127_data_augmentation_using_keras.py

#!/usr/bin/env python
__author__ = "Sreenivas Bhattiprolu"
__license__ = "Feel free to copy, I appreciate if you acknowledge Python for Microscopists"


# https://youtu.be/ccdssX4rIh8
"""
@author: Sreenivas Bhattiprolu

Image shifts via the width_shift_range and height_shift_range arguments.
Image flips via the horizontal_flip and vertical_flip arguments.
Image rotations via the rotation_range argument
Image brightness via the brightness_range argument.
Image zoom via the zoom_range argument.
"""

from keras.preprocessing.image import ImageDataGenerator
from skimage import io

# Construct an instance of the ImageDataGenerator class
# Pass the augmentation parameters through the constructor. 

datagen = ImageDataGenerator(
        rotation_range=45,     #Random rotation between 0 and 45
        width_shift_range=0.2,   #% shift
        height_shift_range=0.2,
        shear_range=0.2,
        zoom_range=0.2,
        horizontal_flip=True,
        fill_mode='constant', cval=125)    #Also try nearest, constant, reflect, wrap


######################################################################
#Loading a single image for demonstration purposes.
#Using flow method to augment the image

# Loading a sample image  
#Can use any library to read images but they need to be in an array form
#If using keras load_img convert it to an array first
x = io.imread('monalisa_einstein/monalisa/monalisa.jpg')  #Array with shape (256, 256, 3)

# Reshape the input image because ...
#x: Input data to datagen.flow must be Numpy array of rank 4 or a tuple.
#First element represents the number of images
x = x.reshape((1, ) + x.shape)  #Array with shape (1, 256, 256, 3)

i = 0
for batch in datagen.flow(x, batch_size=16,  
                          save_to_dir='augmented', 
                          save_prefix='aug', 
                          save_format='png'):
    i += 1
    if i > 20:
        break  # otherwise the generator would loop indefinitely  


####################################################################
#Multiple images.
#Manually read each image and create an array to be supplied to datagen via flow method
dataset = []

import numpy as np
from skimage import io
import os
from PIL import Image

image_directory = 'test_folder/'
SIZE = 128
dataset = []

my_images = os.listdir(image_directory)
for i, image_name in enumerate(my_images):
    if (image_name.split('.')[1] == 'jpg'):
        image = io.imread(image_directory + image_name)
        image = Image.fromarray(image, 'RGB')
        image = image.resize((SIZE,SIZE))
        dataset.append(np.array(image))

x = np.array(dataset)

#Let us save images to get a feel for the augmented images.
#Create an iterator either by using image dataset in memory (using flow() function)
#or by using image dataset from a directory (using flow_from_directory)
#from directory can beuseful if subdirectories are organized by class
   
# Generating and saving 10 augmented samples  
# using the above defined parameters.  
#Again, flow generates batches of randomly augmented images
"""   
i = 0
for batch in datagen.flow(x, batch_size=16,  
                          save_to_dir='augmented', 
                          save_prefix='aug', 
                          save_format='png'):
    i += 1
    if i > 20:
        break  # otherwise the generator would loop indefinitely  
"""


#####################################################################
#Multiclass. Read dirctly from the folder structure using flow_from_directory

i = 0
for batch in datagen.flow_from_directory(directory='monalisa_einstein/', 
                                         batch_size=16,  
                                         target_size=(256, 256),
                                         color_mode="rgb",
                                         save_to_dir='augmented', 
                                         save_prefix='aug', 
                                         save_format='png'):
    i += 1
    if i > 31:
        break 

#Creates 32 images for each class. 
        
#Once data is augmented, you can use it to fit a model via: fit.generator
#instead of fit()
#model = 
#fit model on augmented data
#model.fit_generator(datagen.flow(x))