pca.py

import numpy as np
import cv2
import os

# path for image database
image_db = 'images'

# loading the image into numpy array
image_files = []
for f in os.listdir(image_db):
    if f.endswith('.pgm'):
        image_files.append(f)
images = []
for file in image_files:
    img = cv2.imread(os.path.join(image_db, file),0)  # load image in grayscale
    img = cv2.resize(img, (70, 70))  # resize image to lower resolution it is still large as it will be 1X10000 better to use 50X60
    images.append(img)

# converting images to numpy array
X = np.array(images)

# reshappign images to 1D array
X_flat = X.reshape(X.shape[0], -1)

# calculating mean
mean = np.mean(X_flat, axis=0)

# finding mean shift image
X_centered = X_flat - mean
print('came here 1') # cheking the completeness of the steps successfully

# covariance matrix
cov = np.cov(X_centered.T) 
print('came here 2')
# calculating eigenvectors and eigenvalues of covariance matrix taking most of calculation time when image dimension is large
eigenvalues, eigenvectors = np.linalg.eig(cov)
print('came here 3')

# sort eigenvectors by eigenvalues in descending order
idx = np.argsort(eigenvalues)[::-1]
eigenvectors = eigenvectors[:, idx]
print('came here 4')
# selecting top k eigenvectors (principal components) to use for dimensionality reduction
k = 50
V = eigenvectors[:, :k]
print('came here 5')
#now projecting the image data (centered data) onto principal components to obtain dimensionality-reduced data
X_reduced = np.dot(X_centered, V) # dot product of meanshift images and eigenvectors to reduce the dimension from N to K space
print('came here 6')

# reconstructing the  original data from dimensionality-reduced data
X_reduced = X_reduced.astype(np.float64)
V = V.astype(np.float64) 
X_reconstructed = np.dot(X_reduced, V.T) +mean  # transposing V.T the matrix and dot product for back projection this will give approx image
print('came here 7')
# reshaping reconstructed images back to original shape
X_reconstructed = X_reconstructed.reshape(X.shape)
print('came here 8')

# for showing comparison between constructed and original
import matplotlib.pyplot as plt
fig, axes = plt.subplots(nrows=2, ncols=5 )
print('came here 9')
for i in range(5):
    axes[0, i].imshow(X[i], cmap='gray')
    axes[0, 2].set_title('Original',color='0')
    axes[1, i].imshow(X_reconstructed[i], cmap='gray')
    axes[1, 2].set_title('Reconstructed',color='0')
plt.tight_layout()
plt.show()


# for principla component
print('Number of principal components used:', k)