OCR with CV

ocr.py

import cv2
import cv2.cv as cv
import tesseract

def show(im):
    msg = 'press any key to continue'
    cv2.namedWindow(msg, cv2.WINDOW_NORMAL)
    cv2.imshow(msg, im)
    cv2.waitKey(0)
    cv2.destroyAllWindows()

def main():
    # https://raw.githubusercontent.com/eidge/ruby-captcha-breaker/master/noaa_captchas/noaa_captcha_1.gif
    im = cv2.imread('noaa_captcha_1.jpg')
    h, w, _ = im.shape
    im = cv2.cvtColor(im, cv2.COLOR_BGR2GRAY)
    im[im > 240] = 0
    _, im = cv2.threshold(im, 128, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)
    kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (3, 3))
    im = cv2.dilate(im, kernel, iterations = 2)
    im = cv2.erode(im, kernel, iterations = 2)
    im = cv2.GaussianBlur(im, (3, 3), 0)
    show(im)

    api = tesseract.TessBaseAPI()
    api.Init(".","eng",tesseract.OEM_DEFAULT)
    api.SetVariable("tessedit_char_whitelist", "ABCDEFGHIJKLMNOPQRSTUVWXYZ")
    api.SetPageSegMode(tesseract.PSM_SINGLE_BLOCK)

    h, w = im.shape
    ipl = cv.CreateImageHeader((w,h), cv.IPL_DEPTH_8U, 1)
    cv.SetData(ipl, im.tostring(), im.dtype.itemsize * 1 * (w))
    tesseract.SetCvImage(ipl, api)
    text = api.GetUTF8Text().strip()
    print text

main()