Palette-based image recoloring methods have attracted increasing attention in recent years. By editing a palette that reflects the color distribution of an image, it is possible to recolor the image with specified colors. However, existing methods cannot establish different color mappings for various regions of the image, limiting the richness of the colors and the overall quality of the recolored image. To address this issue, we propose a novel palette-based recoloring method in this paper. Unlike previous methods, we first use a color clustering approach based on color families to extract the palette of the source image. Second, we design and train a deep learning network (DS-Net) that predicts the local color complexity around each pixel. Finally, we propose a color transfer function that converts colors into different ones based on varying local color complexities, ensuring that the transferred colors achieve better visual quality at their respective pixel locations. Our experimental results demonstrate the effectiveness of our method.