def predict(img_name, model, dir_name, new_size=None):
image_loader = ImageLoader(dir_name=dir_name)
img = image_loader.get_one_image(new_size=new_size, path=img_name)
image_preprocessor = ImagePreprocessing(img_size=new_size)
gray_scaled = image_loader.rgb2gray(img["data"])
# sobeled = image_loader.sobel_image(gray_scaled)
# deskewed = deskew(sobeled, new_size[0])
hog = image_preprocessor.get_hog(np.uint8(gray_scaled))
return model.predict([hog])
def train_sgd(n_epoch, n_samples, clf, train_dir, img_size, classes):
image_loader = ImageLoader(dir_name=train_dir)
image_preproc = ImagePreprocessing(img_size=img_size)
for epoch in range(n_epoch):
for i in range(n_samples):
img = image_loader.get_one_image(labels=classes, new_size=img_size)
gray_scaled = image_loader.rgb2gray(img["data"])
train_sample = image_preproc.simple_preproc(gray_scaled)
clf.partial_fit([train_sample], [img["target"]],
classes=list(classes.values()))
del img
if epoch % 10 == 0:
test_data = []
test_labels = []
for _ in range(1000):
test_img = image_loader.get_one_image(labels=classes,
new_size=img_size)
gray_scaled = image_loader.rgb2gray(test_img["data"])
test_data.append(image_preproc.simple_preproc(gray_scaled))
test_labels.append(test_img["target"])
print(
f"accuracy {accuracy_score(clf.predict(test_data), test_labels)*100}"
)
del test_data, test_labels
def load_images(n, img_dir, image_size=None, class_labels=None):
data = []
img_labels = []
image_loader = ImageLoader(dir_name=img_dir)
image_preprocessor = ImagePreprocessing(img_size=image_size)
for i in range(n):
img = image_loader.get_one_image(labels=class_labels,
new_size=image_size)
gray_scaled = image_loader.rgb2gray(img["data"])
# sobeled = image_loader.sobel_image(gray_scaled)
# deskewed = deskew(sobeled, image_size[0])
hog = image_preprocessor.get_hog(np.uint8(gray_scaled))
data.append(hog)
img_labels.append(img["target"])
return np.array(data), np.array(img_labels)
