{"id":38760,"date":"2024-11-02T00:56:52","date_gmt":"2024-11-02T00:56:52","guid":{"rendered":"http:\/\/atmokpo.com\/w\/?p=38760"},"modified":"2024-11-26T06:43:25","modified_gmt":"2024-11-26T06:43:25","slug":"opencv-%ea%b0%95%ec%a2%8c-%ec%96%bc%ea%b5%b4-%eb%a7%a4%ec%b9%ad%ea%b3%bc-%ec%9d%b8%ec%8b%9d%ec%9c%a8-%ed%96%a5%ec%83%81-%ea%b8%b0%eb%b2%95","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/38760\/","title":{"rendered":"OpenCV \uac15\uc88c, \uc5bc\uad74 \ub9e4\uce6d\uacfc \uc778\uc2dd\uc728 \ud5a5\uc0c1 \uae30\ubc95"},"content":{"rendered":"

<\/p>\n

OpenCV(\uc624\ud508 \uc18c\uc2a4 \ucef4\ud4e8\ud130 \ube44\uc804 \ub77c\uc774\ube0c\ub7ec\ub9ac)\ub294 \uc774\ubbf8\uc9c0 \ubc0f \ube44\ub514\uc624\ub97c \ucc98\ub9ac\ud558\ub294 \ub370 \uc0ac\uc6a9\ud560 \uc218 \uc788\ub294 \uac15\ub825\ud55c \ub3c4\uad6c\uc785\ub2c8\ub2e4. \uc774 \uae00\uc5d0\uc11c\ub294 \uc5bc\uad74 \uc778\uc2dd\uc5d0\uc11c \ub9e4\uce6d\uc744 \uc218\ud589\ud558\uace0, \uc778\uc2dd\uc728\uc744 \ud5a5\uc0c1\uc2dc\ud0a4\uae30 \uc704\ud55c \ub2e4\uc591\ud55c \uae30\ubc95\uc744 \uc18c\uac1c\ud569\ub2c8\ub2e4.<\/p>\n

1. \uc5bc\uad74 \uc778\uc2dd\uc758 \uae30\ucd08<\/h2>\n

\uc5bc\uad74 \uc778\uc2dd\uc740 \ucef4\ud4e8\ud130 \ube44\uc804\uc758 \uc911\uc694\ud55c \ubd84\uc57c \uc911 \ud558\ub098\ub85c, \uc774\ubbf8\uc9c0\ub97c \ubd84\uc11d\ud558\uc5ec \uc0ac\ub78c\uc758 \uc5bc\uad74\uc744 \uac10\uc9c0\ud558\uace0 \uc778\uc2dd\ud558\ub294 \uae30\uc220\uc785\ub2c8\ub2e4. OpenCV\uc5d0\uc11c\ub294 Haar Cascade \ubd84\ub958\uae30, LBPH(LBP \ud788\uc2a4\ud1a0\uadf8\ub7a8), DNN(Deep Neural Network) \uac19\uc740 \uc5ec\ub7ec \uac00\uc9c0 \uc5bc\uad74 \uc778\uc2dd \uc54c\uace0\ub9ac\uc998\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4.<\/p>\n

2. \uc5bc\uad74 \ub9e4\uce6d\uc758 \uae30\ucd08<\/h2>\n

\uc5bc\uad74 \ub9e4\uce6d(face matching)\uc740 \ub450 \uac1c\uc758 \uc5bc\uad74 \uc774\ubbf8\uc9c0\uac00 \uac19\uc740 \uc0ac\ub78c\uc778\uc9c0 \uc5ec\ubd80\ub97c \ud655\uc778\ud558\ub294 \uacfc\uc815\uc785\ub2c8\ub2e4. \uc778\uc2dd\ub41c \uc5bc\uad74\uc744 \uc11c\ub85c \ube44\uad50\ud558\uc5ec \uc720\uc0ac\uc131\uc744 \ud3c9\uac00\ud569\ub2c8\ub2e4. \uc774\ub294 \ubcf4\ud1b5 \ud2b9\uc9d5\uc810\uc744 \uc0ac\uc6a9\ud558\uc5ec \ub9e4\uce6d \uc131\ub2a5\uc744 \ud5a5\uc0c1\uc2dc\ud0a4\ub294 \ubc29\uc2dd\uc73c\ub85c \uc218\ud589\ub429\ub2c8\ub2e4.<\/p>\n

