Я пытаюсь написать Ocr для чисел в Python, используя OpenCV и Knn-Algorithm. Код работает довольно хорошо, но я хочу расширить свои входные данные, чтобы перейти к рукописным цифрам. Обучение / ввод данных работает следующим образом: вы запускаете сценарий, где путь к изображению, затем изображение открывается с прямоугольником вокруг числа, и вам нужно нажать число на клавиатуре. В конце он сохраняет классификации и сведенные изображения в текстовый файл.
Проблема в том, что он переопределяет старые текстовые файлы, поэтому эти данные теряются.
Возможно / работает ли добавление нового и старого текстового файла?
Это код:
import sys
import numpy as np
import cv2
import os
MIN_CONTOUR_AREA = 35
RESIZED_IMAGE_WIDTH = 20
RESIZED_IMAGE_HEIGHT = 30
def main():
imgTrainingNumbers = cv2.imread("training_chars.png")
if imgTrainingNumbers is None: # if image was not read successfully
print "error: image not read from file \n\n" # print error message to std out
os.system("pause") # pause so user can see error message
return # and exit function (which exits program)
# end if
imgGray = cv2.cvtColor(imgTrainingNumbers, cv2.COLOR_BGR2GRAY) # get grayscale image
imgBlurred = cv2.GaussianBlur(imgGray, (5,5), 0) # blur
# filter image from grayscale to black and white
imgThresh = cv2.adaptiveThreshold(imgBlurred, # input image
255, # make pixels that pass the threshold full white
cv2.ADAPTIVE_THRESH_GAUSSIAN_C, # use gaussian rather than mean, seems to give better results
cv2.THRESH_BINARY_INV, # invert so foreground will be white, background will be black
11, # size of a pixel neighborhood used to calculate threshold value
2) # constant subtracted from the mean or weighted mean
cv2.imshow("imgThresh", imgThresh) # show threshold image for reference
imgThreshCopy = imgThresh.copy() # make a copy of the thresh image, this in necessary b/c findContours modifies the image
imgContours, npaContours, npaHierarchy = cv2.findContours(imgThreshCopy, # input image, make sure to use a copy since the function will modify this image in the course of finding contours
cv2.RETR_EXTERNAL, # retrieve the outermost contours only
cv2.CHAIN_APPROX_SIMPLE) # compress horizontal, vertical, and diagonal segments and leave only their end points
# declare empty numpy array, we will use this to write to file later
# zero rows, enough cols to hold all image data
npaFlattenedImages = np.empty((0, RESIZED_IMAGE_WIDTH * RESIZED_IMAGE_HEIGHT))
intClassifications = [] # declare empty classifications list, this will be our list of how we are classifying our chars from user input, we will write to file at the end
# possible chars we are interested in are digits 0 through 9, put these in list intValidChars
intValidChars = [ord('0'), ord('1'), ord('2'), ord('3'), ord('4'), ord('5'), ord('6'), ord('7'), ord('8'), ord('9'),
ord('-'), ord('o'), ord('c')]
for npaContour in npaContours: # for each contour
if cv2.contourArea(npaContour) > MIN_CONTOUR_AREA: # if contour is big enough to consider
[intX, intY, intW, intH] = cv2.boundingRect(npaContour) # get and break out bounding rect
# draw rectangle around each contour as we ask user for input
cv2.rectangle(imgTrainingNumbers, # draw rectangle on original training image
(intX, intY), # upper left corner
(intX+intW,intY+intH), # lower right corner
(0, 0, 255), # red
2) # thickness
imgROI = imgThresh[intY:intY+intH, intX:intX+intW] # crop char out of threshold image
imgROIResized = cv2.resize(imgROI, (RESIZED_IMAGE_WIDTH, RESIZED_IMAGE_HEIGHT)) # resize image, this will be more consistent for recognition and storage
cv2.imshow("imgROI", imgROI) # show cropped out char for reference
cv2.imshow("imgROIResized", imgROIResized) # show resized image for reference
cv2.imshow("training_numbers.png", imgTrainingNumbers) # show training numbers image, this will now have red rectangles drawn on it
intChar = cv2.waitKey(0) # get key press
if intChar == 27: # if esc key was pressed
sys.exit() # exit program
elif intChar in intValidChars: # else if the char is in the list of chars we are looking for . . .
intClassifications.append(intChar) # append classification char to integer list of chars (we will convert to float later before writing to file)
npaFlattenedImage = imgROIResized.reshape((1, RESIZED_IMAGE_WIDTH * RESIZED_IMAGE_HEIGHT)) # flatten image to 1d numpy array so we can write to file later
npaFlattenedImages = np.append(npaFlattenedImages, npaFlattenedImage, 0) # add current flattened impage numpy array to list of flattened image numpy arrays
# end if
# end if
# end for
fltClassifications = np.array(intClassifications, np.float32) # convert classifications list of ints to numpy array of floats
npaClassifications = fltClassifications.reshape((fltClassifications.size, 1)) # flatten numpy array of floats to 1d so we can write to file later
print "\n\ntraining complete !!\n"
np.savetxt("classifications.txt", npaClassifications) # write flattened images to file
np.savetxt("flattened_images.txt", npaFlattenedImages) #
cv2.destroyAllWindows()
return
Спасибо