招商网站建设需要什么,网站建设平台排行榜,我做微信淘宝客网站有哪些,wordpress 图片中文名称转为时间文章目录 傅里叶变换频谱矩形圆菱形黑白相间的亮带去除图纹#xff08;反傅里叶变换#xff09;去除图纹滤波器处理 Halcon 频域空间域检测缺陷Halcon 频域差分空间域 缺陷检测#xff08;lines_gauss 提取线#xff09;Halcon 频域差分空间域#xff08;blob特征#xf… 文章目录 傅里叶变换频谱矩形圆菱形黑白相间的亮带去除图纹反傅里叶变换去除图纹滤波器处理 Halcon 频域空间域检测缺陷Halcon 频域差分空间域 缺陷检测lines_gauss 提取线Halcon 频域差分空间域blob特征案例Halcon Blob特征处理缺陷Halcon 频域空间域检测划痕Halcon 傅里叶变换转换为功率图频域blob差分Halcon 训练学习缺陷检测以神经网络mlp为例 傅里叶变换频谱
傅里叶去除纹理
矩形
read_image (Image1, C:/Users/Augustine/Desktop/频谱图形/1.png)
fft_image(Image1,ImageFFT)圆
read_image (Image2, C:/Users/Augustine/Desktop/频谱图形/2.png)
fft_generic(Image2, ImageFFT, to_freq, -1, sqrt, dc_center, complex)菱形
read_image (Image3, C:/Users/Augustine/Desktop/频谱图形/3.png)
fft_image(Image3, ImageFFT)黑白相间的亮带
read_image (Image4, C:/Users/Augustine/Desktop/频谱图形/4.png)
fft_generic(Image4, ImageFFT, to_freq, -1, sqrt, dc_center, complex)去除图纹反傅里叶变换
read_image (Image4, C:/Users/Augustine/Desktop/频谱图形/4.png)
* 图片转换为灰度图
rgb1_to_gray(Image4, GrayImage)
fft_image(GrayImage, ImageFFT)
gen_rectangle1 (ROI_0, 12.9631, 372.108, 377.501, 710.691)
gen_rectangle1 (ROI_0, 17.1055, 11.9479, 373.359, 347.212)
union2(ROI_0, ROI_0, RegionUnion)
gen_rectangle1 (ROI_0, 412.712, 13.6076, 754.467, 345.552)
union2(ROI_0, ROI_0, RegionUnion1)
gen_rectangle1 (ROI_0, 412.712, 368.788, 768.966, 709.031)
union2(ROI_0, ROI_0, RegionUnion2)
* 区域填充
paint_region(RegionUnion2, ImageFFT, ImageResult,0, fill)
* 反傅里叶变换
fft_image_inv(ImageResult, ImageFFTInv)去除图纹滤波器处理
低通滤波器、高通滤波器和中通滤波器都是数字图像处理中常用的滤波器它们的作用是将输入图像信号进行滤波处理以达到去除噪声、增强图像特征等目的。低通滤波器Low-Pass Filter低通滤波器可以去除图像中高频部分的信息保留低频部分的信息。在频域上看低通滤波器会将图像高频成分的能量减弱从而产生模糊的效果。低通滤波器一般用于平滑图像、去除噪声等应用场景。高通滤波器High-Pass Filter高通滤波器可以去除图像中低频部分的信息保留高频部分的信息。在频域上看高通滤波器会将图像低频成分的能量减弱从而使高频细节更加突出。高通滤波器一般用于图像增强、边缘检测等应用场景。中通滤波器Band-Pass Filter中通滤波器可以去除图像中低频和高频部分的信息保留中频部分的信息。中通滤波器一般用于分离出图像中特定频率范围内的信息例如检测特定大小的物体。read_image (Image4, C:/Users/Augustine/Desktop/频谱图形/4.png)
* 图片转换为灰度图
rgb1_to_gray(Image4, GrayImage)
*fft_image(GrayImage, ImageFFT)
fft_generic(ImageConvol, ImageFFT1, to_freq, -1, sqrt, dc_center, complex))
get_image_size(ImageFFT, Width, Height)
* 高通滤波器低频段挡住为黑色
gen_highpass(ImageHighpass, 0.2, none, dc_center, Width, Height)
* 频域中对两个经过傅里叶变换的图像进行卷积操作
convol_fft(ImageFFT, ImageHighpass, ImageConvol)
* 反傅里叶变换 第三 参数改成from_freq,1和-1要反着来
fft_generic(ImageConvol, ImageFFT1, from_freq, 1, sqrt, dc_center, complex))Halcon 频域空间域检测缺陷 高斯滤波主要用于祛除图像中的高频成分低通滤波器也就是去除图像中的细节和噪声。通过在图像上应用高斯核进行卷积操作高频部分会被削弱从而使图像变得更加平滑。 由于高斯核的特性其在中心位置具有最大值并且随着离中心的距离逐渐减小。这意味着高斯滤波会更强调图像中的低频信息即图像中相对较平均和较大尺度的变化。而高频信息例如细节和噪声由于高斯核的衰减作用会在滤波过程中被抑制或消除。 因此高斯滤波的主要效果是减少图像中的高频成分实现图像的平滑和模糊化。这种平滑可以有效地去除图像中的噪声并在某些情况下有助于提高图像处理的结果。 * This program demonstrates how to detect small texture
