Image Data Augmentation and AI

Image Data Augmentation and AI

Image Data Augmentation and AI

 

By Venkateswara Reddy Pusapati

 

Why Image Data Augmentation?

State-of-the-art AI models require around 500 images per category during their training phase. If the available training dataset has fewer images, then a data augmentation technique can be used.

This technique expands the small training dataset by adding additional images which are variations of available images in the dataset. Training a model with a large dataset (containing available + transformed images) will improve the performance and ability of the AI model to generalize.

Many variations of an image can be generated by using several transformations like Flipping, Zooming, Rotating, Shifting etc.

Types of Image Data Augmentation

We have different types of image data augmentations. In this blog the following augmentation types are further elaborated.

  • Shift Augmentation
    • Horizontal Shift Augmentation
    • Vertical Shift Augmentation
  • Flip Augmentation
    • Horizontal Flip Augmentation
    • Vertical Flip Augmentation
  • Random Rotation Augmentation
  • Random Brightness Augmentation
  • Random Zoom Augmentation

Shift Augmentation

Shifting the entire pixels of an image from one position to another position is called as shift augmentation.

We have two types of shift augmentation.

  1. Horizontal shift augmentation
  2. Vertical shift augmentation

Horizontal Shift Augmentation

Shifting all pixels of an image in horizontal direction is called Horizontal shift augmentation.

In the above example width shift of –400 to 400 pixels is performed on the input (left side) image and returns the image (right side) as result after shifting the entire pixels of an image in horizontally.

Vertical Shift Augmentation

Shifting all pixels of an image in vertical direction is called vertical shift augmentation.

In the above example height shift of 0.5 pixels is performed on the input (left side) image and returns the image (right side) as result after shifting the entire pixels of an image in vertically.

Flip Augmentation

Flipping means rotating an image in a horizontal or vertical axis.

In horizontal flip, the flipping will be on vertical axis, In Vertical flip the flipping will be on horizontal axis.

Horizontal Flip Augmentation

Reversing the entire rows and columns of an image pixels in horizontally is called horizontal flip augmentation.

In the above example horizontal flip is performed on the input (left side) image and returns the image (right side) as result after flipping the entire pixels of an image.

Vertical Flip Augmentation

Reversing the entire rows and columns of an image pixels in vertically is called Vertical flip augmentation.

In the above example vertical flip is performed on the input (left side) image and returns the image (right side) as result after flipping the entire pixels of an image.

Random Rotation Augmentation

Random rotation augmentation will randomly rotate the images from 0 to 360 degrees in clock wise direction.

In the above example random rotation is performed on the input (left side) image and returns the image (right side) as result after rotation of the entire pixels of an image.

Random Brightness Augmentation

In Random brightness the image brightness can be augmented to bright or dark based on the given brightness range.

The brightness range which has less than 1.0 % darkens the image.

In the above example random brightness is performed on the input (left side) image with the range of 0.2 to 0.6 % and returns the image (right side) as result after brightening the entire pixels of an image.

Random Zoom Augmentation

Zoom Augmentation will randomly zoom the image and adds new pixels for the image.

In the above example random zoom augmentation is performed on the input (left side) image with the range of 0.5 to 1.0 % and returns the image (right side) as result after zooming the entire pixels of an image.

 

 

 

No Comments

Post a Comment

Comment
Name
Email
Website

five × four =

Stay Updated!

Sign up to receive our latest blogs and news. Join our community to receive our content first.