图像处理在DIC分析中的作用 Processing Images for DIC Analysis

Posted by Micah Simonsen, Last modified by Micah Simonsen on 13 October 2016 01:04 PM

用于二维或三维数字图像相关性测试的采集图像有时可能存在过暗、噪点过多、模糊或混叠等问题。我们时常被问及：使用Photoshop等软件处理这些图像是否能解决上述问题？

Sometimes, images acquired for use in 2D or 3D digital image correlation tests can be too dark; noisy; blurry; or aliased. From time to time we are asked whether it would be beneficial to process these images in software such as Photoshop to fix these issues.

简而言之，此类处理毫无益处。图像信息在从相机传输至电脑的瞬间即已固定；后续调整虽可改变图像外观，却无法提升相关性分析结果，甚至可能造成损害。

In short, this is never beneficial. The information content of the image was set in stone as soon as it was sent from the camera to the PC; further adjustments can change the look of the image, but will not improve your correlation results and may possibly hurt them.

关于亮度问题，Vic-2D和Vic-3D会自动进行图像标准化处理；人为增加数字增益不仅无益，还可能引入量化误差。其他处理操作（如尺寸调整、旋转、滤波）可能引发问题，因为Vic-2D和Vic-3D的插值算法是为模拟已知物理相机特性而精心设计的。任何软后处理都可能产生其他伪影。

In the case of brightness, Vic-2D and Vic-3D will normalize the image internally; adding digital gain to the image will provide no benefit and may introduce quantization errors. Other processing (resizing, rotation, filtering) may cause problems because the interpolation algorithms in Vic-2D and Vic-3D are carefully designed to approximate known physical camera behaviors. Any soft postprocessing may create other artifacts.

若图像过暗导致难以绘制AOI或查看结果叠加层，可使用工具栏中的直方图控制项调整图像外观。此操作不影响相关性计算，但能显著提升图像可视性与绘图便利性。

If an image is too dark to see well for the purposes of drawing an AOI or viewing result overlays, you can use the Histogram control in the toolbar to adjust the look of the image. This does not affect the correlation, but can make the image much easier to see and draw on.

Posted by Elisha Byrne, Last modified by Micah Simonsen on 13 October 2016 01:08 PM

DIC中的连续表面 Continuous Surfaces in DlC 

Because of how we track and match images in order to obtain data, and because of the fundamentals of strain theory, we must treat the area of interest as a continuous surface.  Discontinuous surfaces can result in unreliable data and erroneous strains.

由于我们通过追踪和匹配图像来获取数据的方式，以及应变理论的基本原理，必须将感兴趣区域视为连续表面。不连续表面可能导致数据不可靠和应变计算错误。

追踪与匹配数据 Tracking and Matching Data

我们通过随机斑点图案追踪感兴趣区域（即需要获取数据的样本区域）。追踪对象并非单个斑点，而是称为子集的像素组。 子集大小由用户定义，此处以21像素为例。若子集为21×21像素，则表示在试样表面追踪21×21像素的正方形区域。为实现子集点的追踪与匹配，子集本身必须保持连续性。若子集发生分离或出现不连续行为，则无法进行追踪。 通常断裂的子集（例如表面开裂时）会被剔除，但等高线图中包含的数据部分取决于子集大小和阈值设置（可在运行菜单中调整）。更棘手的是：有时看似子集在裂纹附近被剔除，但裂纹内部的部分错误数据却被纳入了最近的子集。 我们采用子集中心点数据进行报告，因此若子集为21x21像素，等高线图边缘将包含图外10个像素的子集。这意味着部分看似位于裂缝外的数据可能实际包含裂缝区域，导致数据不可靠。由于应变理论假设表面连续性，裂缝内部数据的可靠性将受到影响。

We apply a random speckle pattern to track our Area of Interest (this is the region of the specimen that you wish to obtain data).  We do not track the individual speckles, rather we track groups of pixels that we call Subsets.  The Subset size is user defined, but in this case let's consider a subset size of 21.  If we have a subset size of 21 pixels, this means that we are tracking squares of 21x21 pixels throughout the surface of the specimen.  In order to track and match these subset points, the subsets themselves must remain continuous.  If the subsets break apart or have discontinuous behavior, then we cannot track them.  Typically subsets that break apart (for example, when the surface cracks) will be dropped from the data, however what data is included in the contour plots is partially determined by factors like subset size and thresholding (which you can change in the Run menu).  What is even more problematic is that sometimes it looks like the subset was dropped near a crack, but some of that false data inside the crack is included in the nearest subset.  We report data in the center of the subset, so if the subset is 21x21, the edge of the contour plot will be a subset that includes 10 pixels outside of where the plot is drawn.  So some of the data that looks like it is outside of the crack, could potentially be including the crack, which would result in unreliable data.  The data within that crack is unreliable because strain theory assumes a continuous surface.

