Automated colony counting on agar plates remains a bottleneck in microbiology workflows, with manual counting prone to fatigue-driven error and existing automated systems often requiring fixed imaging setups. We present a distance-invariant colony counting system built on classical computer vision, designed to operate reliably under variable camera positioning and backlit plate illumination. The pipeline detects the plate boundary via Hough circle transform, flattens illumination gradients, and applies watershed segmentation to separate touching colonies before contour-based morphometric filtering rejects non-colony artifacts. A two-layer anomaly detection system flags abnormal colonies: a statistical layer operating from the first plate, and a machine learning classifier designed to activate as labelled data accumulates. The system is organized around organism-by-plate-type profiles spanning 39 organisms and 11 media types, allowing profile-aware interpretation of colony morphology. Central to the design is a human-in-the-loop validation workflow, where users confirm, flag, or correct automated counts through a web-based interface, generating the ground-truth dataset required for the supervised learning layer. We discuss the architecture, design rationale for prioritizing measurement invariance and validation infrastructure, and the system's readiness for biological validation in ongoing work.
