PaperlayerView exact revision on arXiv

Source-linked AI summary

Chameleon: A Color-Adaptive Web Browser for Mobile OLED Displays

Mian Dong, Lin Zhong

arXiv:1101.1240v1cs.GRcs.HC

TL;DR

OLED web browsing can consume substantial energy because display power varies by color and web pages are often bright or unsuitable for mobile optimization. Chameleon adapts web-page colors under user-supplied constraints using device-specific models and browser-engine integration, reducing measured browsing power without noticeable delay. The paper also reports that two weeks of usage can produce transformations close to optimal for at least ten weeks.

  • Problem

    OLED displays use substantially different power for different colors, while much web content is bright and browser-level fixed color schemes can make multi-colored pages unusable.

  • Method

    Chameleon is a color-adaptive mobile web browser that applies device-specific, constraint-aware color transformations while optimizing which tasks run in real time.

  • Results

    41% lower average system power consumption was measured for OLED smartphone web browsing without user-noticeable delay.

  • Takeaways & Limitations

    Two weeks of web usage was enough to derive a transformation close to optimal for at least ten weeks, allowing collection to run only periodically.

  • Takeaways & Limitations

    Chameleon does not currently consider video, although the paper describes a proposed darkening approach for videos.

Abstract

from arXiv · show

Displays based on organic light-emitting diode (OLED) technology are appearing on many mobile devices. Unlike liquid crystal displays (LCD), OLED displays consume dramatically different power for showing different colors. In particular, OLED displays are inefficient for showing bright colors. This has made them undesirable for mobile devices because much of the web content is of bright colors. To tackle this problem, we present the motivational studies, design, and realization of Chameleon, a color adaptive web browser that renders web pages with power-optimized color schemes under user-supplied constraints. Driven by the findings from our motivational studies, Chameleon provides end users with important options, offloads tasks that are not absolutely needed in real-time, and accomplishes real-time tasks by carefully enhancing the codebase of a browser engine. According to measure-ments with OLED smartphones, Chameleon is able to re-duce average system power consumption for web browsing by 41% and reduce display power consumption by 64% without introducing any noticeable delay.

1. Introduction

Chameleon targets OLED web-browsing energy use by adapting web-page colors while preserving user-specified perceptual constraints. Its design combines motivational studies with browser integration and reports substantial measured power savings without noticeable delay.

  • Motivation: OLED displays consume power according to displayed colors, making bright web content especially costly for mobile browsing.Web browsing is widely used, most web content is white, and the display contributed about 60% of Nexus One’s average CNN mobile browsing power.
  • Motivation: Web-site redesigns are unlikely to solve the problem soon because nearly half of mobile users’ visited pages are not optimized for mobile devices.Browser-level fixed color schemes are also inadequate because pages commonly use multiple text and background colors, so a single style can harm usability.
  • Motivation: Real-time client-side color transformation is challenging because it must process roughly 10^5 pixels and 10^7 colors efficiently.A straightforward implementation could consume enough computation to defeat the intended energy savings.
  • Design: Chameleon performs device-specific, constraint-aware color transformation while applying only necessary tasks in real time.It supports consistent site-level transformations, user color and fidelity preferences, locally constructed OLED power models, and optimized real-time execution.
  • Evaluation: 41% lower average system power consumption was measured during OLED smartphone web browsing without user-noticeable delay.The paper presents this evaluation as evidence that Chameleon’s design and realization address the client-side feasibility challenge.
  • Contributions: The paper contributes motivational studies, the Chameleon design, and efficient realizations based on open-source mobile browser engines.The listed realizations include Android WebKit and Fennec.

2. Background and Related Work

The paper introduces OLED color-power behavior, color transformation under perceptual constraints, and the browser-rendering stages that Chameleon modifies. It motivates device-specific and content-aware transformations rather than uniform display darkening or fixed browser styles.

