Abstract
In this paper, we present a Double-Anchoring Based Tone Mapping (DABTM) algorithm for displaying high dynamic range (HDR) images. First, two anchoring values are obtained using the double-anchoring theory. Second, we use the two values to formulate the compressing operator, which can achieve the aim of tone mapping directly. A new method based on accelerated K-means for the decomposition of HDR images into groups (frameworks) is proposed. Most importantly, a group of piecewise-overlap linear functions is put forward to define the belongingness of pixels to their locating frameworks. Experiments show that our algorithm is capable of achieving dynamic range compression, while preserving fine details and avoiding common artifacts such as gradient reversals, halos, or loss of local contrast.
Introduction
High Dynamic Range (HDR) images capture the vast range of light levels found in the real world, often exceeding the 8-bit per channel capacity of standard consumer displays. Tone mapping is the process of squashing this HDR data into a Low Dynamic Range (LDR) format suitable for monitors and printers. However, simple global compression often results in "washed out" images, while local operators frequently introduce artifacts such as halos (bright glows around dark objects) and gradient reversals. The Double-Anchoring Based Tone Mapping (DABTM) algorithm addresses these issues by mimicking the Human Visual System’s (HVS) perception of lightness.
Methods
The DABTM algorithm is built on three main technical pillars designed to replicate human lightness constancy—the ability to recognize a white surface regardless of illumination.
Accelerated Framework Decomposition
The image is first divided into "frameworks," which are regions with similar illumination (e.g., a shadow area versus a sunlit area). To maintain efficiency, the authors employ an accelerated K-means algorithm utilizing the triangle inequality to group pixels into three categories: Dark, Visible, and Light.
Piecewise-Overlap Linear Functions
To prevent harsh boundaries between frameworks, which are a primary cause of halos, the algorithm uses overlapping membership functions. A pixel can partially belong to two frameworks, ensuring smooth transitions across gradients.
Double-Anchoring and Entropy Weighting
Based on the work of Gilchrist et al., the algorithm uses two "anchors" to estimate lightness: the Highest Luminance Anchor (the brightest part of a framework, excluding outliers) and the Surrounding Luminance Anchor (the weighted average luminance of a pixel's neighborhood). Furthermore, the algorithm calculates the relative entropy of each framework. Frameworks with higher entropy—indicating more information and detail—are given more weight in the final compression, ensuring fine details are prioritized.
Results and Evaluation
The authors evaluated DABTM using standard benchmark HDR images, including courtyard and cathedral scenes, which feature extreme lighting contrasts.
Quantitative evaluation was performed using the HDR-VDP (High Dynamic Range Visual Difference Predictor), which predicts the probability of a human noticing differences between the original and mapped images. DABTM consistently outperformed established algorithms like Drago and Reinhard. For instance, in a courtyard scene, DABTM showed a probability of detection ($P > 95\%$) of only 6.81%, compared to 14.51% for Reinhard and 24.37% for Drago. Qualitatively, DABTM preserves significantly more detail in deep shadows and bright highlights without introducing common artifacts.
Conclusion
DABTM provides a robust solution for HDR tone mapping by integrating HVS principles like double-anchoring and fuzzy framework membership. While the current model uses a fixed framework count ($k=3$) and empirical parameters, it demonstrates superior performance in detail preservation and artifact avoidance. Future work will focus on adaptively determining framework counts and compression parameters based on specific image characteristics.
Summary updated