training.py

from sklearn.neighbors import KNeighborsClassifier
from sklearn.metrics import accuracy_score
from skimage import exposure
from skimage import feature
from imutils import paths
from keras.datasets import fashion_mnist
import numpy as np
from matplotlib import pyplot as plt
import argparse
import imutils
import cv2
from sklearn.externals import joblib
#setup
app = argparse.ArgumentParser()
app.add_argument("-d", "--training", required=True, help="Path to the logos training dataset")
app.add_argument("-t", "--test", required=True, help="Path to the test dataset")
args = vars(app.parse_args())

print ("[INFO] extracting features by using HOG...")
data = []
labels = []

for imagePath in paths.list_images(args["training"]):
	
	make = imagePath.split("/")[-2] 
	image = cv2.imread(imagePath)
	#cv2.imshow("Original Image", image)
	imgUMat = image
	gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
	cv2.imshow("Gray Image", gray)
	(thresh, im_bw) = cv2.threshold(gray, 128, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)
	edged = cv2.Canny(gray,50, 100)
	cv2.imshow("After canny edge detection", edged)
	edgedBW = cv2.Canny(im_bw,50, 100)
	logo = cv2.resize(edged, (200, 100))
	(H, hogImage) = feature.hog(logo, orientations=9, pixels_per_cell=(10, 10),
		cells_per_block=(2, 2), transform_sqrt=True, block_norm="L1",visualise=True)
 

	data.append(H)
	labels.append(make)
	hogImage = exposure.rescale_intensity(hogImage, out_range=(0, 256))
	hogImage = hogImage.astype("uint8")
	cv2.imshow("HOG Image", hogImage)


	
	image = cv2.resize(image, (200, 100))
	cv2.putText(image, str(labels), (10, 10), cv2.FONT_HERSHEY_SIMPLEX, 1.0,
		(0, 255, 0), 2)
	cv2.imshow("Image", image)


# "train" the nearest neighbors classifier
print("[INFO] training classifier...")
model = KNeighborsClassifier(n_neighbors=2)
model.fit(data, labels)
print("[INFO] evaluating and saving...")
joblib.dump(model, 'saved_model.pkl') 


for (i, imagePath) in enumerate(paths.list_images(args["test"])):
	
	image = cv2.imread(imagePath)
	gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
	logo = cv2.resize(gray, (200, 100))
	(H, hogImage) = feature.hog(logo, orientations=9, pixels_per_cell=(10, 10),
		cells_per_block=(2, 2), transform_sqrt=True, block_norm="L1", visualise=True)
	pred = model.predict(H.reshape(1, -1))[0]
	predlist = []
	print(pred)
	
	# accuracy = accuracy_score(make, pred)
	# print("Accuracy")
	# print(accuracy)

	
	# hogImage = exposure.rescale_intensity(hogImage, out_range=(0, 256))
	# hogImage = hogImage.astype("uint8")
	# cv2.imshow("HOG Image #{}".format(i + 1), hogImage)


	
	# image = cv2.resize(image, (200, 100))
	# cv2.putText(image, pred.title(), (10, 35), cv2.FONT_HERSHEY_SIMPLEX, 1.0,
	# 	(0, 255, 0), 3)

	# cv2.imshow("Test Image #{}".format(i + 1), image)
	# k = cv2.waitKey(5000)