  • OLED Displays: OLED pixels are emissive, and differing red, green, and blue luminance efficiencies make pixel color directly affect display power.The paper contrasts this with LCD backlights and presents OLED power models as linear functions of linear RGB values.
  • Color-Power Model: Chameleon’s color-power computation counts pixels for each color and applies a pixel power model to estimate region power.The model operates on colors represented in linear RGB or CIELAB space.
  • Related Work: Uniform display darkening changes only lightness, whereas color transformation changes lightness and chromaticity.Under the same perception constraints, the paper reports that color transformation can reduce display power three times more than display darkening.
  • Color Transformation: Color transformation replaces original colors with transformed colors to minimize display power while satisfying fidelity or usability constraints.Fidelity constraints limit total color distortion, whereas usability constraints preserve relationships such as contrast between colors.
  • Color Transformation: Arbitrary transformations can maximize power reduction for independent colors, while linear transformations preserve relative color positions and gradients.The choice therefore depends on whether preserving spatial color relationships or maximizing reduction is more important.
  • Web Browser: A browser loads and parses resources, formats styles, calculates layout, and paints a render tree into a framebuffer bitmap.Chameleon uses image-node position and size information, intercepts paint functions, and modifies image and graphics-library interfaces to transform colors.

3. Motivational Studies

The motivational studies establish that OLED power depends strongly on color, mobile web usage is concentrated and often visually complex, and users need fidelity- and preference-aware transformations.

  • OLED Display Power: OLED display power is a linear function of linear RGB intensity levels, with R2 above 0.95 for all three measured devices.Different devices nevertheless have different red–green power characteristics, motivating device-specific models.
  • OLED Display Power: 5X separates the power consumption of two colors with identical lightness, making chromaticity changes more effective than display darkening alone.Power also increases with lightness at fixed chromaticity.
  • Browsing Behavior: Approximately 50% of mobile-user web pages are not optimized for mobile devices, motivating client-based color transformation.The finding comes from browser traces collected from 25 iPhone 3GS users over three months.
  • Browsing Behavior: 80%-100% of each user’s web usage comes from their 20 most visited sites, averaging 90%.This supports maintaining site-specific transformations for those sites and using a universal scheme or no transformation elsewhere.
  • Web Content: 65% of visited-page pixels are white, while pages use about 1500 colors on average and 6500 at most.The relatively small site-specific color sets make contribution collection and color transformation feasible.
  • Web Content: Images account for about 15% of average-page pixels and require color fidelity, whereas GUI objects cover approximately 85% and can usually be modified if usability and aesthetics remain acceptable.User-defined CSS and indiscriminate inversion can lose visual information or make images unusable.
  • User Preference: 25%, 34%, 66%, and 72% are the reported power reductions for Dark, Green, Inversion, and Arbitrary transformations under the same perceptual constraint.Different users and sites preferred different transformations, supporting user-selectable algorithms.

4. Design of Chameleon

Chameleon’s design applies transformations selectively and contextually: it preserves image fidelity, maintains consistency within heavily used sites, adapts to each device, and exposes user controls while moving expensive work offline.

  • Treat GUI Objects and Images Differently: Chameleon applies display darkening to foreground images but color transformation to GUI objects, background images, and optionally logo images.This preserves fidelity while supporting incremental rendering of large photos.
  • Keep Color Consistency per Web Site: Chameleon applies the same color transformation to all pages from each of a user’s top 20 web sites.It avoids cross-site consistency because users may prefer different transformations for different sites.
  • Generate Device Specific OLED Power Model: Chameleon builds a device-specific OLED power model using power readings from the device’s battery interface.The model accounts for differing power characteristics across OLED displays.
  • Calculate Color Maps Offline: Chameleon computes compute-intensive color-map optimization offline and performs only color contribution collection and painting in real time.This design separates expensive mapping from latency-sensitive browsing operations.
  • Give User Options: Users can select transformation algorithms, color preferences, perceptual constraints, and logo-image handling for each web site.Precomputed maps for all options let users see selection effects immediately.

4.2 Architecture

Chameleon estimates future browsing energy from device power and site color-use data, optimizes color maps offline under user options, and applies them through four browser-integrated modules.

