Abstract: Objectives In this paper, to simultaneously address feature detection failure and matching performance decline problems caused by significant illumination variation in harbor scenarios, a feature matching scheme, termed illumination-invariant feature transform (IIFT), is innovatively proposed. Methods Firstly, a Gaussian pyramid combined with an image block strategy based on information entropy is employed to select candidate regions rich in texture information. Secondly, leveraging the insensitivity of Phase Congruency (PC) theory to illumination changes, keypoints with high robustness are detected. Finally, a PC-FREAK (Phase Congruency - Fast Retina Keypoint) descriptor is designed. By replacing the grayscale values in the neighborhood of keypoints with phase congruency values, the robustness and discriminability of the descriptor under varying illumination are significantly enhanced. Results The experimental results demonstrate that the proposed IIFT algorithm exhibits strong robustness in extreme port scenarios. Compared to traditional gradient-based algorithms and SuperPoint, the average repeatability rate achieves maximum improvements of 17.21% and 15.08%, respectively. The average F1-Score shows an improvement ranging from 7.00% to 18.61%, while the average RMSE is maintained within 2.2 pixels. Conclusions This provides a reliable solution for feature extraction tasks under severe illumination changes in such environments.