An exponential intensity mapping method for grayscale drift suppression in high-temperature forging images

High-temperature forging images often suffer from non-stationary grayscale drift caused by the formation and spallation of oxide scale. This drift degrades the robustness of fixed-threshold Canny edge detection and similarity-based region growing segmentation. To address this issue, an exponential intensity mapping preprocessing method is proposed that suppresses grayscale drift by selectively enhancing the high-intensity forging region while attenuating the low-intensity background, thereby increasing foreground-background separability. The mapping parameters are calibrated on representative production images using a two-point anchoring strategy: stable foreground and background intensity anchors are estimated from quantiles, and the gain and offset are solved analytically under a fixed enhancement strength. Experiments on 79 manually annotated images show that the median region-growing coverage improves from 4.26×10−4 to 0.9438, with the 90th percentile reaching 0.9912. For contour extraction, the band-limited Boundary F1 of Canny increases from 0.7117 to 0.7740. These results demonstrate that the proposed preprocessing effectively compensates for oxide-scale-induced grayscale drift and improves the stability of contour extraction and segmentation without modifying downstream algorithms or adding hardware.