Recently, Implicit Neural Representation (INR) has become a notable approach for signal representation and processing, showing remarkable results in a variety of computer vision applications. To improve the quality of reconstruction, Positional Encoding (PE) was proposed, which transforms the original low-dimensional input into high-dimensional input through trigonometric functions. Nonetheless, current methods of PE in the INR domain treat all images the same, ignoring that different image characteristics need to be matched with different base frequency components. Therefore, this paper introduces Stagewise Positional Encoding as an improvement to existing PE methods in the field of image reconstruction. We introduce segmented PE into INR for the first time to achieve differentiated base frequency parameter assignment for high- frequency and low-frequency components. Additionally, we propose using the fractal dimension measured by the box-counting method as a criterion to determine the most suitable base frequency components for the images, enabling adaptive base frequency parameter adjustment. Experiments’ result demonstrates that our method enhances reconstruction quality without adding the data volume or computational complexity in image reconstruction.