2.1 Haar Cascade\ub85c \uc5bc\uad74 \uac10\uc9c0\ud558\uae30<\/h3>\n

Haar Cascade\ub294 \uac04\ub2e8\ud558\uace0 \ube60\ub978 \uc5bc\uad74 \uac10\uc9c0 \uc54c\uace0\ub9ac\uc998\uc73c\ub85c, \uc0ac\uc804\uc5d0 \ud6c8\ub828\ub41c \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc774\ubbf8\uc9c0\uc5d0\uc11c \uc5bc\uad74\uc744 \ucc3e\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c\uc740 \uae30\ubcf8\uc801\uc778 \uc5bc\uad74 \uac10\uc9c0 \uc608\uc81c\uc785\ub2c8\ub2e4.<\/p>\n

import cv2\n\n# Haar Cascade \ubd84\ub958\uae30 \ub85c\ub4dc\nface_cascade = cv2.CascadeClassifier(cv2.data.haarcascades + 'haarcascade_frontalface_default.xml')\n\n# \uc774\ubbf8\uc9c0 \uc77d\uae30\nimage = cv2.imread('face_image.jpg')\ngray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)\n\n# \uc5bc\uad74 \uac10\uc9c0\nfaces = face_cascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5)\n\n# \uc5bc\uad74 \uc0ac\uac01\ud615 \uadf8\ub9ac\uae30\nfor (x, y, w, h) in faces:\n    cv2.rectangle(image, (x, y), (x + w, y + h), (255, 0, 0), 2)\n\n# \uacb0\uacfc \ubcf4\uc5ec\uc8fc\uae30\ncv2.imshow('Detected Faces', image)\ncv2.waitKey(0)\ncv2.destroyAllWindows()<\/code><\/pre>\n
3. \uc5bc\uad74 \uc778\uc2dd \uae30\ubc95<\/h2>\n
Faces can be recognized through different algorithms. In this article, we will focus on the most commonly used methods: Eigenfaces, Fisherfaces, and LBPH.<\/p>\n
3.1 Local Binary Patterns Histograms (LBPH)<\/h3>\n
LBPH\ub294 \ud6a8\uc728\uc801\uc774\uace0 \uac15\ub825\ud55c \ubc29\ubc95\uc73c\ub85c, \uc5bc\uad74 \uc774\ubbf8\uc9c0\uc758 \uc9c0\uc5ed \ud2b9\uc9d5\uc744 \ucd94\ucd9c\ud558\uace0 \ud788\uc2a4\ud1a0\uadf8\ub7a8\uc744 \uc0dd\uc131\ud558\uc5ec \uc5bc\uad74 \uc778\uc2dd\uc744 \uc218\ud589\ud569\ub2c8\ub2e4. \uc774\ub294 \uc870\uba85 \ubcc0\ud654\ub098 \uc5bc\uad74\uc758 \ud68c\uc804\uc5d0 \uac15\ud55c \ud2b9\uc9d5\uc744 \uac00\uc9c0\ubbc0\ub85c \ub9ce\uc740 \uc2e4\uc81c \uc751\uc6a9\uc5d0\uc11c \uc0ac\uc6a9\ub429\ub2c8\ub2e4.<\/p>\n
import cv2\nimport numpy as np\n\n# LBPH \uc5bc\uad74 \uc778\uc2dd\uae30 \ucd08\uae30\ud654\nrecognizer = cv2.face.LBPHFaceRecognizer_create()\n\n# \ud6c8\ub828 \ub370\uc774\ud130 \uc900\ube44\ndef prepare_training_data(data_folder_path):\n    # \uc774\ubbf8\uc9c0\uc640 \ub808\uc774\ube14\uc744 \uc800\uc7a5\ud560 \ub9ac\uc2a4\ud2b8\n    faces = []\n    labels = []\n    label = 0\n\n    for name in os.listdir(data_folder_path):\n        person_folder_path = os.path.join(data_folder_path, name)\n        if not os.path.isdir(person_folder_path):\n            continue\n\n        for filename in os.listdir(person_folder_path):\n            img_path = os.path.join(person_folder_path, filename)\n            