Hey Dziri,
I understand that you want to classify images using one shot learning technique. One of the most popular one shot learning networks are Siamese Networks.
A Siamese network is a unique type of neural network architecture built to learn the similarity between pairs of inputs by processing them through two identical subnetworks that share weights. Each subnetwork processes an input image and produces a feature embedding. These embeddings are then compared to calculate a similarity score, often by computing the distance between them. During training, the network learns to produce embeddings that are close together for similar pairs (e.g., both normal or both abnormal) and far apart for dissimilar pairs (e.g., normal vs. abnormal). This setup is particularly effective for tasks with limited data per class, as it doesn’t require a large dataset of labeled examples but instead relies on pairs of images for learning similarity.
For the medical image classification problem of identifying images as normal or abnormal, a Siamese network can be highly effective. By training the network on pairs of normal and abnormal images, the model learns to distinguish the subtle differences between the two classes.
Once trained, the Siamese network can classify a new image by comparing its embedding to those of known reference images (prototypes) for each class. If the new image’s embedding is closer to the prototype for “normal,” it is classified as normal; otherwise, it’s classified as abnormal. This approach enables effective classification even with a limited number of labeled training images, making it suitable for scenarios in medical imaging where obtaining a large dataset can be challenging.
Please go through the following article to learn more about the network,
Also you can go through this MathWorks documentation to learn more about implementing these kind of networks in MATLAB,
Hope this helps.
