Classifier.py

00001 #!/usr/bin/env python
00002 
00003 import numpy as np
00004 import re
00005 import scipy.stats
00006 from sklearn import svm, metrics, model_selection
00007 from skimage import io, feature, filters, exposure, color
00008 from joblib import dump, load
00009 
00010 
00011 
00012 class ImageClassifier:
00013 
00014     def __init__(self):
00015         self.classifer = None
00016 
00017     def imread_convert(self, f):
00018         return io.imread(f).astype(np.uint8)
00019 
00020     def load_data_from_folder(self, dir):
00021         # read all images into an image collection
00022         ic = io.ImageCollection(dir + "*.bmp", load_func=self.imread_convert)
00023 
00024         # create one large array of image data
00025         data = io.concatenate_images(ic)
00026 
00027         # extract labels from image names
00028         labels = np.array(ic.files)
00029         for i, f in enumerate(labels):
00030             m = re.search("_", f)
00031             labels[i] = f[len(dir):m.start()]
00032 
00033         return (data, labels)
00034 
00035     def extract_image_features(self, data):
00036         # Please do not modify the header above
00037 
00038         # extract feature vector from image data
00039 
00040         ########################
00041         ######## YOUR CODE HERE
00042         ########################
00043 
00044         feature_data = []
00045 
00046         for image in data:
00047             feature_data.append(
00048                 feature.hog(image, pixels_per_cell=(25, 25), cells_per_block=(3, 3), feature_vector=True,
00049                             block_norm="L2-Hys"))
00050 
00051         # Please do not modify the return type below
00052         return (feature_data)
00053 
00054     def train_classifier(self, train_data, train_labels):
00055         # Please do not modify the header above
00056 
00057         # train model and save the trained model to self.classifier
00058 
00059         ########################
00060         ######## YOUR CODE HERE
00061         ########################
00062 
00063         # this param distribution dictionary was obtained from the sklearn documentation page
00064         param_distribution = {'C': scipy.stats.expon(scale=100), 'gamma': scipy.stats.expon(scale=.1),
00065                               'kernel': ['rbf']}
00066 
00067         model = model_selection.RandomizedSearchCV(svm.SVC(), param_distribution, cv=3, n_iter=6)
00068 
00069         self.classifer = model.fit(train_data, train_labels)
00070 
00071     def predict_labels(self, data):
00072         # Please do not modify the header
00073 
00074         # predict labels of test data using trained model in self.classifier
00075         # the code below expects output to be stored in predicted_labels
00076 
00077         ########################
00078         ######## YOUR CODE HERE
00079         ########################
00080 
00081         predicted_labels = self.classifer.predict(data)
00082 
00083         # Please do not modify the return type below
00084         return predicted_labels
00085 
00086 
00087 def main():
00088     img_clf = ImageClassifier()
00089 
00090     # load images
00091     print("Loading Images")
00092     (train_raw, train_labels) = img_clf.load_data_from_folder('./train/')
00093     (test_raw, test_labels) = img_clf.load_data_from_folder('./test/')
00094 
00095     # convert images into features
00096     print("Extracting features from images")
00097     train_data = img_clf.extract_image_features(train_raw)
00098     test_data = img_clf.extract_image_features(test_raw)
00099 
00100     # train model and test on training data
00101     print("Training Model")
00102     img_clf.train_classifier(train_data, train_labels)
00103     predicted_labels = img_clf.predict_labels(train_data)
00104     print("\nTraining results")
00105     print("=============================")
00106     print("Confusion Matrix:\n", metrics.confusion_matrix(train_labels, predicted_labels))
00107     print("Accuracy: ", metrics.accuracy_score(train_labels, predicted_labels))
00108     print("F1 score: ", metrics.f1_score(train_labels, predicted_labels, average='micro'))
00109 
00110     # test model
00111     predicted_labels = img_clf.predict_labels(test_data)
00112     print("\nTest results")
00113     print("=============================")
00114     print("Confusion Matrix:\n", metrics.confusion_matrix(test_labels, predicted_labels))
00115     print("Accuracy: ", metrics.accuracy_score(test_labels, predicted_labels))
00116     print("F1 score: ", metrics.f1_score(test_labels, predicted_labels, average='micro'))
00117 
00118     dump(img_clf, "model.joblib")
00119 
00120 
00121 if __name__ == "__main__":
00122     main()