Classification with two input images using transfer learning

I have a 3-class classification problem. However, the classification is based on two images rather than the typical one image. How can I use/modify transfer learning models, or otherwise build a model from scratch, that accepts two images as input concurrently.

[Image Analyst]

Not sure what you mean. Attach some images to explain. Maybe you can just stitch the images together to form one single image and train with those. Or else you can use SegNet or U-net to do a pixel-by-pixel classification of things in the images.

[Mohammad Fraiwan]

Thank you for your comment and suggestions. I am not sure specific images will help in understanding my question, so i will elaborate further. To determine the specie of certain parasites, vets need images using the microscope of many subject’s body parts, and it is not possible to take a snapshot of all the parts together from the microscope. Once they view all the parts of interest then they are able to correctly classify the parasite.

[Mohammad Fraiwan]

In my case study, we determine using the image of the mouth, and the image of the genitals.

[Image Analyst]

See these links on panoramic stitching to build a single image from all your subimages:

Feature Based Panoramic Image Stitching - MATLAB & Simulink

Panorama and Image Registration - MATLAB & Simulink

Panoramic photograph stitching

https://www.crisluengo.net/archives/76/

[Mohammad Fraiwan]

Thank you very much. It would be interesting to compare stitching to feeding the two images to a custom CNN.

Answer 1

[yanqi liu]

yes，sir，may be use image fuse or image mosaic to make two image into one，and then use cnn as normal

[Mohammad Fraiwan]

Stitching will not produce good performance due to the small resolution of the CNN input. My question needs a deep learning design answer