New Mexico Tech

1
Reducing Processing Time for a LCD Motion Blur
Reduction Algorithm
Eric Martinez IEEE Student Paper Contest In
Conjunction with Shay Har-Noy Professor Truong
Nguyen

• New Mexico Tech

2
Introduction

• Liquid Crystal Display (LCD)

3
Introduction

• Liquid Crystal Display (LCD)
• Popular Consumer Devices
• Low Cost
• Low Power
• Great Picture

4
Introduction

• Liquid Crystal Display (LCD)
• Popular Consumer Devices
• Low Cost
• Low Power
• Great Picture
• Creates Motion Blur

5
Outline

• Background
• Signal Processing Approach
• Convert Matlab to C/C
• Results

6
Background

• LCD

• CRT

7
Background

• Blurry due to Sample and Hold Nature

8
Background

• Phenomenon of Motion Blur
• LCD output
• Human Visual System (HVS)

9
Background

• Phenomenon of Motion Blur
• LCD output
• Human Visual System (HVS)

10
Previous Attempts

• Flashing Backlight

11
Previous Attempts

• Flashing Backlight
• Increase Frame Rate
• Insert Black Frame

12
Previous Attempts

• Flashing Backlight
• Increase Frame Rate
• Insert Black Frame
• Interpolate Frame

13
Signal Processing Approach

• Invert the LCD response and Human Visual System
• Perceive a sharp image

14
Signal Processing Approach

• Invert the LCD response and Human Visual System
• Perceive a sharp image

Without Inversion
With Inversion
15
Implementing Inversion

• LCD and HVS modeled as a Sinc Function
• Many zeros

16
Implementing Inversion

• LCD and HVS modeled as a Sinc Function
• Many zeros
• Filter Bank Approach
• Reduce Computations

17
Implementing Inversion

• LCD and HVS modeled as a Sinc Function
• Many zeros
• Filter Bank Approach
• Reduce Computations
• Compute Inverting Filter Coefficients
• Depends on only Motion Vector Length

18
Implementing Inversion

• LCD and HVS modeled as a Sinc Function
• Many zeros
• Filter Bank Approach
• Reduce Computations
• Compute Inverting Filter Coefficients
• Depends on only Motion Vector Length
• Scaled Gradient Magnitude
• Emphasize Important Areas

19
Computing Inversion

• Process Algorithm using Matlab
• High Level Language
• Toolbox for Computing Inverse Filter Coefficients
• Slow Processing Time

20
Computing Inversion

• Process Algorithm using Matlab
• High Level Language
• Toolbox for Computing Inverse Filter Coefficients
• Slow Processing Time
• One Frame 12 Seconds
• One Minute of Video 12 hours

21
Use C/C

• Use Computer Visual Library (OpenCV)

22
Use C/C

• Use Computer Visual Library (OpenCV)
• Much Faster Computation Time
• Not As Many Functions
• Hard to Debug

23
Results

• One Frame .15 seconds
• One Minute of Video 9 Minutes
• 80x Faster

24
Results

• One Frame .15 seconds
• One Minute of Video 9 Minutes
• 80x Faster
• Used Pointers Instead of Loops
• Reduced Unnecessary Computations
• Read Coefficients From a Table

25
Example Original Frame
26
Simulated Motion Blur
27
Original Frame
28
Inverted Image
29
Perceived Image
30
Output Comparison

• With Inversion

• Without Inversion

31
Future Work

• Calculate Motion Vector Error
• Test Subjects
• Incorporate Deblurring with 120 Hz LCD

32
Acknowledgements

• Shay and Prof. Nguyen
• STARS Program
• UCSD AGEP

33
Reducing Processing Time for a LCD Motion Blur
Reduction Algorithm
Eric Martinez IEEE Student Paper Contest In
Conjunction with Shay Har-Noy Professor Truong
Nguyen

• New Mexico Tech

34
Modeling LCD and HVS
Where
- perceived image
- continuous image
- direction of motion
- display hold time
35
Filter Bank Inversion

36
Computing Filter Coefficients
- is the tradeoff between regularization and
inversion
- corresponds to the distortion function
37
Calculating Error