Image Data Augmentation in Keras

0
1137
server room 3d illustration with node base programming data design element.concept of big data storage and cloud computing technology.

When dealing with Deep Learning on images, the first question that arises in your mind is where you will get this particular type of image in a huge amount? The answer is you can with a small image dataset and perform some augmentations on this image dataset.

What is Image Augmentation?

Image Augmentation is an image processing technique where we expand the training dataset size by creating new images through some changes in the existing photos. Image sample generated from data augmentation increases the current data by two times or three times, helping you build more generalized models.

In Keras, we have ImageDataGenerator API, which generates the images in batches with real-time data augmentation.

The import command for Image Data Generator is:

from tensorflow.keras.preprocessing.image import ImageDataGenerator

This Image Data Generator has various capabilities to produce new images such as :

  • Applying rotation,
  • shearing,
  • zoom,
  • Changing dimensions,
  • Feature standardization
  • Sample standardization
  • Save the augmented image
  • Moreover, we can define custom augmentation functions for images.

For all the above processing on images, we have the parameters in Data Generator. Let’s discuss some of them.

Rotating images:

We can rotate the image to a random degree through the rotation_angle parameter in the image data generator.

from tensorflow.keras.datasets import mnist
(X_train, y_train), (X_test, y_test) = mnist.load_data() #loading data

X_train = X_train.reshape((X_train.shape[0], 28, 28, 1))
X_test = X_test.reshape((X_test.shape[0], 28, 28, 1))

X_train = X_train.astype('float32')
X_test = X_test.astype('float32')

data_generator = ImageDataGenerator(rotation_range=90) #defining datagenerator

data_generator.fit(X_train)

for X_batch, y_batch in data_generator.flow(X_train, y_train, batch_size=9):
    for i in range(0, 9):
        plt.subplot(330 + 1 + i)
        plt.imshow(X_batch[i].reshape(28, 28), cmap=plt.get_cmap('gray'))
    plt.show()
    break
After rotating images by 90 degree

Shearing and Zooming 

Shearing an image means shifting a part of an image to one direction and the other part in another direction.

datagen = ImageDataGenerator(shear_range = 0.2,zoom_range = 0.3)
# prepare iterator
it = datagen.flow(data, batch_size=1)
# generate samples and plot
for i in range(9):
	# define subplot
	plt.subplot(330 + 1 + i)
	# generate batch of images
	batch = it.next()
	# convert to unsigned integers for viewing
	image = batch[0].astype('uint8')
	# plot raw pixel data
	plt.imshow(image)
# show the figure
plot.show()
original image
Output Images

Like the above parameters, there are various parameters with which you can play. To learn more, visit: https://www.tensorflow.org/api_docs/python/tf/keras/preprocessing/image/ImageDataGenerator/

There is a parameter where we can pass the user-defined image augmentation function. To learn more about custom image augmentation, visit:

https://stepup.ai/custom_data_augmentation_keras

To learn more about image segmentation, click the link below: https://machinelearningmastery.com/image-augmentation-deep-learning-keras.

LEAVE A REPLY

Please enter your comment!
Please enter your name here

This site uses Akismet to reduce spam. Learn how your comment data is processed.