GIGO: Automated Recovery of Price Time Series from Chart Images via Anchor-Point Calibration

Online price-tracking platforms such as CamelCamelCamel render historical price data exclusively as rasterised chart images, without exposing structured, machine-readable time series. For economists studying retail price dynamics, inflation pass-through, or the construction of online price indices, this limitation represents a significant barrier. We present GIGO, an end-to-end pipeline that combines automated web scraping with calibrated computer vision to reverse-engineer the underlying data from these chart images. The core insight is that metadata already published on the tracking page provides exact anchor points for automatic axis calibration, entirely bypassing the need for error-prone OCR of axis labels. A Selenium-driven scraper first collects product metadata, price extremes, and chart screenshots from Amazon Italy via CamelCamelCamel; a computer-vision module then detects colour-coded calibration references via the probabilistic Hough transform, establishes pixel-to-value mappings for both axes, and traces the price curve through column-wise colour scanning. Applied to a panel of consumer-electronics products listed on Amazon Italy, the system recovers over 3000 daily price observations per product with sub-pixel temporal resolution and calibration-bounded price accuracy, offering a scalable alternative to both manual chart digitisation and costly API subscriptions.