halcon学习2识别蓝底白字车牌号码

1. 想针对蓝底白字的车牌做一个提取蓝色区域的适应度更广的halcon程序，进行了第二次车牌识别程序编写
2. 
       图1 蓝底白字，有倾斜度，两个固定钉子
3. 
           图2  蓝底白字，无倾斜度，四个固定钉
4. 相对上一篇的程序，使用了select_shape算子提取蓝色区域，再进行抠图，程序如下
   

   dev_update_window ('off')
   *读取图像
   * Image Acquisition 01: Code generated by Image Acquisition 01
   read_image (Image, 'C:/Users/Administrator/Pictures/车牌5.JPG')
   * read_image (Image, 'C:/Users/Administrator/Pictures/车牌4.PNG')
   get_image_size(Image, Width, Height)
   
   *打开图像窗口，并显示图像
   dev_close_window()
   dev_open_window(0, 0, Width, Height, 'black', WindowHandle)
   dev_display (Image)
   disp_continue_message (WindowHandle, 'black', 'true')
   stop ()
   
   *按颜色分层，灰度化，二值化
   decompose3 (Image, Red, Green, Blue)
   trans_from_rgb(Red, Green, Blue, ImageResult1, ImageResult2, ImageResult3, 'hsv')
   threshold (ImageResult2, Regions, 225, 255)
   
   
   *填成方块
   fill_up(Regions, RegionFillUp)
   connection(RegionFillUp, ConnectedRegions)
   *当图片中的车牌像素不一样时，select_shape就不能顺利按照area选择
   select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 14368.3, 50000)
   * select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 1503.56, 10000)
   shape_trans(SelectedRegions,RegionTrans,'rectangle2')
   disp_continue_message (WindowHandle, 'black', 'true')
   stop ()
   
   *测量车牌是否有倾斜
   orientation_region(RegionTrans, Phi)
   area_center(RegionTrans, Area, Row, Column)
   *Phi<0,Angle=0;Phi>0,Angle=3.14
   vector_angle_to_rigid(Row, Column, Phi, Row, Column, 3.14, HomMat2D)
   *将原图中的车牌进行旋转
   affine_trans_image(Image, ImageAffineTrans, HomMat2D, 'constant', 'false')
   *对区域旋转
   affine_trans_region(RegionTrans, RegionAffineTrans, HomMat2D, 'nearest_neighbor')
   *对被旋转之后的车牌进行裁剪
   reduce_domain(ImageAffineTrans, RegionAffineTrans, ImageReduced)
   
   *halcon训练时是拿黑色字符训练的，所以识别时也要拿黑色的字符区域去进行识别，即转换为白底黑字
   invert_image(ImageReduced, ImageInvert)
   
   rgb1_to_gray(ImageReduced, GrayImage)
   threshold (GrayImage, Regions1, 180, 255)
   connection(Regions1, ConnectedRegions1)
   
   select_shape (ConnectedRegions1, SelectedRegions1, 'area', 'and', 567.78, 1000)
   sort_region(SelectedRegions1, SortedRegions, 'character', 'true', 'column')
   
   read_ocr_class_mlp('Document_0-9A-Z_NoRej.omc',OCRHandle)
   do_ocr_multi_class_mlp(SortedRegions, ImageInvert, OCRHandle, Class, Confidence)
   
   dev_display (Image)
   disp_message(WindowHandle, Class, 'Image', 120, 60, 'green', 'true')

5. 图1检测结果
   
   

   

6. 图2检测两处报错后，放弃尝试
   

   

   上图为图2车牌被钉子影响，无法做下一步算子处理，下图为图1 理想结果

7. 1）大小不一样，蓝色区域的灰度值数量不一样；2）钉子将车牌截成两段
8. 总结：1）要对图像预处理，将图像大小，分辨率等进行归一化，才能有效实现适应度增加；2）要增加判断语句，进行膨胀或开处理等形态学处理。还要继续努力