DIH-Cell-Augmentor

A setup for augmentation of biological cells captured via Digital Inline Holography

Aim

• A script to augment the cell-lines obtained via DIHM.
• It generates rotated copies of the input cell sample, while maintaining the same image dimensions and near-similar sample background, using OpenCV.
• The augmented samples preserve the spatial features as well as the statistical distribution present in the input sample.

Method

• Load dataset, having images grouped under labels
• Augment cell sample by rotation at specific angle
• Store augmented images in a csv

Augmentation Procedure

• Create a mask to extract the cell background
• Negate the above mask to extract the cell
• Rotate the image by specified degree, achieved via getRotationMatrix2D and warpAffine functions in cv2
• Extract the rotated cell signature using the negated mask
• Mask it with the extracted cell background to create an augmented image, with same shape as original image

Note: The code is applicable only for augmentation of objects/cells lying within the maximum fitting circle which can be fit into the image

Pre-requisites:

• cv2
• glob, or alternatively use os
• numpy

Demo:

Input Image:

Augmented Images:

Augmented Images on rotation by 0, 10, 20, 30,…..,350 degrees :

Note: Black dot has been added for easier vizualization

Git: DIH Cell Augmentor