  • Energy Model: The energy expression weights each supported color’s pixel power by its time-integrated pixel count.The integral captures spatial and temporal contribution, including longer viewing and optional fading of older records.
  • Energy Model: OLED pixel power is modeled as a linear function of linear RGB values: P_pixel(x) = a·R + b·G + c·B.The color map X' transforms each original color, while D records each color’s time-integrated contribution.
  • Architecture Overview: Chameleon minimizes estimated energy E by constructing a power model M, gathering contribution data D, calculating a color map X', and applying it to future pages.The architecture contains four modules that interact with the browser engine.
  • Modules: The model construction module uses the smart battery interface, and the contribution collection module gathers per-site color contributions from browser layout and painting stages.Contribution collection is event-driven.
  • Modules: The offline mapping optimization module computes X' for all user options, while the execution module applies X' to web-page colors.Only contribution collection and execution must run in real time.

4.3 Color Power Model Construction

Chameleon constructs an OLED color-power model automatically by displaying uniform-color benchmark images, measuring system power, and fitting the resulting measurements with linear regression.

  • Model Construction: Chameleon constructs its model automatically using the limited power-measurement capability of the smart battery interface.No external assistance is required, allowing immediate deployment.
  • Measurement Procedure: The model construction module displays benchmark images whose pixels share one color and measures whole-system power for each image.The measurements isolate the relationship between displayed RGB values and system power, including black-screen power.
  • Model Fitting: Linear regression obtains the OLED model coefficients a, b, and c after all benchmark-image measurements are collected.The system-power model includes pixel count n and black-screen power P_black.
  • Calibration: The model is recalibrated a few times per year because OLED displays age and their color-power properties change over time.The aging process is described as slow and recalibration does not engage the user.

4.4 Color Contribution Collection

Chameleon collects color-contribution statistics from displayed pixels in real time, using paint events and framebuffer access to balance accuracy and efficiency. A user study guides a one-second timeout, while block processing and sampling reduce overhead.

  • Contribution statistics: Chameleon builds a color contribution vector by tracking how many pixels display each color and for how long.The vector covers GUI objects, background images, and optionally logo images.
  • Event-driven time counting: Real-time collection captures the actual visible screen despite scrolling and zooming, which offline browser-history analysis cannot accurately represent.
  • Event-driven time counting: Paint-function events trigger collection, while a timeout identifies when a sequence of paint calls has ended.
  • Timeout selection: A ten-user, 300-minute Nexus One browsing study evaluates inter-series intervals and series durations across timeout values.Inter-series interval measures time between collection executions; series duration measures updates not counted during a paint sequence.
  • Timeout selection: A one-second timeout is selected from the reasonable one-to-two-second range, trading off frequent collection against unaccounted display updates.
  • Pixel counting: Framebuffer copying and block-based hash tables support pixel counting, while sampling trades contribution-vector accuracy for efficiency.Contribution-collection execution takes at most 66ms without sampling and 17ms with sampling.

4.5 Color Mapping Optimization

Chameleon separates foreground-image handling from color-scheme transformation and computes color maps offline for user-selected perceptual constraints. Precomputing option combinations enables immediate changes to the resulting scheme.

  • Transformation policies: Chameleon darkens foreground images but transforms GUI objects, background images, and optionally logo images under perceptual constraints.
  • Offline optimization: Given the OLED power model and color contribution vector, Chameleon computes color maps offline for all combinations of user options and perceptual constraints.The implementation stores 20 color maps for each website.

4.6 Color Transformation Execution

Chameleon applies precomputed mappings during browser rendering for GUI objects, background images, and selected logo images, while foreground photos receive brightness scaling. These operations modify rendering or decompression parameters rather than pixels individually.

  • GUI and background content: Chameleon replaces each original GUI, background, or selected logo color with its corresponding transformed color during the Painting stage.Paint-function parameters are modified so an entire updated region is transformed at once.
  • Foreground photos: Foreground photo transformation darkens every pixel by multiplying its RGB components by λ, where λ∈[0,1].The scaling is integrated into image decompression immediately after pixel RGB values are calculated.

5. Implementation of Chameleon

Chameleon is realized by modifying browser-engine source code, with implementations for Fennec and Android WebKit. Its design combines in-browser collection and transformation with offline server-side mapping optimization, while contribution collection need not remain continuously active.

