Frappe Image Search

Visual similarity search for ERPNext Items — upload a product image and instantly find matching items ranked by similarity score.

---

---

Installation

bench get-app https://github.com/your-org/frappe_image_search
bench install-app frappe_image_search
bench --site <site> migrate

Quick Start

1. Open any Item in ERPNext and set an image in the Image field.
2. Build the search index (run once, or after bulk item updates):

   bench --site <site> execute frappe_image_search.search.rebuild_index

3. Go to /app/image-search in the Frappe Desk.
4. Drag-and-drop or upload a product image → click Search.
5. Matching items appear with a Match % similarity score.

Saving an Item with an image automatically updates its index entry.

How Similarity Works

Each item image is converted to a 1286-dimensional vector combining:

• Spatial colour histograms — where colours appear in the image
• HOG (Histogram of Oriented Gradients) — shape and edge structure
• LBP (Local Binary Patterns) — surface texture (e.g. chain links vs smooth metal)
• Grayscale thumbnail — coarse intensity layout

Search computes cosine similarity between the uploaded image and all indexed items.

License

MIT