image = cv2.imread(img_path, cv2.IMREAD_GRAYSCALE)\n            faces.append(image)\n            labels.append(label)\n\n        label += 1\n\n    return faces, np.array(labels)\n\n# \ud6c8\ub828 \ub370\uc774\ud130 \ub85c\ub4dc\nfaces, labels = prepare_training_data('path_to_training_data_folder')\n\n# \ubaa8\ub378 \ud6c8\ub828\nrecognizer.train(faces, labels)\n\n# \ubaa8\ub378 \uc800\uc7a5\nrecognizer.save('face_recognizer.yml')<\/code><\/pre>\n
3.2 \uc5bc\uad74 \uc778\uc2dd \uc2e4\ud589<\/h3>\n
\ud6c8\ub828\uc774 \uc644\ub8cc\ub41c \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc0c8\ub85c\uc6b4 \uc774\ubbf8\uc9c0\ub97c \uc778\uc2dd\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
def recognize_person(image, recognizer):\n    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)\n    faces = face_cascade.detectMultiScale(gray, scaleFactor=1.1, minNeighbors=5)\n\n    for (x, y, w, h) in faces:\n        id_, confidence = recognizer.predict(gray[y:y+h, x:x+w])\n        if confidence < 100:\n            print(\"ID : {}, Confidence : {}\".format(id_, confidence))\n        else:\n            print(\"Unknown\")\n\n# \ud14c\uc2a4\ud2b8 \uc774\ubbf8\uc9c0\uc5d0\uc11c \uc5bc\uad74 \uc778\uc2dd\ntest_image = cv2.imread('test_face_image.jpg')\nrecognize_person(test_image, recognizer)<\/code><\/pre>\n
4. \uc778\uc2dd\uc728 \ud5a5\uc0c1 \uae30\ubc95<\/h2>\n
\uc5bc\uad74 \uc778\uc2dd \uc2dc\uc2a4\ud15c\uc740 \ub2e4\uc591\ud55c \uc694\uc778\uc73c\ub85c \uc778\ud574 \uc778\uc2dd\uc728\uc774 \ub0ae\uc544\uc9c8 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc778\uc2dd\uc728\uc744 \ud5a5\uc0c1\uc2dc\ud0a4\uae30 \uc704\ud55c \uba87 \uac00\uc9c0 \uae30\ubc95\uc744 \uc18c\uac1c\ud569\ub2c8\ub2e4.<\/p>\n
4.1 \ub370\uc774\ud130 \uc99d\uac15(Data Augmentation)<\/h3>\n
\ub370\uc774\ud130 \uc99d\uac15\uc740 \ud6c8\ub828 \ub370\uc774\ud130\uc14b\uc758 \ubcc0\ud615\uc744 \ud1b5\ud574 \uc778\uc2dd \ubaa8\ub378\uc758 \uc77c\ubc18\ud654 \ub2a5\ub825\uc744 \ud5a5\uc0c1\uc2dc\ud0a4\ub294 \uae30\ubc95\uc785\ub2c8\ub2e4. \uc774\ubbf8\uc9c0 \ud68c\uc804, \ud06c\uae30 \uc870\uc808, \uc0c9\uc0c1 \ubcc0\ud615 \ub4f1\uc758 \ubc29\ubc95\uc744 \uc0ac\uc6a9\ud569\ub2c8\ub2e4.<\/p>\n
from keras.preprocessing.image import ImageDataGenerator\n\ndatagen = ImageDataGenerator(rotation_range=20, width_shift_range=0.2,\n                             height_shift_range=0.2, shear_range=0.2,\n                             zoom_range=0.2, horizontal_flip=True,\n                             fill_mode='nearest')\n\n# \uc608\uc2dc: \ud2b9\uc815 \uc774\ubbf8\uc9c0\uc5d0 \ub370\uc774\ud130 \uc99d\uac15 \uc218\ud589\nimage = cv2.imread('face_image.jpg')\nimage = np.expand_dims(image, axis=0)\n\ni = 0\nfor batch in datagen.flow(image, batch_size=1):\n    cv2.imshow('Augmented Image', batch[0])\n    i += 1\n    if i > 20:\n        break\ncv2.waitKey(0)\ncv2.destroyAllWindows()<\/code><\/pre>\n
