Data Augmentation (데이터 증강)

from PIL import Image # 이미지를 생성하고 조작하기 위한 라이브러리 import numpy as np # 배열 및 행렬 연산을 위한 라이브러리 def create_random_image(size=(100, 100)): return Image.fromarray(np.random.randint(0, 256, size + (3,), dtype=np.uint8)) def show_image(image, title="Image"): import matplotlib.pyplot as plt # 함수 내부에서 import하여 필요할 때만 로드 plt.imshow(image) plt.title(title) plt.axis('off') plt.show() image = create_random_image() # create_random_image((200, 200)) show_image(image, "Random Image")

def rotate_image(image, angle): return image.rotate(angle) rotated_image = rotate_image(image, 45) show_image(rotated_image, "Rotated Image")

from PIL import ImageOps def flip_image(image, direction): if direction == 'horizontal': return ImageOps.mirror(image) elif direction == 'vertical': return ImageOps.flip(image) flipped_image = flip_image(image, 'horizontal') show_image(flipped_image, "Flipped Image")

def translate_image(image, x, y): array = np.array(image) # np.array() 입력 데이터를 NumPy 배열로 변환 translated_array = np.roll(array, shift=(y, x), axis=(0, 1)) # shift=(y, x): 이미지를 x축 방향으로 30픽셀, y축 방향으로 50픽셀 이동, # axis=(0, 1) : 기준 축을 0,1로 이동 return Image.fromarray(translated_array) translated_image = translate_image(image, 50, 30) show_image(translated_image, "Translated Image")

def resize_image(image, scale): width, height = image.size return image.resize((int(width * scale), int(height * scale))) print(f"Original Image Size: {original_image.size}") resized_image = resize_image(original_image, 1.5) print(f"Resized Image Size: {resized_image.size}") show_image(resized_image, "Resized Image")

from PIL import ImageEnhance # PIL(Python Imaging Library)은 파이썬에서 이미지를 처리하기 위한 라이브러리 def change_brightness(image, factor): enhancer = ImageEnhance.Brightness(image) return enhancer.enhance(factor) brightened_image = change_brightness(image, 1.5) show_image(brightened_image, "Brightened Image")

def add_noise(image): array = np.array(image) noise = np.random.normal(0, 25, array.shape).astype(np.uint8) # 0 : 노이즈의 평균값, 25: 노이즈의 평균편차 noisy_image = np.clip(array + noise, 0, 255).astype(np.uint8) return Image.fromarray(noisy_image) # 노이즈가 추가된 배열을 다시 이미지 객체로 변환하여 반환합니다. noisy_image = add_noise(image) # 노이즈가 추가된 이미지를 화면에 표시합니다. show_image(noisy_image, "Noisy Image")

# 1. 필요한 라이브러리 설치 및 불러오기 from PIL import Image, ImageEnhance, ImageOps import numpy as np import matplotlib.pyplot as plt # 2. 이미지 불러오기 image_path = 'Lenna.jpg' # 이미지 경로를 입력하세요 image = Image.open(image_path) # 3. 이미지 증강 기법 적용 # 이미지 회전 def rotate_image(image, angle): return image.rotate(angle) # 이미지 뒤집기 def flip_image(image, direction): if direction == 'horizontal': return ImageOps.mirror(image) elif direction == 'vertical': return ImageOps.flip(image) # 이미지 이동 (이미지 크기 변경 없이 이동 효과를 위해 배경 추가) def translate_image(image, x, y): array = np.array(image) translated = np.roll(array, shift=(y, x), axis=(0, 1)) return Image.fromarray(translated) # 이미지 확대/축소 def resize_image(image, scale): width, height = image.size return image.resize((int(width * scale), int(height * scale))) # 색상 변형 def change_brightness(image, factor): enhancer = ImageEnhance.Brightness(image) return enhancer.enhance(factor) # 노이즈 추가 def add_noise(image): array = np.array(image) noise = np.random.normal(0, 25, array.shape) noisy_image = np.clip(array + noise, 0, 255).astype(np.uint8) return Image.fromarray(noisy_image) # 증강된 이미지들 보여주기 fig, ax = plt.subplots(2, 3, figsize=(15, 10)) ax[0, 0].imshow(rotate_image(image, 45)) ax[0, 0].set_title("Rotated Image") ax[0, 1].imshow(flip_image(image, 'horizontal')) ax[0, 1].set_title("Flipped Image") ax[0, 2].imshow(translate_image(image, 50, 30)) ax[0, 2].set_title("Translated Image") ax[1, 0].imshow(resize_image(image, 1.5)) ax[1, 0].set_title("Resized Image") ax[1, 1].imshow(change_brightness(image, 1.5)) ax[1, 1].set_title("Brightness Changed") ax[1, 2].imshow(add_noise(image)) ax[1, 2].set_title("Noisy Image") plt.tight_layout() plt.show()

# 1. 필요한 라이브러리 설치 및 불러오기 import tensorflow as tf from tensorflow.keras.preprocessing.image import ImageDataGenerator from tensorflow.keras import layers, models # 2. 데이터셋 불러오기 및 증강 (train_images, train_labels), (test_images, test_labels) = tf.keras.datasets.cifar10.load_data() # 데이터 증강을 위한 ImageDataGenerator 설정 datagen = ImageDataGenerator( rotation_range=20, width_shift_range=0.2, height_shift_range=0.2, horizontal_flip=True, brightness_range=[0.8, 1.2], zoom_range=0.2 ) # 데이터셋 준비 train_images = train_images.astype('float32') / 255.0 test_images = test_images.astype('float32') / 255.0 # 3. 모델 생성 및 학습 model = models.Sequential([ layers.Conv2D(32, (3, 3), activation='relu', input_shape=(32, 32, 3)), layers.MaxPooling2D((2, 2)), layers.Conv2D(64, (3, 3), activation='relu'), layers.MaxPooling2D((2, 2)), layers.Conv2D(64, (3, 3), activation='relu'), layers.Flatten(), layers.Dense(64, activation='relu'), layers.Dense(10, activation='softmax') ]) model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy']) # 증강된 데이터를 사용하여 모델 학습 train_generator = datagen.flow(train_images, train_labels, batch_size=64) model.fit(train_generator, epochs=10, validation_data=(test_images, test_labels)) # 4. 모델 평가 test_loss, test_acc = model.evaluate(test_images, test_labels) print(f'Test accuracy: {test_acc}')

# 필요한 라이브러리 임포트 import numpy as np import matplotlib.pyplot as plt from tensorflow.keras.datasets import cifar10 from tensorflow.keras.preprocessing.image import ImageDataGenerator # CIFAR-10 데이터셋 로드 (x_train, y_train), (x_test, y_test) = cifar10.load_data() # ImageDataGenerator 설정 datagen = ImageDataGenerator( rotation_range=20, width_shift_range=0.2, height_shift_range=0.2, shear_range=0.2, zoom_range=0.2, horizontal_flip=True, fill_mode='nearest' ) # 데이터 피팅 datagen.fit(x_train) # 증강된 이미지 생성 augmented_images = datagen.flow(x_train[:1], batch_size=1) # 증강된 이미지 시각화 plt.figure(figsize=(10, 10)) for i in range(9): plt.subplot(3, 3, i+1) batch = next(augmented_images) image = batch[0].astype('uint8') plt.imshow(image) plt.axis('off') plt.show()