应变理论 Strain Theory

应变是量化连续体变形程度的方法。简而言之，它代表长度变化的百分比，可衡量材料的延展性或刚性（但软件中存在不同类型的应变张量）。 若存在孔洞或裂纹等不连续表面，当我们在空隙或裂纹区域进行相关分析时，实质上是在裂纹上安装应变计。因此当材料分离时，可能显示出巨大“应变”值，但这并非真实应变——而是材料断裂导致的裂纹开口位移。当表面出现裂纹或不连续性时，我们的软件通常会自动剔除这些点。 再次强调，我们可通过子集大小和阈值控制纳入的数据量。但观察存在不连续表面的应变时必须谨慎，因材料行为的不连续性可能导致数据失真。

Strain is a way to quantify how a continuous body deforms.  It is, very simply, the percent change in elongation and it is a measure of how ductile or rigid a material is (however, there are different types of strain tensors available in the software).  If you have a discontinuous surface, like a hole or a crack, if we correlate over that void or crack, then we are essentially putting strain gauges over the crack.  So as a material separates, it can show a huge "strain," but the material has broken so that is not strain, rather than just simply displacements of crack openings. When cracks or discontinuities in the surface occur, our software is designed to typically drop those points.  Again, we can control how much data we choose include with subset size and thresholding.  But we must be careful when looking at strains where there are discontinuous surfaces, because those could be erroneous due to the discontinuous material behavior.

特殊情况：复合材料与纺织品  Special Cases: Composites and Textiles

某些具有微观结构的材料（如复合材料和纺织品）在宏观层面表现为连续材料，但在更小尺度上存在纤维滑移和不连续现象。需特别注意的是，材料间滑移可能产生看似剪切应变的现象，但本质上仅是滑移而非应变。 然而在宏观尺度上，材料仍以连续表面形态发生变形。研究纤维材料时，必须同时理解其在微观与宏观层面的变形机制。

Some materials with a micro-structure, such as composites and textiles, can behave as a continuous material on a macrolevel but then have fibers slipping past each other and discontinuities on a smaller scale.  This is something to note because we could potentially see a shear strain due to materials slipping past each other, when it's, in fact, just slippage and not strain.  On a more macro level, however, the material is deforming as a continuous surface.  It's important to have an idea of how the material is deforming on both a micro and macro scale when looking at fibrous materials.

Posted by Elisha Byrne, Last modified by Elisha Byrne on 23 August 2017 10:50 AM

Vic Snap“显示焦点/对比度”功能 Vic Snap "Show Focus/contrast" feature

该功能此前曾标注为“显示协方差”，后改为“显示标准差估计值”。

This feature was previously labeled "Show co-variance" and then "Show sigma estimate."

在Vic-Snap中，可通过右键点击实时图像选择“显示焦点/对比度”功能。 同一菜单中还提供三种不同子集尺寸选项，该功能可作为对焦辅助工具。具体而言，此工具显示Vic-2D/3D分析结果中呈现的标准差置信区间（sigma估计值），该估计值反映在给定焦点/照明条件下搜索和追踪子集的难易程度。为实现实时反馈，该σ估计值的计算时间远短于Vic-2D/3D分析中的σ结果。由于采用不同且更简化的计算方式，其数值可能与结果中的σ值存在差异（因此我们认为原“σ估计值”的命名易造成混淆，故将其更名为“聚焦/对比度”）。

The "Show Focus/Contrast" feature may be selected by right clicking the live image in Vic-Snap.  The user is also given the option for three different subset sizes in the same menu.  This is used as a focus tool.  Specifically, this tool shows an estimate of the sigma (the one standard-deviation confidence interval) that is displayed in Vic-2D/3D results.  This sigma estimate simply estimates how easy it'll be to search and track subsets for the given focus/lighting. In order to provide live feedback, the sigma estimate requires less computing time than the sigma results that are provided in the Vic-2D/3D analysis.  Since it's computed in a different and simpler way, it will likely not match the sigma in the results (which is why we determined that labeling the feature "sigma estimate" was confusing and renamed it "focus/contrast"). 

该工具旨在为用户提供实时对焦反馈（同时兼具图案质量与光照反馈功能）。具体而言，此功能通过测量黑白之间的梯度变化来评估图像质量。我们认为从白到黑的锐利梯度意味着良好的对比度和对焦效果，因此可将其作为对焦辅助工具。这也解释了为何移动光源或调整图像亮度会改变“显示对焦/对比度”图谱。若将其用作对焦工具，请确保在对焦过程中保持光源（及曝光时间）恒定。

The purpose of this tool is to provide the user some live feedback for focus (and also provides some feedback on pattern quality and lighting).  Specifically, this feature is a measure of the gradient between white and black. We assume that a sharp gradient from white to black means that we have good contrast and good focus. This is why it can be used as a focus tool. This is also why if the lights are moved or the brightness of the image is adjusted, it will results in a change in the "show focus/contrast" map. So if you are using it as a focus tool, you'll want the lighting (and exposure time) to remain fixed as you focus.

