Candlestick Pattern Matching Algorithm

I have an internal library of roughly 550 JPEG snapshots of candlestick formations. I need a small, self-contained tool that can scan this collection and return both exact matches and the closest-looking patterns when I supply a fresh chart image. The comparison has to balance precision with tolerance for slight visual differences, so a hybrid of traditional image-processing (OpenCV, feature descriptors, etc.) and a lightweight machine-learning similarity model is perfectly acceptable—whatever gets reliable results without over-engineering.

The images are all JPEG, and I will be adding new examples over time by dropping additional JPEGs into a folder. A simple “re-index” or “update database” button is enough; automatic feeds are not required. A concise UI (command-line, notebook, or minimal web page) that shows thumbnail previews of the top matches plus a similarity score will make validation much easier.

Deliverables and acceptance criteria
• Script/application source code with clear setup instructions (Python preferred, but any widely-supported language is fine).
• Function that ingests the current image directory, builds/updates the searchable index, and stores it in a reusable format.
• Search routine that accepts a single candlestick JPEG and returns:
– Exact duplicates, if any, ranked first.
– The top N visually similar patterns with their similarity scores.
• README outlining how similarity is computed and how to retrain or fine-tune the model when I manually add new images.
• A short video or screenshots demonstrating the tool matching at least one exact duplicate and one near-match from the existing 550-image set.

Once these points are met, I can quickly test the tool on my side and sign off.