Neon Sign Generator
Computer vision system that converts sketches and logos into photorealistic neon sign visualizations with automated price estimation for a real client.
Overview
Second-year group project at Universiteit van Amsterdam, built for Your Neon — a company that creates custom neon LED signs. The client's problem: customers would submit sketch requests, the client would spend hours manually designing neon signs and calculating prices, only for many customers to back out after seeing the cost. We automated both the visualization and price estimation.
How It Works
Given an input image (logo, text, or sketch) and a desired width, the system:
- Preprocessing: Greyscale conversion and adaptive thresholding to isolate the salient image object from background noise
- Edge/Skeleton Extraction: Two modes depending on the design type:
- Contour mode: Traces the outer edges of the design (for outlined signs)
- Skeleton mode: Extracts the centerline through the design (for filled shapes)
- Length Measurement: Converts the pixel-wide tracing to real-world meters using the customer's specified width
- Price Estimation: Feeds the length into the client's pricing formula (LED strip cost, packaging, shipping, materials)
- Neon Visualization: Renders a photorealistic glowing neon effect over the extracted edges
Price Model
We initially attempted a neural network for length-to-price estimation but lacked sufficient training data. Instead, we fitted separate RANSAC linear regression models for contour and skeleton images, which significantly outperformed least-squares regression by handling outliers in the client's historical pricing data.
Technologies
Python, OpenCV, Canny Edge Detection, RANSAC, Linear Regression, Tkinter GUI