Scope
This paper focuses on energy-aware image computation, addressing the pressing need to reduce energy consumption associated with digital displays, especially considering the significant environmental impact of digital technologies. The authors propose a taxonomy of methods to modify image content to achieve this goal, considering factors like fidelity to the original content and energy consumption reduction. They elucidate the correlation between image content and energy consumption in emissive displays, emphasizing the role of brightness in energy usage. Different methodologies are explored, including histogram-based, scaling-based, and deep learning-based methods. Each category is dissected, highlighting various approaches and their implications on image fidelity and power reduction. An objective evaluation of 11 methods uses two datasets, focusing on metrics like SSIM and LPIPS for fidelity assessment and actual power measurements for energy reduction. The results showcase the efficacy of certain methods, such as PVR in scaling-based approaches and DeepPVR in deep learning-based approaches, in achieving significant power reduction while maintaining high fidelity to the original content. Furthermore, the study delves into the adaptability of methods across different images and proposes colour correction techniques to enhance fidelity further. The conclusions drawn underscore the potential for substantial energy savings without compromising image quality, although the need for further subjective evaluation and consideration of artistic intent is acknowledged.
Summary
This paper explores, evaluates, and compares methods for computing energy-aware images. It specifically focuses on techniques for reducing the energy consumption of digital displays while maintaining acceptable fidelity to the original content.
The comparison includes two specific technologies:
Emissive Displays (OLED): These displays emit light directly from individual pixels, offering advantages such as improved viewing angles and colour contrast. The paper highlights OLED displays due to their ability to independently control pixel intensities, which impacts energy consumption based on the displayed content.
Non-Emissive Displays (LCD): LCD displays use an external light source to illuminate the pixels. Unlike OLED displays, they do not have independent control of pixel intensities. The paper contrasts OLEDs with LCDs to illustrate how different display technologies contribute to energy consumption, especially concerning brightness levels and content types.
Key aspects:
Taxonomy of Methods: The paper proposes a taxonomy categorising various methodologies for energy-aware image computation, including histogram-based, scaling-based, and deep learning-based approaches.
Objective Evaluation: An objective analysis of 11 methods is conducted, utilizing metrics such as SSIM, LPIPS, and actual power measurements to assess fidelity and energy reduction capabilities across different datasets.
Structural Similarity Index Measure (SSIM): SSIM is a metric used to measure the similarity between two images. It evaluates the original and modified images’ structural information, luminance, and contrast similarity. Higher SSIM values indicate greater similarity between images.
Learned Perceptual Image Similarity (LPIPS): LPIPS is a perceptual metric that quantifies the difference between two images based on features learned by a deep neural network. It assesses the perceptual similarity between images, considering factors such as color, texture, and structure.
Power Reduction Rate (R): This metric quantifies the reduction in energy consumption achieved by energy-aware image computation methods. It is calculated as the ratio of the modified image’s power consumption to the original image’s power consumption, expressed as a percentage.
The paper compares the performance of different methods within each category, highlighting strengths and weaknesses based on objective evaluation results. It explores the adaptability of methods across diverse images and proposes techniques such as a colour correction to enhance fidelity further. The paper concludes with insights into the potential for significant energy savings without compromising image quality while acknowledging the need for further subjective evaluation and consideration of artistic intent in future research endeavours.
Relevance for EXIGENCE
The relevance of his paper is not straight forward, because focus is given to display energy consumption, but an adaptation of the concept of power reduction rate to the projects scenarios could be beneficial to show case the energy efficiency of the proposed solutions maybe for requirements and scenarios and energy measurements.