  • Browser integration: Browser plug-ins cannot control page rendering, so Chameleon is implemented by modifying browser source code instead.
  • Browser integration: Chameleon is implemented on Fennec and Android WebKit, with Fennec avoiding Android system rebuilding but requiring over 45MB of storage.
  • Browser integration: The Fennec implementation changes 15 files and adds or changes 387 lines of code.
  • Device modeling: Nexus One model-building uses controlled operating conditions, including airplane mode and shutting down other third-party applications.
  • Runtime collection: Contribution collection reads updated screen regions from the framebuffer into main memory using OpenGL GLReadPixels, with copying taking at most 16ms on Nexus One.
  • Runtime collection: Two weeks of color-contribution collection lead to near-optimal power reduction for at least the next three months for an average user.
  • Offline optimization: An Internet service performs mapping optimization from the device power model and 20 contribution vectors representing the user’s most visited websites.
  • Rendering integration: GUI colors are transformed through Fennec’s color interface, while image colors are transformed in format-specific image libraries after decompression.

6. EVALUATION

Chameleon was evaluated with measurements, trace-based emulation, and field trials, showing substantial power savings with low overhead and no noticeable slowdown. Two weeks of training remained close to optimal over at least ten weeks, while the emulation was approximate because traces did not capture displayed content.

  • Evaluation methodology: The power model’s average absolute error was 3%, with 95% of errors within ±10% against DAQ measurements on 100 web pages.This comparison used randomly selected pages from the LiveLab traces.
  • Evaluation methodology: The LiveLab emulation was an approximation because its traces did not capture what appeared on the display.The emulation used the LiveLab traces, the Nexus One display power model, and the arbitrary color transformation.
  • Training effectiveness: 64% average display power reduction was achieved after two to four weeks of training, which performed similarly and better than one week.The two-to-four-week results remained close to the optimal reduction of approximately 70% for at least ten weeks.
  • Training effectiveness: Two weeks of training was sufficient on average for Chameleon to maintain power reduction close to optimal over the remaining traces.The authors suggest retraining for two weeks only a few times per year to maintain effectiveness.
  • Overhead: Chameleon introduced less than 5% system-power difference in a control trial using an unchanged color map, negligible relative to the 41% reduction.The control comparison used the same LiveLab URL list with Fennec and Chameleon.
  • Field trials: No user noticed slowdown in a seven-day field trial comparing original Fennec with Chameleon on Nexus One smartphones.The trial involved five users of Nexus One and Samsung Galaxy S, with the latency assessment performed on two Nexus One phones.

7. Discussions

Chameleon currently excludes video from its considered content, although the discussion describes a CSS-opacity-based method for darkening videos if support is added.

  • Scope: Chameleon does not currently consider video because mobile browsers did not widely support it at the time of the work.The discussion gives iPhone Safari’s lack of Flash support as an example.
  • Possible extension: Videos could be darkened by overlaying a black image with CSS opacity 1 − λ, where λ is the desired scale factor.The overlay would be placed on top of a video at the same size.

8. Conclusions

Chameleon is a color-adaptive mobile browser designed to reduce OLED-system energy use, informed by user and web-usage studies. It achieves large measured savings, requires limited retraining, and can offload mapping optimization to the cloud for immediate option previews.

  • Conclusion: Chameleon reduces OLED smartphone system power consumption by over 41% during web browsing without noticeable user delay.This is the paper’s reported conclusion from its design and realization.
  • Conclusion: User and field web-usage studies directly motivated many of Chameleon’s optimization techniques and contributed to its design and success.The authors identify studying users and their web usage as instrumental.
  • Conclusion: Two weeks of web usage produced a color transformation close to optimal for at least ten weeks in LiveLab-trace emulation.The conclusion states that contribution collection therefore needs to run only for two weeks once or twice a year, while execution runs continuously.
  • Conclusion: Cloud-offloaded mapping optimization lets users view the visual impact of different color options without delay.Chameleon obtains possible color maps after offloading compute-intensive optimization.
Loading 1101.1240v1…