Masking Algorithm for CCD Bleeding in Korea Microlensing Telescope Network Images

Abstract

Astronomical CCD images are often affected by bleeding from bright sources, producing column-aligned streaks that degrade source detection and photometry. We present a pixel-level masking algorithm for bleeding in Korea Microlensing Telescope Network (KMTNet) images, in which these streaks extend along detector columns away from the readout register. The algorithm identifies bleed-generating pixels using a conservative bleeding threshold and a bleeding index that quantifies excess signal along the expected bleeding direction. The extent of each bleed trail is then determined by termination criteria based on a background-relative detection threshold and a continuity condition. To optimize these termination parameters, we generate multiple candidate masks and evaluate them with an F-beta-like score that favors high purity while preserving recovery of sources whose photometry is biased by bleeding. Using the final mask, we apply interpolation-based cleaning and assess the performance in terms of completeness and photometric accuracy in a crowded KMTNet field, using Gaia DR3 as a reference. The source completeness improves from 94.08% to 98.61%, primarily by enabling proper aperture placement that was previously hindered by bleeding. For selected bleeding-affected sources, the root-mean-square photometric offset decreases from 0.623 mag to 0.068 mag. These results demonstrate that the algorithm provides a practical framework for mitigating bleeding artifacts in KMTNet images. We further examine the stability of the algorithm across different observing conditions and discuss limitations of the current masking and cleaning procedure.