th1nkk1d / gocr

OCR implementation with Golang

gOCR

OCR Implementation with Golang

This project was mainly focus on Thai language and tested with some Thai characters, English characters and Arabic digits

Implemented functions

• Template generator
• Image adjustment (automatic thresholding and median filter)
• Rows and characters segmentation (horizontal histrogram and blob coloring)
• Character recognition (image resizing and template matching)

How to use

Generate the Templates

gocr --gentemp <charlist.txt> <fontfile.ttf>

Templates of character list in charlist.txt with fontfile.ttf fontface will be generated in /templates directory

OCR

gocr <charlist.txt> <inputImage>

The inputImage and charlist.txt (the same one that use to generate templates) will be use for OCR. The output text file and images of each process will be saved in /outputs directory.

Example

Input image

Median filter + automatic thresholding

Rows segmentation

Characters segmentation

Text output

234sohappy
sawasdeewansao
ลูกกิดมาล้าว
สุขใจหรรษา
อิอิzaa55บวก

NOTE: You can notice some incorrect character recognition here

Terminal output

Test cases

No.	Font style	Capture method	Correctness
1	Same as template	Screen capture	58/58
2	Same as template, double font size	Screen capture	58/58
3	Same as template	Phone camera	57/58
4	Different font	Screen capture	32/58
5	Different font	Phone camera	34/58

What should be improve

• Support multi-object characters (ex. ะ, ญ)
• Better similar character recognition ( Dictionary? Machine Learning? Character type consideration?)
• A space in between word recognition
• Order recognition for lower and upper alphabet
• Handle system error (ex. can't open file)

Main Resources

• Golang : an open source programming language
• GoCV : allow Go to access OpenCV 3
• OpenCV : a computer vision library

About the project

Created by Withee Poositasai

Simple Thai OCR Project, CPE489 Image Processing and Computer Vision, KMUTT

gocr's Issues

Help please

github.com/Th1nkK1D/gocr

Th1nkK1D\gocr\common.go:85:2: too many arguments to return
have (gocv.Mat, error)
want (gocv.Mat)
Th1nkK1D\gocr\main.go:98:21: cannot use rowImg (type gocv.Mat) as type *gocv.Mat in argument to gocv.Rectangle
Th1nkK1D\gocr\rowsegment.go:113:12: cannot use img (type gocv.Mat) as type *gocv.Mat in argument to gocv.Line
Th1nkK1D\gocr\rowsegment.go:114:12: cannot use img (type gocv.Mat) as type *gocv.Mat in argument to gocv.Line
Th1nkK1D\gocr\threshold.go:49:16: cannot use imgOut (type gocv.Mat) as type *gocv.Mat in argument to gocv.Threshold

undefined

Hi display this to compile, help

.\main.go:37:5: undefined: GenTemplate
.\main.go:37:17: undefined: ReadCharList
.\main.go:56:11: undefined: AutoThreshold
.\main.go:63:15: undefined: MedianFilter
.\main.go:63:28: undefined: GetImgArray
.\main.go:64:11: undefined: GetImgMat
.\main.go:71:18: undefined: ReadTemplate
.\main.go:71:31: undefined: ReadCharList
.\main.go:76:19: undefined: SplitLine
.\main.go:78:5: undefined: DrawRowSegment
.\main.go:78:5: too many errors

Fonts

In the reader it seems that you got problems with different fonts.
Did you use the trained data from the OCR part of opencv ?
They have trained on 4000000 documents and thousands of fonts