字符OCR识别，新人求助

我是新人练习OCR字符识别，字符一直无法识别，不知道是问题出现在哪里，求助各位高手指点迷津谢谢！

1. dev_close_window ()
   
2. dev_open_window (0, 0, 512, 512, 'black', WindowHandle)
   
3. 
   
4. read_image (Image, 'C:/Users/WP/Desktop/测试图像.jpg')
   
5. rotate_image (Image, Image1, 4, 'constant')
   
6. *画矩形
   
7. draw_rectangle1 (WindowHandle, Row1, Column1, Row2, Column2)
   
8. gen_rectangle1 (ModelRegion, Row1, Column1, Row2, Column2)
   
9. *分割图像
   
10. reduce_domain (Image1, ModelRegion, ModelRegionImage)
    
11. 
    
12. *将分割后的图像放大显示
    
13. crop_domain (ModelRegionImage, ImagePart)
    
14. *图像增强
    
15. equ_histo_image (ImagePart, ImageEquHisto)
    
16. *增强图像对比度
    
17. * emphasize (ImagePart, ImageEmphasize, 7, 7, 1)
    
18. *图像进行旋转
    
19. * rotate_image (ImageEmphasize, ImagePart, 4, 'constant')
    
20. *将图片写入保存本地
    
21. write_image (ImagePart, 'tiff', 0, 'C:/Users/WP/Desktop/test.bmp')
    
22. decompose3 (ImagePart, r, g, b)
    
23. trans_from_rgb (r, g, b, h, s, v, 'hsv')
    
24. 
    
25. *进行阈值分割
    
26. threshold (b, Regions, 38, 255)
    
27. connection (Regions, ConnectedRegions)
    
28. select_shape (ConnectedRegions, SelectedRegions, ['row','column'], 'and', [69.44,86.11], [212.04,499.07])
    
29. union1 (SelectedRegions, RegionUnion)
    
30. shape_trans (RegionUnion, RegionTrans, 'rectangle1')
    
31. * partition_rectangle (RegionTrans, Partitioned, 50, 50)
    
32. 
    
33. dev_display (b)
    
34. intersection (b, RegionUnion, RegionIntersection)
    
35. sort_region (RegionIntersection, SortedRegions, 'character', 'true', 'column')
    
36. 
    
37. read_ocr_class_mlp ('Pharma_0-9A-Z_NoRej.omc', OCRHandle)
    
38. do_ocr_multi_class_mlp (ConnectedRegions, Image1, OCRHandle, Class, Confidence)
    
39. 
    
40. disp_message (WindowHandle, Class, 'Image', 80, Column1 + 80, 'green', 'false')
    

复制代码

正常识别是白底黑字的，你看你图像反转了么

图片是斜的我旋转了一下

方便的话贴一下原图

原图太大了我处理了一下不过也可以

夜问夜问 发表于 2023-11-28 14:02
原图太大了我处理了一下不过也可以

这边你看下效果，只做了上面大的字体识别

我是不是图像没有处理好呀

夜问夜问 发表于 2023-11-28 14:16
我是不是图像没有处理好呀

你这个例子不太需要什么处理，我只是在我之前测试OCR代码上改了一点点就能识别出来了

夜问夜问 发表于 2023-11-28 14:16
我是不是图像没有处理好呀

1. read_image(Image, '3')
   
2. get_image_size(Image, Width, Height)
   
3. dev_open_window(0, 0, Width, Height, 'black', WindowHandle)
   
4. dev_display(Image)
   
5. decompose3 (Image, Image1, Image2, Image3)
   
6. draw_rectangle1 (WindowHandle, Row1, Column1, Row2, Column2)
   
7. gen_rectangle1 (Rectangle, Row1, Column1, Row2, Column2)
   
8. reduce_domain (Image3, Rectangle, ImageReduced)
   
9. invert_image (ImageReduced, ImageInvert)
   
10. threshold (ImageReduced, Region, 90, 255)
    
11. connection (Region, ConnectedRegions)
    
12. select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 10, 50)
    
13. sort_region(SelectedRegions, SortedRegions1, 'character', 'true', 'row')
    
14. shape_trans (SortedRegions1, Characters, 'rectangle1')
    
15. read_ocr_class_mlp('Document_A-Z+_NoRej.omc', OCRHandle)
    
16. T:=''
    
17. do_ocr_word_mlp(Characters, ImageInvert, OCRHandle, T, 3, 1, Class, Confidence, Word, Score)
    
18. clear_ocr_class_mlp(OCRHandle)
    
19. set_display_font(WindowHandle, 16, 'mono', 'true', 'false')
    
20. disp_message(WindowHandle, Word, 'window', 10, 10, 'black', 'true')

复制代码

代码给你可以看看

Ggg3434 发表于 2023-11-28 14:20
read_image(Image, '3')
get_image_size(Image, Width, Height)
dev_open_window(0, 0, Width, Height, ' ...

还是大佬比较厉害，我这新手都搞不清楚图像处理遵循那些步骤