* defects on the surface of plastic items by using the fast
* fourier transform (FFT).
* First, we construct a suitable filter using Gaussian
* filters. Then, the images and the filter are convolved
* by using fast fourier transforms. Finally, the defects
* are detected in the filtered images by using
* morphology operators.
*
* Initializations
* 1.采集图像
dev_update_off ()
dev_close_window ()
read_image (Image, plastics/plastics_01)
get_image_size (Image, Width, Height)
dev_open_window (0, 0, Width, Height, black, WindowHandle)
set_display_font (WindowHandle, 14, mono, true, false)
dev_set_draw (margin)
dev_set_line_width (3)
dev_set_color (red)
* 针对图像大小进行优化
* Optimize the fft speed for the specific image size
optimize_rft_speed (Width, Height, standard)
*
* Construct a suitable filter by combining two gaussian
* filters
* 产生两个滤波器
Sigma1 : 10.0
Sigma2 : 3.0
* 2.产生滤波器产生高斯滤波
gen_gauss_filter (GaussFilter1, Sigma1, Sigma1, 0.0, none, rft, Width, Height)
gen_gauss_filter (GaussFilter2, Sigma2, Sigma2, 0.0, none, rft, Width, Height)
* 滤波器进行差分形成一个新的滤波器
sub_image (GaussFilter1, GaussFilter2, Filter, 1, 0)
*
* Process the images iteratively
NumImages : 11
for Index : 1 to NumImages by 1* * Read an image and convert it to gray valuesread_image (Image, plastics/plastics_ Index$02)rgb1_to_gray (Image, Image)* 3.傅里叶变换频域变换* Perform the convolution in the frequency domain* 空间域转换为频域rft_generic (Image, ImageFFT, to_freq, none, complex, Width)* 频域中对两个经过傅里叶变换的图像进行卷积操作convol_fft (ImageFFT, Filter, ImageConvol)* 将频域转换为空间域rft_generic (ImageConvol, ImageFiltered, from_freq, n, real, Width)* * Process the filtered image* 4.空间域blob 分析* 灰度范围变化把图像亮的地方更亮暗的地方更暗gray_range_rect (ImageFiltered, ImageResult, 10, 10)min_max_gray (ImageResult, ImageResult, 0, Min, Max, Range)threshold (ImageResult, RegionDynThresh, max([5.55,Max * 0.8]), 255)connection (RegionDynThresh, ConnectedRegions)select_shape (ConnectedRegions, SelectedRegions, area, and, 4, 99999)union1 (SelectedRegions, RegionUnion)closing_circle (RegionUnion, RegionClosing, 10)connection (RegionClosing, ConnectedRegions1)select_shape (ConnectedRegions1, SelectedRegions1, area, and, 10, 99999)* 获取缺陷点的坐标area_center (SelectedRegions1, Area, Row, Column)* * Display the results* 找缺陷dev_display (Image)Number : |Area|if (Number)gen_circle_contour_xld (ContCircle, Row, Column, gen_tuple_const(Number,30), gen_tuple_const(Number,0), gen_tuple_const(Number,rad(360)), positive, 1)ResultMessage : [Not OK,Number defect(s) found]Color : [red,black]dev_display (ContCircle)elseResultMessage : OKColor : forest greenendifdisp_message (WindowHandle, ResultMessage, window, 12, 12, Color, true)if (Index ! NumImages)disp_continue_message (WindowHandle, black, true)stop ()endif
