High-Performance Image Stitching

Optimized Image Stitching Algorithm for ITk Pixel Module Inspection

This project focuses on the development and optimization of a high-precision image stitching algorithm designed for the visual inspection of ATLAS ITk pixel modules during large-scale production. The work was carried out as part of the effort to automate wire bond quality assessment for the HL-LHC upgrade.

A single visual inspection of a pixel module consists of multiple high-resolution microscope images acquired in a grid pattern. These images must be combined into a single, geometrically consistent representation while preserving sub-pixel accuracy, as small distortions directly affect downstream defect detection algorithms.

The stitching problem was formulated as an optimization task over a homography plus translation, where transformation parameters are determined by minimizing cost functions defined on overlapping image regions. Several metrics were evaluated, including image-difference and feature-based similarity measures, to ensure robust alignment under varying lighting conditions.

The optimized stitching pipeline produces a standardized, large-scale image with a known mapping between pixel coordinates and physical module geometry. This significantly simplifies both automated analysis and manual cross-checks, while reducing the operational complexity of the image acquisition process.

Links