CBIR

Why we included this project

If you've ever wanted to see how content-based image retrieval works under the hood without wading through a research paper, this is a good place to start. It's a compact, readable reference implementation of the classic pipeline: extract a feature from a query image, compare it against a pre-built database of image features, and return the closest matches. The code is organized as modular feature extractors, including color histograms, Gabor filters for texture, HOG and DAISY for shape, and deep features from VGG and ResNet, with a feature-fusion step and random projection for dimensionality reduction. Because each feature is its own module, you can swap approaches and see how each one affects retrieval quality. The included evaluation script computes mean average precision, so you can measure which feature combination actually performs best on your own dataset. It's a solid starting point for students and engineers who want to understand and prototype CBIR before moving to a heavier production stack.