Search code examples
image-processingkeras

Keras best Image Data Generator parameters for data augmentation

I'm working on facial expression recognition using Keras, the dataset I'm using does not have a big amount of data available, So I'm going to use Keras's image preprocessing for data augmentation.

I want to know the best parameters of ImageDataGenerator to generate normal faces which I can use to train my neural network with.

Here's the code I'm using for Data augmentation:

def data_augmentation(subdir):

    datagen = ImageDataGenerator(
        featurewise_center=False,
        samplewise_center=False,
        featurewise_std_normalization=False,
        samplewise_std_normalization=False,
        zca_whitening=False,
        rotation_range=30,
        width_shift_range=0.2,
        height_shift_range=0.2,
        horizontal_flip=True,
        vertical_flip=False)

    print ("\nData augmentation...")
    print ("\nProcess...")

    for file in glob.glob(subdir+"*/*.jpg"):
        img = load_img(file)
        print ("\nProcessing..." + str(file))
        x = img_to_array(img)
        x = x.reshape((1,) + x.shape)

        i = 0
        for batch in datagen.flow(x, batch_size=1, save_to_dir='data_aug', save_prefix='Fig', save_format='jpg'):
            i += 1
            if i > 20:
                break

Here's all ImageDataGenerator's parameters

keras.preprocessing.image.ImageDataGenerator(featurewise_center=False,
                samplewise_center=False,
                featurewise_std_normalization=False,
                samplewise_std_normalization=False,
                zca_whitening=False,
                zca_epsilon=1e-6,
                rotation_range=0.,
                width_shift_range=0.,
                height_shift_range=0.,
                shear_range=0.,
                zoom_range=0.,
                channel_shift_range=0.,
                fill_mode='nearest',
                cval=0.,
                horizontal_flip=False,
                vertical_flip=False,
                rescale=None,
                preprocessing_function=None,
                data_format=K.image_data_format())

And here's an example of images generated using my code:

enter image description here

enter image description here

enter image description here

enter image description here

As you can see, the images are distorted and not good enough to train my network.

I want to know what's the best parameters of ImageDataGenerator for human faces or is there any better methods for data augmentation?

Solution

  • If you are dealing with faces only, check this paper: http://www.openu.ac.il/home/hassner/projects/augmented_faces/ It can generate 30 images from a single face image. Also you can use virtual makeup implementations in python. They are easy to use and successful.