Codebox Software
Image Augmentation for Machine Learning in Python
Published:
This is a simple data augmentation tool for image files, intended for use with machine learning data sets. The tool scans a directory containing image files, and generates new images by performing a specified set of augmentation operations on each file that it finds. This process multiplies the number of training examples that can be used when developing a neural network, and should significantly improve the resulting network's performance, particularly when the number of training examples is relatively small.
Run the utility from the command-line as follows:
python main.py <image dir> <transform1> <transform2> ...
The <image dir> argument should be the path to a directory containing the image files to be augmented.
The utility will search the directory recursively for files with any of the following extensions:
jpg, jpeg, bmp, png.
The <transform> arguments determine what types of augmentation operations will be performed,
using the codes listed in the table below:
| Code | Description | Example Values |
|---|---|---|
fliph |
Horizontal Flip | fliph |
flipv |
Vertical Flip | flipv |
noise |
Adds random noise to the image | noise_0.01noise_0.5 |
rot |
Rotates the image by the specified amount | rot_90rot_-45 |
trans |
Shifts the pixels of the image by the specified amounts in the x and y directions | trans_20_10trans_-10_0 |
zoom |
Zooms into the specified region of the image, performing stretching/shrinking as necessary | zoom_0_0_20_20zoom_-10_-20_10_10 |
blur |
Blurs the image by the specified amount | blur_1.5 |
Each transform argument results in one additional output image being generated for each input image. An argument may consist of one or more augmentation operations. Multiple operations within a single argument must be separated by commas, and the order in which the operations are performed will match the order in which they are specified within the argument.
Examples
Produce 2 output images for each input image, one of which is flipped horizontally, and one of which is flipped vertically
python main.py ./my_images fliph flipv
Produce 1 output image for each input image, by first rotating the image by 90° and then flipping it horizontally
python main.py ./my_images rot_90,fliph
Operations
Horizontal Flip
Mirrors the image around a vertical line running through its center
python main.py ./my_images fliph
→
Vertical Flip
Mirrors the image around a horizontal line running through its center
python main.py ./my_images flipv
→
Noise
Adds random noise to the image. The amount of noise to be added is specified by a floating-point numeric value that is included in the transform argument, the numeric value must be greater than 0.
python main.py ./my_images noise_0.01 noise_0.02 noise_0.05
→
Rotate
Rotates the image. The angle of rotation, in degrees, is specified by a integer value that is included in the transform argument
python main.py ./my_images rot_90 rot_180 rot_-90
→
Translate
Performs a translation on the image. The size of the translation in the x and y directions are specified by integer values that are included in the transform argument as follows:
trans_<x distance>_<y distance>
Positive distance values correspond to rightward and downward movement, negative values indicate the opposite.
python main.py ./my_images trans_20_20 trans_0_100
→
Zoom/Stretch
Zooms in to (or out from) a particular area of the image. The top-left and bottom-right coordinates of the target region are specified by integer values included in the transform argument as follows:
zoom_<top-left x>_<top-left y>_<bottom-right x>_<bottom-right y>
By specifying a target region with an aspect ratio that differs from that of the source image, stretching transformations can be performed.
python main.py ./my_images zoom_150_0_300_150 zoom_0_50_300_150 zoom_200_0_300_300
→
Blur
Blurs the image. The amount of blurring is specified by a floating-point value included in the transform argument.
python main.py ./my_images blur_1.0 blur_2.0 blur_4.0
→