endforHalcon 频域差分空间域 缺陷检测lines_gauss 提取线 * this example shows how to detect mura defects
* in blurred images
*
dev_close_window ()
dev_update_off ()
Path : lcd/mura_defects_blur_
read_image (Image, Path 01)
get_image_size (Image, Width, Height)
dev_open_window_fit_size (0, 0, Width, Height, 640, 480, WindowHandle)
set_display_font (WindowHandle, 14, mono, true, false)
dev_set_draw (margin)
dev_set_line_width (3)
dev_set_color (red)
* 设置缩放因子
ScaleFactor : 0.4
* 计算高斯线性参数
calculate_lines_gauss_parameters (17, [25,3], Sigma, Low, High)
for f : 1 to 3 by 1read_image (Image, Path f$.2i)* 获取rgb 通道decompose3 (Image, R, G, B)* correct side illumination* 频域处理fft_generic(B, ImageFFT, to_freq, -1, none, dc_center, complex)*rft_generic (B, ImageFFT, to_freq, none, complex, Width)* 产生高斯过滤*gen_gauss_filter (ImageGauss, 100, 100, 0, n, rft, Width, Height)gen_gauss_filter(ImageGauss, 100, 100, 0, n, dc_center, Width, Height)* 频域中对两个经过傅里叶变换的图像进行卷积操作convol_fft (ImageFFT, ImageGauss, ImageConvol)*从频域转换为空间域fft_generic(ImageConvol, ImageFFT1, from_freq, -1, none, dc_center, byte)*rft_generic (ImageConvol, ImageFFT1, from_freq, none, byte, Width)* 图片相减 mageSubB-ImageFFT1*2100sub_image (B, ImageFFT1, ImageSub, 2, 100)* perform the actual inspection* 按照比例因子缩放zoom_image_factor (ImageSub, ImageZoomed, ScaleFactor, ScaleFactor, constant)* avoid border effects when using lines_gauss()* 由图像转换为区域获取区域get_domain (ImageZoomed, Domain)* 腐蚀运算erosion_rectangle1 (Domain, RegionErosion, 7, 7)* 裁剪区域reduce_domain (ImageZoomed, RegionErosion, ImageReduced)* 空间域检测线条lines_gauss (ImageReduced, Lines, Sigma, Low, High, dark, true, gaussian, true)* 仿射变换hom_mat2d_identity (HomMat2DIdentity)hom_mat2d_scale_local (HomMat2DIdentity, 1 / ScaleFactor, 1 / ScaleFactor, HomMat2DScale)affine_trans_contour_xld (Lines, Defects, HomMat2DScale)* 显示dev_display (Image)dev_display (Defects)if (f 3)disp_continue_message (WindowHandle, black, true)stop ()endif
endforHalcon 频域差分空间域blob特征案例 estimate_background_illumination (B, ImageFFT1) get_image_size (Image, Width, Height)
* 空间域转换为频域
fft_generic(Image, ImageFFT, to_freq, -1, none, dc_center, complex)
*rft_generic (Image, ImageFFT, to_freq, none, complex, Width)
* 产生高斯滤波
*gen_gauss_filter (ImageGauss, 50, 50, 0, n, rft, Width, Height)
gen_gauss_filter(ImageGauss, 50, 50, 0, n, dc_center, Width, Height)
* 卷积
convol_fft (ImageFFT, ImageGauss, ImageConvol)
* 频域转换为空间域
fft_generic(ImageConvol, IlluminationImage, from_freq, 1, sqrt, dc_center, byte)
*rft_generic (ImageConvol, IlluminationImage, from_freq, none, byte, Width)
return ()* This example shows how to detect mura defects
