Scene-Segmentation-Based Exposure Compensation for Tone Mapping of High Dynamic Range Scenes

Yuma Kinoshita,Hitoshi Kiya
2024-10-21
Abstract:We propose a novel scene-segmentation-based exposure compensation method for multi-exposure image fusion (MEF) based tone mapping. The aim of MEF-based tone mapping is to display high dynamic range (HDR) images on devices with limited dynamic range. To achieve this, this method generates a stack of differently exposed images from an input HDR image and fuses them into a single image. Our approach addresses the limitations of MEF-based tone mapping with existing segmentation-based exposure compensation, which often result in visually unappealing outcomes due to inappropriate exposure value selection. The proposed exposure compensation method first segments the input HDR image into subregions based on luminance values of pixels. It then determines exposure values for multi-exposure images to maximize contrast between regions while preserving relative luminance relationships. This approach contrasts with conventional methods that may invert luminance relationships or compromise contrast between regions. Additionally, we present an improved technique for calculating fusion weights to better reflect the effects of exposure compensation in the final fused image. In a simulation experiment to evaluate the quality of tone-mapped images, the MEF-based tone mapping with the proposed method outperforms three typical tone mapping methods including conventional MEF-based one, in terms of the tone mapped image quality index (TMQI).
Computer Vision and Pattern Recognition,Image and Video Processing
What problem does this paper attempt to address?
### Problems the Paper Attempts to Solve This paper aims to address the quality issues of high dynamic range (HDR) images when displayed on low dynamic range (LDR) devices. Specifically, the paper proposes a scene segmentation-based exposure compensation method for tone mapping in multi-exposure image fusion (MEF). ### Background and Motivation 1. **Mismatch between HDR Images and LDR Devices**: - High dynamic range (HDR) images can capture a wide range of brightness in the real world, but most display devices (such as smartphones and regular monitors) can only display a limited dynamic range (LDR). - To display HDR images on these devices, tone mapping is required, which adjusts the brightness range of HDR images to fit within the range of LDR devices. 2. **Limitations of Existing Methods**: - Existing multi-exposure image fusion (MEF) methods can generate high-quality LDR images, but they face difficulties in selecting appropriate exposure values. - Traditional scene segmentation-based exposure compensation methods avoid overexposure or underexposure issues but often lead to brightness relationship inversion and loss of contrast between regions, affecting the final image's visual quality. ### Proposed Method 1. **Scene Segmentation-Based Exposure Compensation**: - The input HDR image is segmented into multiple sub-regions based on pixel brightness values. - Exposure values are determined for each sub-region to maximize inter-region contrast while maintaining relative brightness relationships. 2. **Improved Fusion Weight Calculation**: - A new fusion weight calculation method is proposed to better reflect the effects of exposure compensation, ensuring the quality of the final fused image. ### Experimental Results 1. **Performance Evaluation**: - Through simulation experiments, the tone-mapped image quality index (TMQI) was used to evaluate the LDR images generated by different methods. - Experimental results show that the proposed scene segmentation-based exposure compensation method outperforms three typical tone mapping methods, including traditional MEF methods, in TMQI scores. 2. **Visual Effects**: - Visual comparison results indicate that the proposed method effectively preserves the contrast of HDR images, generating high-quality LDR images. ### Conclusion This paper proposes a scene segmentation-based exposure compensation method for tone mapping in multi-exposure image fusion. By optimizing exposure values and improving fusion weight calculation, the method effectively addresses the issues of brightness relationship inversion and contrast loss present in existing methods, significantly enhancing the visual quality of LDR images.