TensorFlow 예제 - Keras + OpenCV를 사용하여 실제 손글씨 숫자 인식

첫번째 코드를 실행시켜서 손글씨 숫자 MNIST 데이터로 훈련을 시킨 후, 두번째 코드에서 실제 손글씨 이미지를 로드하여 인식시켜보았습니다. 

테스트시 아래 이미지를 사용했습니다. 

 

실행결과는 다음과 같았습니다.

 

[0 1 2 3 4 5 6 7 8 9]

import tensorflow as tf mnist = tf.keras.datasets.mnist (x_train, y_train),(x_test, y_test) = mnist.load_data() x_train, x_test = x_train / 255.0, x_test / 255.0 model = tf.keras.models.Sequential([   tf.keras.layers.Flatten(),   tf.keras.layers.Dense(512, activation=tf.nn.relu),   tf.keras.layers.Dropout(0.2),   tf.keras.layers.Dense(10, activation=tf.nn.softmax) ]) model.compile(optimizer='adam',               loss='sparse_categorical_crossentropy',               metrics=['accuracy']) model.fit(x_train, y_train, epochs=5) model.evaluate(x_test, y_test) #-------------------------------------------------------------------------- model.save_weights('/tmp/mnist_checkpoint')

import tensorflow as tf import cv2 import numpy as np import math from scipy import ndimage def getBestShift(img):     cy,cx = ndimage.measurements.center_of_mass(img)     rows,cols = img.shape     shiftx = np.round(cols/2.0-cx).astype(int)     shifty = np.round(rows/2.0-cy).astype(int)     return shiftx,shifty def shift(img,sx,sy):     rows,cols = img.shape     M = np.float32([[1,0,sx],[0,1,sy]])     shifted = cv2.warpAffine(img,M,(cols,rows))     return shifted # 10개의 이미지를 위한 배열을 생성 images = np.zeros((10, 784)) i = 0 for no in range(10):  # 10개의 이미지를 입력 받음     gray = cv2.imread(str(no) + ".png", 0)     gray = cv2.resize(gray, (28, 28), interpolation=cv2.INTER_AREA)     cv2.imwrite("0_" + str(no) + ".png", gray)     (thresh, gray) = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)     cv2.imwrite("1_" + str(no) + ".png", gray)     while np.sum(gray[0]) == 0:         gray = gray[1:]     while np.sum(gray[:, 0]) == 0:         gray = np.delete(gray, 0, 1)     while np.sum(gray[-1]) == 0:         gray = gray[:-1]     while np.sum(gray[:, -1]) == 0:         gray = np.delete(gray, -1, 1)     rows, cols = gray.shape     cv2.imwrite("2_" + str(no) + ".png", gray)     if rows > cols:         factor = 20.0 / rows         rows = 20         cols = int(round(cols * factor))         # first cols than rows         gray = cv2.resize(gray, (cols, rows))     else:         factor = 20.0 / cols         cols = 20         rows = int(round(rows * factor))         # first cols than rows         gray = cv2.resize(gray, (cols, rows))     colsPadding = (int(math.ceil((28 - cols) / 2.0)), int(math.floor((28 - cols) / 2.0)))     rowsPadding = (int(math.ceil((28 - rows) / 2.0)), int(math.floor((28 - rows) / 2.0)))     gray = np.lib.pad(gray, (rowsPadding, colsPadding), 'constant')     cv2.imwrite("3_" + str(no) + ".png", gray)     shiftx, shifty = getBestShift(gray)     shifted = shift(gray, shiftx, shifty)     gray = shifted     cv2.imwrite("4_" + str(no) + ".png", gray)     flatten = gray.flatten() / 255.0     images[i] = flatten     i += 1 model = tf.keras.models.Sequential([   tf.keras.layers.Flatten(),   tf.keras.layers.Dense(512, activation=tf.nn.relu),   tf.keras.layers.Dropout(0.2),   tf.keras.layers.Dense(10, activation=tf.nn.softmax) ]) model.load_weights('/tmp/mnist_checkpoint') predictions = model.predict(images) with tf.Session() as sess:     print(tf.argmax(predictions, 1).eval())

참고한 코드

 

https://github.com/aymericdamien/TensorFlow-Examples/blob/master/examples/3_NeuralNetworks/neural_network_raw.py  

 

https://github.com/opensourcesblog/tensorflow-mnist/blob/master/mnist.py 

 

https://www.tensorflow.org/tutorials/ 

마지막 업데이트 - 2018. 11. 7