Classification-With-HOG-K-mean-Logistic

Intuition

• Build a Classifier to classify metropolitian & countryside
• Using:
  • K-Means clustering, histogram (HOG) for prerocess data and extract features
  • LogisticsRegression for classify metropolitian and countryside
• Data: metropolitian & countryside

Architecture

• Preprocess Images : Using opencv for imread data with function load_image and resize output image (350,350)

• K-Means Clustering : We will extract dominant colors with k = 5 centroid with function setUPKmean

• HOG (Histogram Oriented Gradients) : Construct the features vector for each image using 5 clustered RGB colors, sorted by color frequency.

• LogisticRegression : We use Logistic Regression model to classify the feature vectors built from the HOG. Implement the Gradient Descent method to optimize the model.

   batch_size = 32
   epochs = 10
   threshold = 0.5
   learning_rate = 0.001
   
  

• Accuracy: 73.8 % with 1206 samples ( validation set = 20% , training set = 80%)

• Output K-means Clustering with centroid = 5:

  

• Vector Histogram:

  

• Ouput HOG:

  

Tree Project

  Classification/
                model/
                      model.sav # model results after training
                resources_data/
                        countryside         # contains data of countryside ( have some image for intuition)
                        metropolitian       # contains data of metropilitian ( have some image for intuition)
                        image_predict       # contains image for test predict
                        output_image_test   # some results of k-mean and hog
                predict.py                  # predict some image from resources_data/image_predict with args pathto model
                preprocess_image.py         # contains imread data, fit k-mean and hog
                setup_model.py              # Build LogisicRegression from scatch
                train.py                    # start training with args path data train
                      

USAGE:

Requirements

   skimage
   opencv-python
   numpy
   matplotlib
   sklearn

Training

python train.py --inputcountry path_to_countrydata/ --inputmetro path_to_metropolitiandata/

prediction

python predict.py --model path_to_model/ --image path_to_image/