MATLAB Image Labeler for Automated Defect Detection

Description

This code is written to provide a semi-automated tool that can:

• Sort bridge images into categories of interest to inspectors
• Label bridge images on a pixel wise basis for structural defects

This semi-automated tool makes use of traditional pixel-wise segmentation techniques in addition to manual adjustments for easy labelling.

This code organizes images in a certain file structure for easy integration with the Keras ImageDataGenerator. The images are organized based on the following category tree (bold indicates a final save directory):

• Visible
  • Material
    • Damage
    • No Damage
  • Object
    • Damage
    • No Damage
  • Structure
• Infrared
  • Material
    • Damage
    • No Damage
  • Object
    • Damage
    • No Damage
  • Structure

If the image receives a label of Damage or No Damage, this code will enable pixel-wise labeling for:

• Visual Images with at least one of:
  • No damage (0)
  • Cracking (1)
  • Spalling (2)
  • Corrosion Staining (3)
• Infrared images with at least one of:
  • No damage (0)
  • Delamination (1)

Note that if the image is labeled as No Damage, the code will automatically label all pixels as No Damage.

If the image is labeled as Damage, the user will follow the following process of pixel-wise labeling for each damage type present in the image:

Figure 1 - Process for expediting the pixel-wise labeling of images

Once labeling is complete for an image, the labeled image mask is saved as a m by n matrix where the pixel is assigned a number the represents a corresponding class.

Instructions

To use the program run file image_labeler.m and follows the steps:

• Select images to be labeled when prompted
• Select visual or infrared image
• Select material type
• Select damaged or not damaged
• If damaged: - Select damage type - Use the slide filters to obtain a first guess of the location of the damage - Manually adjust the image mask using add/remove tools until satisfied - Save and continue to next defect or next image
• If not damaged: - All pixels assigned as not damaged - Continue to next defect or next image

Technical Requirements

• Matlab R2017a