. It shows only the boundaries of BGA substrate conducting paths, which are further decomposed directly by 2D WT. Since most of the wavelet energy is clustered at the edges of image objects, the wavelet transform modulus sum (WTMS) of each edge pixel on BGA substrate conducting path boundaries is initially collected. By comparing the WTMS of an edge pixel on decomposition level j with its WTMS on an adjacent decomposition level j + 1, an across-level ratio can be estimated to verify the irregularity of an edge pixel. That is, an edge pixel is classified as strongly irregular (e.g. potential open or short defects) if its across-level ratio reaches a predefined threshold. The proposed approach is template-free and easy to implement, so it is suitable for small-batch production. Real BGA substrates with synthetic boundary defects are used as test samples to evaluate the performance of the proposed approach. Experimental results show that the proposed method is capable of capturing all the open and short defects on BGA substrate conducting paths without missing any errors by using a selected across-level ratio threshold and appropriate decomposition level.     

Design rules for Background Oriented Schlieren experiments with least-squares based displacement calculation

Background Oriented Schlieren (BOS) is a non-intrusive optical method used to visualize density variations in fluids. A camera is used to measure the distortion of a background image displaying a random pattern due to the refraction of the light induced by the local gradients in the refractive index of the under study. In contrast to conventional schlieren, for which direct observation is possible, the images obtained with the BOS technique have to be processed. In this contribution, a computationally affordable approach based on linear least-squares minimization and adapted from optical flow techniques is presented in detail and tested in the context of BOS with the Matlab GUI application comBOS. A series of synthetic images are first analyzed to devise design rules for the definition of an optimal random Gaussian dot background pattern. Tests are performed for comBOS, the commercial BOS module included in DaVis 10, and the free PIV tool PIVlab. It is found that the diameter of the dot images on the camera sensor should be at least 3 pixels, but can be larger without affecting the accuracy of the measurement with comBOS and DaVis 10, while PIVlab severely degrades the quality of the results for dots larger than 5 pixels. The contrast and the density of dots must also be sufficient such that a minimum of 75 % of the surface of the background is shaded. If these conditions are met, the maximum systematic error in the measured distortions is 0.006 pixel for comBOS and 0.001 pixel with DaVis 10 when local gradients are small. Similar errors are found with PIVlab for dots smaller than 5 pixels in diameter, but they increase rapidly for larger dots. The random error is typically smaller than 0.004 pixel in optimal conditions, but it increases significantly for images affected by a random noise or local gradients of the distortion. Finally, it is concluded with a realistic test case that both comBOS and DaVis 10 are well suited to capture the main features of a complex compressible flow with Schlieren imaging. Thus, the main advantage of comBOS is its affordability and robustness for researchers and educators willing to implement BOS, for example in educational contexts.     

Background removal in scanning tunneling spectroscopy of single atoms and molecules on metal surfaces

Scanning tunneling spectroscopy has developed into a powerful spectroscopic technique that has found wide application in the atomic scale characterization of the electronic properties of clean surfaces as well as adsorbates and defects at surfaces. However, it still lacks the standard methods for data treatment and removal of artifacts in spectra as they are, e.g., common in photoemission spectroscopy. The properties of the atomic scale tip apex--the probe of the instrument--tend to introduce spurious background signals into tunneling spectra. We present and discuss two methods which permit to extract tip-independent information from low temperature tunneling spectra acquired on single atoms and molecules on single crystal surfaces by background subtraction. The methods rely on a characterization of the tip on the clean metal surface. The performance of both methods is demonstrated and compared for simulated and experimental tunneling spectra.     

Steel billet reheat simulation with growth of oxide layer and investigation on zone temperature sensitivity

This paper presents a three-dimensional heat conduction numerical model and simulation of steel billet reheating in a reheat furnace. The model considers the growth of oxide scale on the billet surfaces. Control-volume approach and implicit scheme of finite difference method are used to discretize the transient heat conduction equation. The model is validated with analytical results subject to limited conditions. Simulations are carried out for predictions of three-dimensional temperature filed in the billet and oxide scale growth on the billet surfaces. The model predictions are in agreement with expected trends. It was found that the effect of oxide scale on billet heating is considerable. In order to investigate the effect of zone temperatures on the responses, a parametric sensitivity subject to six responses of interest are carried out using analysis of mean approach. The simulation approach and parametric study presented will be useful and applicable to the steel industry.     

