Abstract: To enhance the quality of image harmonization, we propose a region-aware image harmonization method based on adaptive feature calibration. Our method adopts a multi-level encoder-decoder structure and designs an adaptive feature calibration module at the skip connection. This module leverages partial convolution to extract foreground and background features separately, and calibrates the foreground features with adaptive instance normalization, reducing the impact of locally biased features on the decoder reconstruction process. In addition, we utilize the region-aware Transformer improved by the Swin Transformer, which not only enhances the interaction of contextual information, but also modulates foreground features at different scales according to the statistical information of background features. Our method is evaluated on the publicly available iHarmony4 dataset, and outperforms current state-of-the-art methods significantly in terms of PSNR, MSE, and fMSE metrics.