镜头选择与立体角

发布者：Elisha Byrne，最后修改者：Micah Simonsen

一般而言，相机设置（例如相机之间的立体角和镜头选择）并没有太多硬性规则，因为校准过程会自动计算所有参数，包括立体角和焦距。不过，在搭建立体系统时，仍需注意以下几点。

1. 对于短焦距镜头，应采用较大的立体角。这是因为如果使用短焦距镜头时立体角过小，图像边缘会出现更多的噪声。详细解释请参见此处（从第 57 张幻灯片开始）：DIC-noise-bias.pdf

关于镜头选择/立体角的经验法则如下：

· 对于焦距较短的镜头（8mm、12mm），应使用较大的立体角（至少35度，但越大越好）

· 对于中焦距镜头（17mm），应使用至少25度的立体角

· 对于35mm及以上焦距的镜头，立体角可降至15度（对于超长焦镜头，10度也可接受）。通常情况下，由于长焦镜头/高倍放大倍率固有的景深问题，使用长焦镜头时实际上更倾向于采用较小的立体角。

2. 如果因实验限制必须在短焦距镜头下使用较小的立体角，请将样本置于图像中心。当立体角过小时，图像边缘处的噪声会显著增加。在 Vic-Snap 中，您可以点击“显示辅助线”来帮助将样本定位在图像中心。

3. 使用过大的立体角可能会导致表面与传感器呈过大倾斜角度，从而难以提取图案。这也会给景深带来挑战。当因镜头选择而不得不使用较大立体角时，您需要权衡这些因素。如果图案与传感器呈很大倾斜角度，您可能需要使用“初始猜测”功能来协助软件进行匹配。

4. Correlated Solutions, Inc. 会对镜头和相机进行联合规格确认，以确保它们兼容。我们还会在确定规格前对镜头进行测试，以确保镜头质量适合 DIC 应用。劣质镜头可能会给系统引入噪声。 但是，如果客户希望使用其他镜头，建议先查阅该镜头可覆盖的传感器尺寸范围，并与相机的传感器尺寸进行对比，以确保镜头的成像圈能完全覆盖传感器且无暗角。

5. 对于焦距较短的镜头，您可能需要在“校准对话框”中选择更高的畸变阶数。更多信息请参阅：校准问题故障排除

6. 对于焦距较长的镜头或高倍放大应用，仅当中心 x 和中心 y 值未能正确提取时，才需要在“校准对话框”中选择“高倍放大”。更多信息请参阅：高倍放大校准

Lens Selection and Stereo Angle

Posted by Elisha Byrne, Last modified by Micah Simonsen

Generally speaking, there are not many rules to setting up cameras, such as the stereo angle between cameras and lens selection because the calibration calculates all the parameters, such as stereo angle and focal length.  However there are a few things to keep in mind when setting up your stereo system.

1. For short lenses, use a large stereo angle.  The reason for this is that we get more noise around the edges of the image if a short lens is used with too small of a stereo angle.  This is explained in detail here (starting on slide 57): DIC-noise-bias.pdf

Rules of thumb for lens selection/stereo angle are here:

• For shorter focal length lenses (8mm, 12mm), you should use a large stereo angle (at least 35 degrees, but the higher the better)

• For mid-range focal length lenses (17mm), use at least a 25 degree stereo angle

• For lenses 35mm or longer, it's acceptable to go down to a 15 degree stereo angle (for very long lenses, 10 degrees is OK).  Typically, a smaller stereo angle with longer lenses is actually preferred due to depth-of-field issues inherent to longer lenses/higher magnifications.

2. If you must use a small stereo angle with a short lens due to experimental constraints, keep the specimen in the center of the image.  The noise is much higher along the edges of the image when the stereo angle is too small.  In Vic-Snap, you can click the Toggle Lines on to help position the specimen in the center of the image.

3. Using a very large stereo angle can cause the surface to be so oblique to the sensor that the pattern is hard to extract.  It might present challenges with the depth of field too.  This is something you need to balance when having to use a large stereo angle due to lens selection.  The the pattern is very oblique to the sensor, you might have to use the Initial Guess feature to help the software with the match.

4. Correlated Solutions, Inc. spec's lenses and cameras together, making sure they are compatible.  We also test the lenses before they are spec'ed to make sure the lens quality is suitable for DIC.  Poor quality lenses can introduce noise into the system.  However, if the customer has other lenses they would like to use, it is suggested that the look up what sensor sizes it can cover and compare that to the sensor size of the camera in order to make sure the image circle of the lens properly covers the sensor with no vignetting.

5. For shorter lenses, you might need to select a higher Distortion Order in the Calibration Dialog.  More see: Troubleshooting Calibration Problems

6. For longer lenses/high magnification applications, you might need to select "High Magnification" in the Calibration Dialog if, and only if, the center x and center y values did not extract correctly.  More see: High Magnification Calibration