基于分层模型与局部复原的多聚焦图像融合方法

针对多聚焦图像融合后清晰像素信息损失较为严重的问题,提出一种分层模型与局部复原相结合的方法。首先,根据源图像模糊区域分布情况,对不同灰度特征源图像进行多尺度分解选取恰当的分层模型,并利用此模型对图像进行初步融合;在此基础上,通过计算对应区域的功率谱曲线斜率以推导图像中局部模糊区域像素值,提高图像融合算法的细节提取能力,再结合维纳滤波策略对其进行恢复,以实现图像清晰融合。最后,利用3个实例验证了该方法的有效性与可行性。     

Atomic force microscopy of human hair

The atomic force microscope (AFM) was used to investigate the surface architecture of the entire lengths of cleaned human head hairs. Many features previously seen with the scanning electron microscope (SEM) were identified. However, the AFM has provided much greater detail and, in particular, the hair's cuticular surfaces appear not to be as smooth as had been previously supposed. A consistent feature was of step discontinuities or "ghosts" on the scale surfaces. These delineated the original location of each overlying scale before its edge had been chipped away. There was a change in the longitudinal angular presentation of the surfaces about each ghost. This means the distal ends of each cuticle cell have been synthesised in the follicle to be thicker than where that same cuticle cell is bounded on both sides by other cuticle cells. The undamaged outer cuticular surfaces at the root end of each hair were covered everywhere by longitudinal ridges (striations). Where the hair surface was worn, the striations terminated at a scale edge ghost. The ridges were approximately 9 nm high and were in parallel array with a lateral repeat spacing of about 350 nm. The striations are evidently formed on the outer surface of each cuticle cell following earlier contact in the hair follicle with the inner root sheath. The study of stained transverse sections of hairs in the transmission electron microscope (TEM) is suggested as a means for throwing some light on the underlying structure and chemistry of the striations. Finally, our AFM studies have revealed that the surface of the freshly emergent hair gradually changes over a distance of about 20 mm and that the surface of the hair for most of its length is quite different from that near the root. This is likely to be of import to those engaged in the hair toiletries industry.     

Prioritization analysis and compensation of geometric errors for ultra-precision lathe based on the random forest methodology

Geometric errors remarkably affect the dimensional accuracy of parts manufactured by ultra-precision machining. It is vital to consider the workpiece shape for the identification of crucial error types. This research investigates the prioritization analysis of geometric errors for arbitrary curved surfaces by using random forest. By utilizing multi-body system (MBS) theory, a volumetric error model is initially established to calculate tool position errors. An error dataset, which contains information of 21 geometric errors, workpiece shape, and dimensional errors, is then constructed by discretizing the workpiece surface along the tool path. The problem of identifying crucial geometric errors is translated into another problem of feature selection by applying random forest on the error dataset. Moreover, the influence extent of each geometric error on the dimensional accuracy of four typical curved surfaces is analyzed through numerical simulation, and crucial geometric errors are identified based on the proposed method. Then, an iterative method of error compensation is proposed to verify the reasonability of the determined crucial geometric errors by specifically compensating them. Finally, under compensated and uncompensated conditions, two sinusoidal grid surfaces are machined on an ultra-precision lathe to validate the prioritization analysis method. Findings show that the machining accuracy of the sinusoidal grid surface with crucial geometric error compensation is better than that without compensation.     

Phase Maps Retrieval from Sequences of Phase Shifted Images with Unknown Phase Steps Using Generalized N-Dimensional Lissajous Figures—Principles and Applications

This article presents an alternative approach for retrieving phase information from a sequence of images with unknown phase shifts. A sequence of five or more discrete interferograms with unknown and different phase shifts are used to determine the parameters of N-dimensional Lissajous figures. A set of N-dimensional discrete points are used to calculate the relative phase shifts. Once zthe phase shifts are determined, the phase values for each image pixel are determined from a generalized phase calculation equation for irregular phase shifts. The article presents the physical and mathematical basis of the developed approach and discusses some results obtained from simulated data and real data from controlled experiments.     

一种基于基元匹配的立体图像修复算法

针对立体图像修复中直接采用二维图像修复算法所导致的深度信息修复效果欠佳的问题,提出一种基于基元匹配的立体图像修复算法.通过在立体图像左右视图间进行基元匹配,求得各个像素对应目标的深度信息;然后通过平滑深度求得缺损区域的深度数据;最后通过立体图像成像公式计算出左视图中用于填补右视图缺损区域的对应像素点坐标.通过与Criminisi算法进行对比实验,结果表明,对于多幅缺损的立体图像,该算法均取得了较好的修复效果.