4.2 \ub2e4\uc591\ud55c \ub370\uc774\ud130 \uc0ac\uc6a9<\/h3>\n
\ub2e4\uc591\ud55c \uc870\uba85 \uc870\uac74, \uac01\ub3c4 \ubc0f \uc778\uc885\uc758 \uc778\ubb3c \uc0ac\uc9c4\uc744 \ud3ec\ud568\ud55c \ub113\uc740 \ud6c8\ub828 \ub370\uc774\ud130\uc14b\uc740 \uc778\uc2dd\uc728\uc744 \ub192\uc774\ub294 \ub370 \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4. \uc774\ub294 \ubaa8\ub378\uc774 \ub2e4\uc591\ud55c \ubcc0\ud654\ub97c \ud559\uc2b5\ud560 \uc218 \uc788\ub3c4\ub85d \ub3d5\uc2b5\ub2c8\ub2e4.<\/p>\n
4.3 \ud569\uc131\uacf1 \uc2e0\uacbd\ub9dd(Convolutional Neural Networks)<\/h3>\n
\ub525\ub7ec\ub2dd \uae30\ubc18\uc758 \ud569\uc131\uacf1 \uc2e0\uacbd\ub9dd(CNN)\uc744 \uc0ac\uc6a9\ud558\uba74 \ub354 \ub192\uc740 \uc815\ud655\ub3c4\ub85c \uc5bc\uad74 \uc778\uc2dd\uc744 \ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. TensorFlow \ub610\ub294 PyTorch\uc640 \uac19\uc740 \ud504\ub808\uc784\uc6cc\ud06c\ub97c \uc0ac\uc6a9\ud558\uc5ec \ubcf4\ub2e4 \ubcf5\uc7a1\ud55c \ubaa8\ub378\uc744 \uad6c\ucd95\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
import tensorflow as tf\nfrom tensorflow.keras import layers, models\n\n# \uac04\ub2e8\ud55c CNN \ubaa8\ub378 \uc815\uc758\nmodel = models.Sequential([\n    layers.Conv2D(32, (3, 3), activation='relu', input_shape=(48, 48, 1)),\n    layers.MaxPooling2D((2, 2)),\n    layers.Conv2D(64, (3, 3), activation='relu'),\n    layers.MaxPooling2D((2, 2)),\n    layers.Flatten(),\n    layers.Dense(128, activation='relu'),\n    layers.Dense(num_classes, activation='softmax')\n])\n\nmodel.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=['accuracy'])\n# \ubaa8\ub378 \ud6c8\ub828\nmodel.fit(train_images, train_labels, epochs=10)\n<\/code><\/pre>\n
5. \uacb0\ub860<\/h2>\n
OpenCV\ub294 \uc5bc\uad74 \uc778\uc2dd \ubc0f \ub9e4\uce6d \uc791\uc5c5\uc744 \uc218\ud589\ud558\ub294 \ub370 \ub9e4\uc6b0 \uc720\uc6a9\ud55c \ub3c4\uad6c\uc774\uba70, \uc5ec\ub7ec \uc54c\uace0\ub9ac\uc998\uacfc \uae30\ubc95\uc744 \ud1b5\ud574 \uc778\uc2dd\uc728\uc744 \ub192\uc77c \uc218 \uc788\uc2b5\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \uae30\ubcf8 \uae30\ubc95\uacfc \uc778\uc2dd\uc728 \ud5a5\uc0c1\uc5d0 \ub300\ud55c \ub2e4\uc591\ud55c \ubc29\ubc95\uc744 \uc0b4\ud3b4\ubcf4\uc558\uace0, \uc5ec\ub7ec\ubd84\uc774 \uc9c1\uc811 \uad6c\ud604\ud574 \ubcf4\ub294 \ub370 \ub3c4\uc6c0\uc774 \ub418\uc5c8\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4.<\/p>\n
6. \ucc38\uace0 \uc790\ub8cc<\/h2>\n