* in highly textured images
*
* 图片预处理
dev_close_window ()
dev_update_off ()
Path : lcd/mura_defects_texture_
read_image (Image, Path 01)
get_image_size (Image, Width, Height)
dev_open_window (0, 0, 640, 480, black, WindowHandle)
set_display_font (WindowHandle, 14, mono, true, false)
dev_set_draw (margin)
dev_set_line_width (3)
dev_set_color (red)
for F : 1 to 2 by 1*1.采集图像read_image (Image, Path F$.2i)decompose3 (Image, R, G, B)* Defects are characterized by dark patches. Hence, by substracting the* estimated background illumination from the original image the* defects become more apparent.* 2.频域处理差分estimate_background_illumination (B, ImageFFT1)* sub_image (B, ImageFFT1, ImageSub, 2, 100)* 3.空间域blob特征* Median filter smoothes out the fine texture, simplifying the following* segmentation and final detection of defects.* 中值滤波median_image (ImageSub, ImageMedian, circle, 9, mirrored)watersheds_threshold (ImageMedian, Basins, 20)* Dark patches corresponding to defects have a very low energy.cooc_feature_image (Basins, ImageMedian, 6, 0, Energy, Correlation, Homogeneity, Contrast)*tuple_find(sgn(Energy-0.05),-1,Indices)*select_obj(Basins, Defects, Indices1)* 如果Energy 0.05Mask : Energy [] 0.05* 获取Mask 个数select_mask_obj (Basins, Defects, Mask)* 显示dev_display (Image)dev_display (Defects)count_obj (Defects, NDefects)disp_message (WindowHandle, NDefects \mura\ defects detected, window, 12, 12, red, true)if (F 2)disp_continue_message (WindowHandle, black, true)stop ()endif
endforHalcon Blob特征处理缺陷 * This programm shows the extraction of surface scratches via
* local thresholding and morphological post-processing
*
dev_update_off ()
dev_close_window ()
*
* Step 1: Acquire image
* 1.采集图片
read_image (Image, surface_scratch)
get_image_size (Image, Width, Height)
dev_open_window_fit_image (Image, 0, 0, Width, Width, WindowID)
set_display_font (WindowID, 16, mono, true, false)
dev_set_draw (margin)
dev_set_line_width (4)
dev_display (Image)
Message : This program shows the extraction of
Message[1] : surface scratches via local thresholding
Message[2] : and morphological post-processing
disp_message (WindowID, Message, window, 12, 12, black, true)
disp_continue_message (WindowID, black, true)
stop ()
*
* Step 2: Segment image
* 2.图像分割
* Using a local threshold
* 均值滤波
mean_image (Image, ImageMean, 7, 7)
* 动态二值化
dyn_threshold (Image, ImageMean, DarkPixels, 5, dark)
*
* Extract connected components
* 形成单独的连通域
connection (DarkPixels, ConnectedRegions)
dev_set_colored (12)
dev_display (Image)
dev_display (ConnectedRegions)
Message : Connected components after image segmentation
Message[1] : using a local threshold.
