识别盲文字符

上述代码是使用HALCON软件编写的脚本，用于处理图像并提取盲文字符。HALCON是一款用于机器视觉和图像分析的高级软件，广泛应用于工业自动化、医学图像处理等领域。这段代码主要解决的问题是利用光度立体技术（Photometric Stereo）来重建盲文字符的高度场，并对其进行分割和识别

dev_close_window ()
dev_update_off ()
read_image (Image, ‘photometric_stereo/embossed_01’)
get_image_size (Image, WindowWidth, WindowHeight)
dev_open_window_fit_size (0, 0, WindowWidth, WindowHeight, WindowWidth * 0.6, WindowHeight * 0.6, WindowHandle)
set_display_font (WindowHandle, 14, ‘mono’, ‘true’, ‘false’)
dev_display (Image)
disp_continue_message (WindowHandle, ‘black’, ‘true’)
disp_message (WindowHandle, ‘Extract braille characters using\nphotometric stereo. For this approach\nsix different light setups were used.’, ‘window’, 10, 10, ‘black’, ‘true’)
stop ()
*

• Show input images (illuminated differently)
  FileNames := []
  for I := 1 to 6 by 1
  FileNames := [FileNames,‘photometric_stereo/embossed_’ + I$‘.2’]
  endfor
  read_image (Images, FileNames)
  count_obj (Images, NumImages)
  for I := 1 to NumImages by 1
  select_obj (Images, ObjectSelected, I)
  dev_display (ObjectSelected)
  wait_seconds (0.5)
  endfor
• Apply photometric stereo to determine the height field
  photometric_stereo (Images, Height, Gradient, Albedo, [70,50,70,50,70,50], [0,0,90,90,180,180], ‘all’, ‘rft_cyclic’, [], [])
  mean_image (Height, ImageMean, 61, 61)
  sub_image (Height, ImageMean, ImageSub, 1, 0)
  dev_display (ImageSub)
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  disp_message (WindowHandle, ‘Reconstruct the height field\napplying photometric stereo’, ‘window’, 10, 10, ‘black’, ‘true’)
  stop ()
  dev_display (Image)
• Shade a height field
  disp_message (WindowHandle, ‘Shade the height field\nof the braille dots’, ‘window’, 10, 10, ‘black’, ‘true’)
  wait_seconds (2)
  for I := 0 to 360 by 10
  shade_height_field (ImageSub, ShadedFace, 45, I, 1, 0, ‘false’)
  dev_display (ShadedFace)
  endfor
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  stop ()
• Segment brailles dots
  threshold (ImageSub, Region, 0.35, 1000)
  dev_set_draw (‘margin’)
  dev_set_line_width (4)
  dev_set_color (‘green’)
  dev_display (Image)
  dev_display (Region)
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  disp_message (WindowHandle, ‘Segment single dots’, ‘window’, 10, 10, ‘black’, ‘true’)
  stop ()
• Group brailles dots to characters
  closing_rectangle1 (Region, RegionClosing, 200, 60)
  connection (RegionClosing, ConnectedRegions)
  count_obj (ConnectedRegions, Number)
  dev_display (Image)
  dev_set_colored (12)
  for I := 1 to Number by 1
  select_obj (ConnectedRegions, ObjectSelected, I)
  intersection (ObjectSelected, Region, RegionIntersection)
  partition_dynamic (RegionIntersection, Partitioned, 135, 20)
  dev_display (Partitioned)
  endfor
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  disp_message (WindowHandle, ‘Finally, group the dots that describe\nindividual braille characters’, ‘window’, 10, 10, ‘black’, ‘true’)
  stop ()
• Another way is to use the mean curvature to extract the dots. Note that the Gradient must be calulated only.
• This version is faster because no height field must be calculated.
  photometric_stereo (Images, Height1, Gradient1, Albedo1, [70,50,70,50,70,50], [0,0,90,90,180,180], ‘gradient’, ‘rft_cyclic’, [], [])
  derivate_vector_field (Gradient1, Result, 1, ‘mean_curvature’)
  dev_display (Result)
  disp_message (WindowHandle, ‘Alternative Solution: Use gauss curvature of gradient image\nto segment the dots’, ‘window’, 10, 10, ‘black’, ‘true’)
  stop ()
  threshold (Result, Region, -10, 0)
  connection (Region, ConnectedRegions1)
  fill_up (ConnectedRegions1, RegionFillUp)
  select_shape (RegionFillUp, SelectedRegions1, [‘area’,‘circularity’], ‘and’, [150,0.8], [99999,1])
  count_obj (SelectedRegions1, Number)
  union1 (SelectedRegions1, RegionUnion)
  closing_rectangle1 (RegionUnion, RegionClosing1, 200, 60)
  connection (RegionClosing1, ConnectedRegions)
  count_obj (ConnectedRegions, Number)
  dev_display (Image)
  dev_set_colored (12)
  for I := 1 to Number by 1
  select_obj (ConnectedRegions, ObjectSelected, I)
  intersection (ObjectSelected, RegionUnion, RegionIntersection)
  partition_dynamic (RegionIntersection, Partitioned, 135, 20)
  dev_display (Partitioned)
  endfor
  disp_continue_message (WindowHandle, ‘black’, ‘true’)
  disp_message (WindowHandle, ‘Finally, group the dots that describe\nindividual braille characters’, ‘window’, 10, 10, ‘black’, ‘true’)
  
  
  

以下是代码的详细解释：

dev_close_window() 和 dev_update_off()：关闭当前显示窗口并关闭自动更新显示。

read_image(Image, ‘photometric_stereo/embossed_1’)：读取名为 ‘embossed_1’ 的图像文件。

get_image_size(Image, WindowWidth, WindowHeight)：获取图像的宽度和高度。

dev_open_window_fit_size(…)：根据图像大小打开一个新的显示窗口。

set_display_font(…)：设置显示窗口的字体样式。

dev_display(Image)：在新窗口中显示图像。

disp_continue_message(…) 和 disp_message(…)：在窗口中显示继续消息和文本消息。

stop()：暂停执行，等待用户操作。

接下来的循环读取六张不同光照条件下的图像，并逐个显示它们。

photometric_stereo(…)：应用光度立体技术来确定高度场、梯度和反照率。

mean_image(…) 和 sub_image(…)：计算高度场的平均图像，并从原始高度场中减去平均图像。

shade_height_field(…)：对高度场进行阴影处理，以更清晰地显示盲文点的三维形状。

threshold(…) 和 connection(…)：对图像进行阈值处理，并将连续的区域连接起来。

closing_rectangle1(…) 和 select_shape(…)：使用闭运算和形状选择来进一步处理和筛选区域。

union1(…) 和 partition_dynamic(…)：将区域合并并动态分割。

photometric_stereo(…) 和 derivate_vector_field(…)：另一种解决方案，使用梯度图像的高斯曲率来分割点。

threshold(…)、connection(…) 和 fill_up(…)：对高斯曲率图像进行阈值处理，连接区域，并填充空洞。

select_obj(…) 和 dev_display(…)：选择对象并显示结果。

这段代码通过光度立体技术重建盲文字符的高度场，然后通过图像处理技术（如阈值分割、区域连接、形态学操作等）来分割和识别盲文字符。最后，提供了一种替代方案，使用梯度图像的高斯曲率来直接分割点，这种方法更快，因为不需要计算整个高度场。