点文字识别
图片预处理
目标区域定位(一般固定的方式可以用工具画出大致区域)
灰度化
文字矫正(放射变换)
二值化
膨胀腐蚀等,断开联通区域,
如果没有很好的把每个字符分开,考虑Shape_Trans转换成矩形,再分割矩形
分割后的区域s与二值化后的区域求交, 结果集一般就是一个一个字符了
然后将区域与图片(白底黑字)叫个mlp做识别
![](https://images.cnblogs.com/OutliningIndicators/ContractedBlock.gif)
read_image (Image, 'dot') * dots_image (Image, DotImage,15, 'dark', 0) get_image_size (Image, Width, Height) gen_rectangle1 (ROI_0, 72.7942, 110.25, 466.877, 852.086) text_line_orientation (ROI_0, Image, 25, -0.523599, 0.523599, OrientationAngle) vector_angle_to_rigid (Width/2, Height/2, OrientationAngle, Width/2, Height/2, 0, HomMat2D) affine_trans_image (Image, ImageAffineTrans, HomMat2D, 'constant', 'false') dots_image (ImageAffineTrans, DotImage, 7, 'dark', 0) reduce_domain (DotImage, ROI_0, ImageReduced) binary_threshold (ImageReduced, Region, 'max_separability', 'light', UsedThreshold) dilation_rectangle1 (Region, RegionDilation, 10, 10) connection (RegionDilation, ConnectedRegions) shape_trans (ConnectedRegions, RegionTrans, 'rectangle1') select_shape (RegionTrans, SelectedRegions, ['width','height'], 'and', [20.092,18.807], [98.807,100]) partition_rectangle (SelectedRegions, Partitioned, 30, 80) intersection (Partitioned, Region, RegionIntersection) sort_region (RegionIntersection, SortedRegions, 'character', 'true', 'row') read_ocr_class_mlp ('DotPrint_0-9A-Z_Rej.omc', OCRHandle) invert_image (ImageReduced, ImageInvert) do_ocr_multi_class_mlp (SortedRegions, ImageInvert, OCRHandle, Class, Confidence) smallest_rectangle1 (SortedRegions, Row1, Column1, Row2, Column2) count_obj (SortedRegions, Number) dev_display (ImageInvert) for Index := 1 to Number by 1 disp_message (200000, Class[Index-1], 'image', Row2[Index-1], Column1[Index-1], 'red', 'true') endfor * invert_image (Image, ImageInvert)