disp_message (WindowID, Message, window, 12, 12, black, true)
disp_continue_message (WindowID, black, true)
stop ()
*
* Step 3: Process regions
*
* Select large regions
* 筛选面积大小
select_shape (ConnectedRegions, SelectedRegions, area, and, 10, 1000)
dev_display (Image)
dev_display (SelectedRegions)
disp_message (WindowID, Large Regions, window, 12, 12, black, true)
disp_continue_message (WindowID, black, true)
stop ()
*
* Visualize fractioned scratch
* 在新的窗口打开
open_zoom_window (0, round(Width / 2), 2, 303, 137, 496, 3, WindowHandleZoom)
dev_set_color (blue)
dev_display (Image)
dev_display (SelectedRegions)
set_display_font (WindowHandleZoom, 16, mono, true, false)
disp_message (WindowHandleZoom, Fractioned scratches, window, 12, 12, black, true)
disp_continue_message (WindowHandleZoom, black, true)
stop ()
*
* Merge fractioned scratches via morphology
* 联合
union1 (SelectedRegions, RegionUnion)
* 膨胀
dilation_circle (RegionUnion, RegionDilation, 3.5)
dev_display (Image)
dev_display (RegionDilation)
Message : Region of the scratches after dilation
disp_message (WindowHandleZoom, Message, window, 12, 12, black, true)
disp_continue_message (WindowHandleZoom, black, true)
stop ()
*提取骨架
skeleton (RegionDilation, Skeleton)
* 形成单独的连通域
connection (Skeleton, Errors)
dev_set_colored (12)
dev_display (Image)
dev_display (Errors)
Message : Fractioned scratches merged via morphology
disp_message (WindowHandleZoom, Message, window, 12, 12, black, true)
disp_continue_message (WindowHandleZoom, black, true)
stop ()
*
* Distinguish small and large scratches
close_zoom_window (WindowHandleZoom, Width, Height)
* 筛选出区域面积大划痕
select_shape (Errors, Scratches, area, and, 50, 10000)
select_shape (Errors, Dots, area, and, 1, 50)
* 小划痕
dev_display (Image)
dev_set_color (red)
dev_display (Scratches)
dev_set_color (blue)
dev_display (Dots)
Message : Extracted surface scratches
Message[1] : Not categorized as scratches
disp_message (WindowID, Message, window, 440, 310, [red,blue], true)Halcon 频域空间域检测划痕 * This program shows how to detect defects (scratches) in
* an inhomogeneously illuminated surface by filtering in
* the frequency domain.
* First, a suitable bandpass filter is created. Then, the
* input image is fourier transformed and filtered in the
* frequency domain, so that high frequency information is
* enhanced. Finally, it is transformed back to the
* spatial domain and the enhanced defects are post-processed
* by morphology.
* 1.采集图像
dev_update_off ()
dev_close_window ()
read_image (Image, surface_scratch)
* 图像翻转暗点变成亮点亮点变成暗点
invert_image (Image, ImageInverted)
get_image_size (Image, Width, Height)
dev_open_window (0, 0, Width, Height, black, WindowHandle)
set_display_font (WindowHandle, 16, mono, true, false)
dev_display (Image)
*
* Optimize the speed of the fast fourier transform
* Message : Optimize the speed of the fast fourier transform.
* Message[1] : Please wait...
* disp_message (WindowHandle, Message, window, 12, 12, black, true)
* optimize_rft_speed (Width, Height, standard)
* disp_continue_message (WindowHandle, black, true)
* stop ()
* 2.频域变换
* Enhance the scratches by filtering in the frequency domain
* 形成一个正弦滤波器 可以移到中间
*gen_sin_bandpass (ImageBandpass, 0.4, none, rft, Width, Height)
gen_sin_bandpass(ImageBandpass, 0.4, none, dc_center, Width, Height)
*rft_generic (ImageInverted, ImageFFT, to_freq, none, complex, Width)
fft_generic(ImageInverted, ImageFFT, to_freq, -1, none, dc_center, complex)
* 卷积
convol_fft (ImageFFT, ImageBandpass, ImageConvol)
fft_generic(ImageConvol, Lines, from_freq, 1, n, dc_center, byte)
*rft_generic (ImageConvol, Lines, from_freq, n, byte, Width)
*
* Segment the scratches by using morphology
* blob 分析
threshold (Lines, Region, 5, 255)
connection (Region, ConnectedRegions)
select_shape (ConnectedRegions, SelectedRegions, area, and, 5, 5000)
dilation_circle (SelectedRegions, RegionDilation, 5.5)
union1 (RegionDilation, RegionUnion)
reduce_domain (Image, RegionUnion, ImageReduced)
* 获取线
lines_gauss (ImageReduced, LinesXLD, 0.8, 3, 5, dark, false, bar-shaped, false)
* 共线联合
union_collinear_contours_xld (LinesXLD, UnionContours, 40, 3, 3, 0.2, attr_keep)
select_shape_xld (UnionContours, SelectedXLD, contlength, and, 15, 1000)
* xld 转区域
gen_region_contour_xld (SelectedXLD, RegionXLD, filled)
union1 (RegionXLD, RegionUnion)
dilation_circle (RegionUnion, RegionScratches, 10.5)
*
* Display the results
dev_set_draw (margin)
dev_set_line_width (3)
dev_set_colored (12)
dev_display (Image)
dev_display (RegionScratches)Halcon 傅里叶变换转换为功率图频域blob差分 * This program shows how to separate foreground information
* from a disturbing background texture by filtering in the
* frequency domain.
