Intra-device color constancy calibration for multi-focal-length imaging

Multi-focal-length imaging systems are widely used in modern photography and display-related applications to capture both wide-field context and fine-detail information. However, variations in optical configuration across focal lengths within a single device often lead to noticeable color inconstancy, manifested as shifts in hue, saturation, and brightness between images captured at different focal lengths. Such intra-device color variations degrade visual coherence and can adversely affect subsequent image processing tasks. Unlike conventional color variations caused by illumination changes or inter-device differences, multi-focal-length imaging introduces coupled focal-length-dependent geometric and radiometric variations, including changes in field of view, magnification, spatial misalignment, and heterogeneous color distortion. To systematically investigate this problem, we construct the Multi-Focal-Length Color Distortion (MFCD) dataset, which isolates focal-length-dependent color differences under controlled acquisition conditions. Based on this dataset, we propose Flow-guided Color Constancy Calibration (FCCC), a correspondence-constrained, spatially adaptive calibration framework for intra-device color constancy. FCCC jointly models geometric alignment and color correction by first identifying reliable overlapping regions in the feature domain, and then performing flow-guided dense correspondence propagation with reliability-aware suppression for spatially adaptive color calibration. Extensive experiments demonstrate that FCCC significantly improves perceptual color accuracy and structural fidelity compared with existing methods. Furthermore, when applied as a preprocessing module, FCCC enhances the performance of downstream vision tasks under multi-focal-length imaging, confirming its practical value for real-world imaging systems. To facilitate independent verification of the proposed FCCC method, we provide a packaged executable demo, alongside a set of representative testing samples available at the following public GitHub repository: https://github.com/zd1059391830/-Multi-Focal-Length-Color-Distortion-dataset.