Single Image Detail Enhancement Based on Local Fractal Dimension Maximization

With the increasing demand for higher image quality, various image detail enhancement techniques have continuously emerged. However, local filter-based methods, while fast, often provide only limited detail enhancement; global filter-based methods yield stronger enhancement but incur large computational costs; deep learning-based methods rely heavily on manually annotated data and lack interpretability; and residual learning-based strategies tend to fall into local optima, making it difficult to fully mine potential global-optimal features. To address these challenges, an image detail enhancement algorithm based on local fractal dimension maximization is proposed. The study finds that the fractal dimension of an image can effectively characterize its texture structure to a certain extent, and its spatial distribution exhibits a certain pattern: Edge regions generally have the highest fractal dimension, textured regions follow, and smooth regions the lowest. Based on this characteristic, a mapping between image texture features and fractal dimension is established, and the intrinsic correlation mechanism between fractal dimension and image detail layers is further investigated. Under the premise of maintaining overall structural consistency, the proposed method achieves effective detail enhancement by increasing local fractal dimensions, thereby providing a theoretically grounded new approach to image enhancement. Extensive experimental results show that the method is competitive in both subjective visual perception and objective evaluation metrics. For example, in ×4 enhancement tests on the BSDS200 dataset, the proposed method improves peak signal to noise ratio and structural similarity by 5.20 dB and 0.1456 over the currently popular QWLS method, thereby demonstrating its advantages and strong generalization capability in image detail enhancement tasks.