* First, the image is fourier transformed in order to obtain
* its frequency spectrum. Then, we detect the frequency peaks
* corresponding to the disturbing background texture in the
* frequency spectrum. Afterwards, a filter which eliminates those
* frequencies is built and applied to the spectrum. By applying
* the inverse fourier transform to the filtered spectrum, we
* obtain a filtered image from which the disturbing background
* texture was removed.
*
dev_update_off ()
dev_close_window ()
Scale : [1.0,.65]
MinGray : [50,100]
for Index : 0 to 1 by 1* * Read and display the imageread_image (Image, plan_ (Index 1)$02)get_image_size (Image, Width, Height)dev_open_window (0, 0, Width * Scale[Index], Height * Scale[Index], black, WindowHandle)set_display_font (WindowHandle, 14, mono, true, false)dev_set_part (0, 0, Height - 1, Width - 1)dev_display (Image)disp_message (WindowHandle, Original image, window, 12, 12, black, true)* * Perform fft and display spectrum * 优化速度optimize_fft_speed (Width, Height, standard)* * We used fft_generic sqrt and dc_center mainly* for visualization purposes.* To speed up the program, rft_generic should be used;* but of course, the peak detection algorithm has to be* adjusted in this case.fft_generic (Image, ImageFFT, to_freq, -1, sqrt, dc_center, complex)dev_open_window (0, Width * Scale[Index] 7, Width * Scale[Index], Height * Scale[Index], black, WindowHandle1)dev_set_color (red)dev_set_draw (margin)set_display_font (WindowHandle1, 14, mono, true, false)dev_set_part (0, 0, Height - 1, Width - 1)dev_display (ImageFFT)disp_message (WindowHandle1, Fourier spectrum, window, 12, 12, black, true)disp_cont_message (WindowHandle1, black, true)stop ()* * Detect the eight most significant peaks in the spectrum* 转换为功率图power_real (ImageFFT, PowerSpectrum)* 低通滤波binomial_filter (PowerSpectrum, ImageSmooth, 9, 9)* 二值化threshold (ImageSmooth, Region, MinGray[Index], 100000)* 形成单个连通域connection (Region, ConnectedRegions)select_shape (ConnectedRegions, SelectedRegions, area, and, 5, 200)union1 (SelectedRegions, RegionUnion)reduce_domain (ImageSmooth, RegionUnion, ImageReduced)* 获取局部最大值local_max (ImageReduced, LocalMaxima)* * Next, detect peaks one octave higher, i.e., at twice* the frequency of the most significant peaks* 形成一个包凸shape_trans (LocalMaxima, RegionTrans, convex)* Construct ROI band at twice the frequency* 形成一个缩放矩阵的仿射变换hom_mat2d_identity (HomMat2DIdentity)hom_mat2d_scale (HomMat2DIdentity, 2.1, 2.1, Height / 2, Width / 2, HomMat2DScale)affine_trans_region (RegionTrans, RegionTrans1, HomMat2DScale, nearest_neighbor)hom_mat2d_scale (HomMat2DIdentity, 1.9, 1.9, Height / 2, Width / 2, HomMat2DScale)affine_trans_region (RegionTrans, RegionTrans2, HomMat2DScale, nearest_neighbor)* 进行差分difference (RegionTrans1, RegionTrans2, RegionDifference)* Extract the peaks at twice the frequency* 差分区域进行裁剪reduce_domain (ImageSmooth, RegionDifference, ImageReduced)threshold (ImageReduced, Region, 15, 100000)reduce_domain (ImageSmooth, Region, ImageReduced)* 获取局部最大值local_max (ImageReduced, LocalMaxima2)* * Merge the peaks of both octaves and enlarge them to* integrate the relevant frequencies into the filter* 将之前的到的区域和后面的区域进行联合union2 (LocalMaxima, LocalMaxima2, RegionUnion)* 膨胀处理dilation_circle (RegionUnion, RegionDilation, 15.5)* 将获取的区域得到填充paint_region (RegionDilation, ImageFFT, ImageFFTFiltered, 0, fill)dev_display (ImageFFT)dev_display (RegionDilation)disp_message (WindowHandle1, Frequencies of the\nbackground texture, window, 12, 12, black, true)disp_cont_message (WindowHandle1, black, true)stop ()* * Apply the filter and display the results* 反傅里叶变换fft_generic (ImageFFTFiltered, ImageFiltered, from_freq, 1, sqrt, dc_center, byte)dev_display (ImageFiltered)disp_message (WindowHandle1, Filtered image, window, 12, 12, black, true)* dev_open_window (0, 2 * (Width * Scale[Index]) 14, Width * Scale[Index], Height * Scale[Index], black, WindowHandle2)set_display_font (WindowHandle2, 14, mono, true, false)dev_set_part (0, 0, Height - 1, Width - 1)sub_image (Image, ImageFiltered, ImageTexture, 1, 128)dev_display (ImageTexture)disp_message (WindowHandle2, Removed texture, window, 12, 12, black, true)if (Index 1)disp_cont_message (WindowHandle2, black, true)stop ()dev_close_window ()dev_close_window ()dev_close_window ()endif
endforHalcon 训练学习缺陷检测以神经网络mlp为例 create_class_mlp 创建 add_samples_image_class_mlp添加样本 set_rejection_params_class_mlp 设置拒绝参数 train_class_mlp 训练 write_class_mlp 保存 classify_image_class_mlp 识别 * This example program shows how to use the MLP classifier for novelty
* detection to perform a web inspection task. To perform the novelty detection,
* a rejection class is trained internally.
* For the web inspection task, the MLP can subsequently be used to detect
* textures that do not correspond to the texture of the trained good objects.
* 1.采集图像
dev_update_off ()
*
ReadPretrainedClassifier : false
* Uncomment the following line to read the pretrained classifier from
* disk. The training may last up to half a minute.
* ReadPretrainedClassifier : true
SaveClassifier : false
* Uncomment the following line to write the MLP classifier to disk after training.
* SaveClassifier : true
*
read_image (Image, plastic_mesh/plastic_mesh_01)
get_image_size (Image, Width, Height)
dev_close_window ()
dev_open_window (0, 0, Width, Height, black, WindowHandle)
dev_set_color (red)
set_display_font (WindowHandle, 16, mono, true, false)
* 获取系统信息
get_system (example_dir, HalconExamples)
* The texture filters used for the classification will return artifacts at the image
* borders because the images of the plastic mesh to be inspected do not
* contain an integer number of mesh cells. Because this would lead to wrongly
* detected errors at the image borders, we must exclude the area close to the
* image border from the training and classification. This is done with the following
* rectangle. Note that the image is later scaled down by a factor of two.
