<\/a><\/p>\n\uc774 \uc774\ubbf8\uc9c0\uc5d0\ub294 \uc190\uc73c\ub85c \uc791\uc131\ud55c 5000\uac1c\uc758 0~9\uae4c\uc9c0\uc758 \ubb38\uc790\uac00 \ub2f4\uaca8 \uc788\uc2b5\ub2c8\ub2e4. \ubb38\uc790 \ud558\ub098\ub2f9 500\uac1c\uc529 \uae30\ub85d\ub418\uc5b4 \uc788\uc73c\uba70, \uac00\ub85c\uc640 \uc138\ub85c\ub85c \uac01\uac01 100\uac1c\uc529, 5\uac1c\uc529 \ud45c\uae30\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ubbf8\uc9c0\uc5d0\uc11c \ubb38\uc790 \ud558\ub098\uac00 \ucc28\uc9c0\ud558\ub294 \ud06c\uae30\ub294 20×20 \ud53d\uc140\uc785\ub2c8\ub2e4. \uac00\uc7a5 \uba3c\uc800 \uc774 \uc774\ubbf8\uc9c0\uc5d0\uc11c 5000\uac1c\uc758 \ubb38\uc790 \ub2e8\uc704\ub85c \uc798\ub77c\ub0b4\uc57c \ud569\ub2c8\ub2e4. \uadf8\ub9ac\uace0 \uc774 20×20 \ud53d\uc140 \ud06c\uae30 \ubb38\uc790 \uc774\ubbf8\uc9c0\ub97c 400 \ud06c\uae30\uc758 \ub2e8\uc77c \ud589\uc73c\ub85c \ub9cc\ub4ed\ub2c8\ub2e4. \uc774 \ub370\uc774\ud130\uac00 \ubaa8\ub4e0 \ud53d\uc140\uc5d0 \ub300\ud55c \ud654\uc18c\uac12\uc744 \uac00\uc9c0\ub294 \ud53c\uccd0\uc14b(Feature Set)\uc785\ub2c8\ub2e4. \uc6b0\ub9ac\uac00 \uc0dd\uc131\ud560 \uc218 \uc788\ub294 \uac00\uc7a5 \ub2e8\uc21c\ud55c \ud53c\uccd0\uc14b\uc785\ub2c8\ub2e4. \uc774 \ub370\uc774\ud130\uc5d0\uc11c \uac01 \ubb38\uc790\uc758 250\uac1c\uc5d0 \ud574\ub2f9\ud558\ub294 \ubd80\ubd84\uc740 train_data\ub85c \uc0ac\uc6a9\ud558\uace0 \ub098\uba38\uc9c0 250\uac1c\ub294 test_data\ub85c \uc0ac\uc6a9\ud569\ub2c8\ub2e4.<\/p>\n
\uc774\uc81c \ucf54\ub4dc\ub97c \uc791\uc131\ud574 \ubcf4\uba74..<\/p>\n
\r\nimport numpy as np\r\nimport cv2\r\n\r\nimg = cv2.imread('.\/data\/digits.png')\r\ngray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)\r\n\r\n# Now we split the image to 5000 cells, each 20x20 size\r\ncells = [np.hsplit(row,100) for row in np.vsplit(gray,50)]\r\n\r\n# Make it into a Numpy array. It size will be (50,100,20,20)\r\nx = np.array(cells)\r\n\r\n# Now we prepare train_data and test_data.\r\ntrain = x[:,:50].reshape(-1,400).astype(np.float32) # Size = (2500,400)\r\ntest = x[:,50:100].reshape(-1,400).astype(np.float32) # Size = (2500,400)\r\n\r\n# Create labels for train and test data\r\nk = np.arange(10)\r\ntrain_labels = np.repeat(k,250)[:,np.newaxis]\r\ntest_labels = train_labels.copy()\r\n\r\n# Initiate kNN, train the data, then test it with test data for k=1\r\nknn = cv2.ml.KNearest_create()\r\nknn.train(train, cv2.ml.ROW_SAMPLE, train_labels)\r\nret, result, neighbors, dist = knn.findNearest(test, k=5)\r\n\r\n# Now we check the accuracy of classification\r\n# For that, compare the result with test_labels and check which are wrong\r\nmatches = result==test_labels\r\ncorrect = np.count_nonzero(matches)\r\naccuracy = correct*100.0\/result.size\r\nprint(accuracy)\r\n<\/pre>\n\ucf54\ub4dc\ub97c \uc124\uba85\ud558\uba74, 8\ubc88\uc740 digits.png \uc774\ubbf8\uc9c0\ub97c \uac00\ub85c\ub85c 100\uac1c, \uc138\ub85c\ub85c 50\ub85c \uc798\ub77c \uc870\uac01\ub0b4\uc5b4 cells \ubcc0\uc218\uc5d0 \uc800\uc7a5\ud558\ub294\ub370, \uac01\uac01\uc758 \uc870\uac01 \uc774\ubbf8\uc9c0\uc5d0\ub294 \ubb38\uc790 \ud558\ub098\uac00 \ub2f4\uaca8 \uc788\uc2b5\ub2c8\ub2e4. 11\ubc88 \ucf54\ub4dc\ub294 \ub2e4\uc2dc \uc774 cells\ub97c NumPy\uc758 \ubc30\uc5f4\ub85c \ub9cc\ub4e4\uc5b4 x \ubcc0\uc218\uc5d0 \uc800\uc7a5\ud569\ub2c8\ub2e4. 14\ubc88 \ucf54\ub4dc\ub294 \ubc30\uc5f4 x \uc911 \uc808\ubc18\uc744 \ud559\uc2b5 \ub370\uc774\ud130\ub85c \uc0ac\uc6a9\ud558\uace0 \ub098\uba38\uc9c0 \uc808\ubc18\uc744 \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\ub85c \uc0ac\uc6a9\ud558\uace0\uc790 \uac01\uac01 train\uacfc test \ubcc0\uc218\uc5d0 \ub2f4\uc2b5\ub2c8\ub2e4. train \ubcc0\uc218\uc5d0 \uc800\uc7a5\ub41c \ubb38\uc790\uc5d0 \ub300\ud574 0~9\uae4c\uc9c0\uc758 \uac12\uc73c\ub85c \ub77c\ubca8\ub9c1\ud574\uc918\uc57c \ud558\ub294\ub370, 18-19\ubc88 \ucf54\ub4dc\uac00 \uadf8\uc5d0 \ud574\ub2f9\ud569\ub2c8\ub2e4. \ubc14\ub85c \uc774 train \ub370\uc774\ud130\uc640 train_labels \ub370\uc774\ud130\uac00 \ud559\uc2b5 \ub370\uc774\ud130\ub77c\uace0 \ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub807\uac8c \ud559\uc2b5\ub41c \ub370\uc774\ud130\ub97c \ud1a0\ub300\ub85c test \ubcc0\uc218\uc5d0 \uc800\uc7a5\ub41c \ubb38\uc790\ub4e4\uc774 0~9\uae4c\uc9c0 \uc911 \ubb34\uc5c7\uc5d0 \ud574\ub2f9\ud558\ub294\uc9c0 kNN \uc54c\uace0\ub9ac\uc998\uc73c\ub85c \ud30c\uc545\ud558\ub294 \uac83\uc774 23~25\ubc88 \ucf54\ub4dc\uc785\ub2c8\ub2e4. \ucd5c\uc885\uc801\uc73c\ub85c \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uac00 \uc815\ud655\ud788 \uc778\uc2dd\ub418\uc5c8\ub294\uc9c0 \ud655\uc778\ud558\ub294 \ucf54\ub4dc\uac00 29~32\ubc88 \ucf54\ub4dc\uc785\ub2c8\ub2e4. \ucd9c\ub825\uac12\uc740 91.76\uc778\ub370, \uc989 \uc131\uacf5\ub960\uc774 91.76%\ub77c\ub294 \uc758\ubbf8\uc785\ub2c8\ub2e4.<\/p>\n
