Research

  • Embedded systems and software
  • Display systems and image processing (focusing AMOLED display)
  • Low-power technology, power modeling (focusing AMOLED display)
  • Human visual system (HVS)-aware image quality assessment (IQA)
  • GPU acceleration
  • Deep learning applications

 

  • Why low-power?

 

     

  • Current research


    • 1. Accurate power model for AMOLED displays (DAC 2017)
    - AMOLED display is one of big power consumers
    - Existing power models do not consider channel dependencies

    2. Low-power and HVS-aware Color Transformation (MTAP 2018)
    - OLED display’s power consumption varies depending on the contents: The brighter, the more power.
    - Darkening is always good? How much dark?
      -> Human visual perceptuality is important!
     


    3. Color Transformation-based Dynamic Voltage Scaling (ELEX, 2015)

    - Color transformation finds a optimal direction for both HVS-awareness and power saving, producing a reference luminance
    - DVS achieves more power saving based on this luminance



    4. DVS using SCD for Video Playback on Mobile AMOLED Displays (ISLPED 2016, JEDS 2017)

    - A more accurate and lower overhead scene change detection (SCD) method than prior work is proposed
    - Employs YCbCr entropy values of the macroblocks in the decoding process rather than using the RGB information


    - A novel DVS combined with the proposed SCD method (ESC-DVS) is designed




    5. LGC-DVS: Local Gamma Correction-Based Dynamic Voltage Scaling (JEDS 2017)

    - The first practical and effective DVS scheme on a commercial smartphone
    - Automated battery-aware DVS with both HVS-aware image quality and high power saving based on a user study
    - Sophisticated implementation of the proposed DVS scheme at the Android HAL level through an in-depth study of the Android platform








    6. HVS-aware Image Quality Assessment

    - Full reference (FR) IQAs still have failed to obtain good HVS-aware image similarity
    - The performance often depends on the distortion type or image type
      -> We still require both a robust and consistent IQA to the HVS
     - CNN or ML may be used for better performance






    7. Deep Learning on Embedded Systems with GPU Acceleration Supports

    - OpenCL or CUDA based acceleration for performance enhancement
    - Memory usage is also a big issue


    8. IQA Acceleration

    - Implementation of IQA metrics and their acceleration on mobile embedded systems
    - Performance vs. FPS (frame per second)

 

  • Past research


    • 1. Low-power and high quality LED backlight based LCD systems
    - Research on low-power and high-fidelity dimming
    - Development of human visual system(HVS)-aware technology
    - Both simulator and board-based research

    low-power LCD dimming research


    2. Low-power and high-performance SATA hard disk

    - Design of an optimal cache algorithm

    - Advanced awareness of NCQ of SATA

    - Implementation and evaluation of a practical SATA disk simulator

    SATA disk simulator architecure