* 产生矩形窗口
gen_rectangle1 (Rectangle, 10, 10, Height / 2 - 11, Width / 2 - 11)
* 如果没有分类器则创建分类器
if (ReadPretrainedClassifier)* Read the pretrained classifier from disk.dev_display (Image)disp_message (WindowHandle, Reading classifier from disk..., window, 10, 10, black, true)read_class_mlp (HalconExamples /hdevelop/Segmentation/Classification/novelty_detection.gmc, MLPHandle)wait_seconds (1.5)
else* Create the MLP classifier.* 创建分类器* 第一个参数为特征变量的数量和通道数对应- gen_texture_image (ImageZoomed, ImageTexture)create_class_mlp (5, 9, 2, softmax, principal_components, 3, 42, MLPHandle)* The training is based on five images that contain no errors.* 形成一个空的区域gen_empty_region (EmptyRegion)* 将Rectangle和EmptyRegion 放入到 ObjectsConcat 中concat_obj (Rectangle, EmptyRegion, ObjectsConcat)for J : 1 to 5 by 1read_image (Image, plastic_mesh/plastic_mesh_ J$02)* The images are zoomed down because the resolution of the mesh is very* high. This saves a large amount of processing time.* 等比例缩放图片zoom_image_factor (Image, ImageZoomed, 0.5, 0.5, constant)dev_display (ImageZoomed)disp_message (WindowHandle, Adding training samples..., window, 10, 10, black, true)* Generate the texture image.* 形成一个纹理图像gen_texture_image (ImageZoomed, ImageTexture)* Add the samples to the classifier.* 添加样本add_samples_image_class_mlp (ImageTexture, ObjectsConcat, MLPHandle)endfordev_display (ImageZoomed)* Now configure the MLP that a rejection class will be added during training.* 设置不符合要求的参数必须要设置set_rejection_params_class_mlp (MLPHandle, sampling_strategy, hyperbox_ring_around_each_class)set_rejection_params_class_mlp (MLPHandle, rejection_sample_factor, .3)* Train the MLP.disp_message (WindowHandle, Training MLP..., window, 10, 10, black, true)* 训练train_class_mlp (MLPHandle, 200, 1, 0.01, Error, ErrorLog)if (SaveClassifier)*保存分类器write_class_mlp (MLPHandle, HalconExamples /hdevelop/Segmentation/Classification/novelty_detection.gmc)endif
endif
* Now detect errors in the plastic meshes.
dev_set_draw (margin)
dev_set_line_width (3)
for J : 1 to 14 by 1read_image (Image, plastic_mesh/plastic_mesh_ J$02)zoom_image_factor (Image, ImageZoomed, 0.5, 0.5, constant)dev_display (ImageZoomed)dev_set_color (white)dev_display (Rectangle)gen_texture_image (ImageZoomed, ImageTexture)reduce_domain (ImageTexture, Rectangle, ImageTextureReduced)* Classify samples belonging to the trained class with the MLP.* 识别classify_image_class_mlp (ImageTextureReduced, ClassRegions, MLPHandle, 0.5)* Post process the returned raw errors to remove insignificant parts of the* detected errors.select_obj (ClassRegions, Correct, 1)select_obj (ClassRegions, Errors, 2)opening_circle (Errors, ErrorsOpening, 2.5)closing_circle (ErrorsOpening, ErrorsClosing, 12.5)connection (ErrorsClosing, ErrorsConnected)* 筛选出缺陷select_shape (ErrorsConnected, FinalErrors, area, and, 20, 1000000)* 缺陷放入容器count_obj (FinalErrors, NumErrors)dev_set_color (red)dev_display (FinalErrors)if (NumErrors 0)disp_message (WindowHandle, Mesh not OK, window, 10, 10, red, true)elsedisp_message (WindowHandle, Mesh OK, window, 10, 10, forest green, true)endifif (J 14)disp_continue_message (WindowHandle, black, true)endifstop ()
endfor产生滤波算子
gen_texture_image
* The texture image is a five-channel image that contains the result of applying
* five different Laws filters, which basically correspond to first and second
* derivatives, and smoothing them sufficiently.
* 纹理滤波
texture_laws (Image, ImageEL, el, 5, 5)
texture_laws (Image, ImageLE, le, 5, 5)
texture_laws (Image, ImageES, es, 1, 5)
texture_laws (Image, ImageSE, se, 1, 5)
texture_laws (Image, ImageEE, ee, 2, 5)
* 合成一个通道
compose5 (ImageEL, ImageLE, ImageES, ImageSE, ImageEE, ImageLaws)
smooth_image (ImageLaws, ImageTexture, gauss, 3)
return ()
