The method of my project is basic but out-of-date, I highly recommend this project, hope you'll enjoy it ::joy::.

Vehicle_Lincense_Plate_Recognition

Data Distribution:

components	number
integers	4679
alphabets	9796
Chinese_letters	3974
percentage	training_set : testing_set == 4 : 1

Usage:

1. To see the data_set accuracy: python restore_lenet.py, and change the parameters of the test_lst in main function.

2. To predict the plate of a image containing part of a car including its vehicle-plate: python predict_plate.py SRC_IMAGE.

3. Any problems on displaying Chinese characters on matplotlib windows, please refer to this.

In saved model:

matches	Validation Accuracy
字母+数字+汉字	98.98%
字母+数字	99.35%
汉字	97.48%
字母	99.40%
数字	99.27%

• Extract Characters:
  • Effects:

    • 

    • 

    •  1. Canny
       2. Denoise
       3. Morphology
       4. Find_contours
       5. get_rects
       6. Select_the_very_rect
       7. Cut_out_the_plate_area

    • 

    • 

    •  8. Border_denoise
       9. Thresholding
       10. Denoise
       11. Kick_out_white_circle_dot
       12. Get_vertical_split_lines
       13. Split_characters

    • 

• Character Recognition:

  • Data Transformation:

    • img <--> tfrecords -> array

  • Construct Lenet-5

  • Restore Lenet-5

    In saved model:

    matches	Validation Accuracy
    字母+数字+汉字	98.98%
    字母+数字	99.35%
    汉字	97.48%
    字母	99.40%
    数字	99.27%
    Accuracies on training set and test set
    Recall Rates on Alphabet, integers and Chinese_letters
    Recall Rates on Alphabets and Integers
    Recall Rates on Chinese_letters
    Recall Rates on Alphabet
    Recall Rates on Integers

• Combination:
  •