\uc778\uc2dd \uc815\ud655\ub3c4\ub97c \uac1c\uc120\ud558\uae30 \uc704\ud574\uc11c\ub294 \uc778\uc2dd\uc774 \uc2e4\ud328\ud55c \ub370\uc774\ud130\ub97c \ud559\uc2b5\uc2dc\ucf1c train \ubcc0\uc218\uc5d0 \ucd94\uac00\ud558\uace0 \ub2e4\uc74c\uc5d0 \uc774 \ubcc0\uc218\ub97c \uc7ac\ud65c\uc6a9\ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc774\ub97c \uc704\ud574 \ud30c\uc77c\uc5d0 \uc800\uc7a5\ud558\uace0 \ub2e4\uc74c\uc5d0 \uc800\uc7a5\ub41c \ud30c\uc77c\ub85c\ubd80\ud130 \ubd88\ub7ec\uc624\ub294 \ud568\uc218\uac00 \ud544\uc694\ud569\ub2c8\ub2e4. \ud559\uc2b5 \ub370\uc774\ud130\ub97c \uc800\uc7a5\ud558\ub294 \uc608\ub294 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4.<\/p>\n
\r\nnp.savez('knn_data.npz',train=train, train_labels=train_labels)\r\n<\/pre>\n\ub370\uc774\ud130 \ud30c\uc77c\uc744 \ubd88\ub7ec\uc624\ub294 \uc608\ub294 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4.<\/p>\n
\r\nwith np.load('knn_data.npz') as data:\r\n print data.files\r\n train = data['train']\r\n train_labels = data['train_labels']\r\n<\/pre>\n","protected":false},"excerpt":{"rendered":"\uc774 \uae00\uc758 \uc6d0\ubb38\uc740 https:\/\/opencv-python-tutroals.readthedocs.io\/en\/latest\/py_tutorials\/py_ml\/py_knn\/py_knn_opencv\/py_knn_opencv.html \uc785\ub2c8\ub2e4. \uc774 \uae00\uc740 \uc190\uc73c\ub85c \uadf8\ub9b0 \uae00\uc790\ub97c \ud310\ub3c5\ud558\ub294 \uae30\ub2a5\uc5d0 \ub300\ud55c \uac83\uc785\ub2c8\ub2e4. \uc774\ub97c \uc704\ud574 \uba87\uac00\uc9c0 \ud6c8\ub828 \ub370\uc774\ud130(train_data)\uc640 \uc2dc\ud5d8 \ub370\uc774\ud130(test_data)\uac00 \ud544\uc694\ud569\ub2c8\ub2e4. \uc544\ub798\uc640 \uac19\uc740 2000×1000 \ud53d\uc140 \ud06c\uae30\uc758 digits.png \ud30c\uc77c\uc744 \uc0ac\uc6a9\ud569\ub2c8\ub2e4. \uc774 \uc774\ubbf8\uc9c0\uc5d0\ub294 \uc190\uc73c\ub85c \uc791\uc131\ud55c 5000\uac1c\uc758 0~9\uae4c\uc9c0\uc758 \ubb38\uc790\uac00 \ub2f4\uaca8 \uc788\uc2b5\ub2c8\ub2e4. \ubb38\uc790 \ud558\ub098\ub2f9 500\uac1c\uc529 \uae30\ub85d\ub418\uc5b4 \uc788\uc73c\uba70, \uac00\ub85c\uc640 \uc138\ub85c\ub85c \uac01\uac01 100\uac1c\uc529, 5\uac1c\uc529 \ud45c\uae30\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ubbf8\uc9c0\uc5d0\uc11c \ubb38\uc790 \ud558\ub098\uac00 \ucc28\uc9c0\ud558\ub294 … <\/p>\n