EE133 Final Project Averaging Video Frames for Noise Reduction

1
EE-133 Final ProjectAveraging Video Frames for
Noise Reduction

• James Pringle

2
Background

• Low cost cameras have poor performance in low
  light conditions and the CMOS sensors used
  produce more noise than the CCD sensors in more
  expensive cameras.

3
Description

• Matlab was used to read in frames from an AVI
  captured from a low cost USB camera under low
  light conditions. The frames were averaged to
  produce one image with less noise.

4
Theory

• Averaging can reduce sensor noise.
• Averaging multiple frames avoids the blurring
  that results from applying an averaging filter to
  one image.
• Summing can be used to brighten dark images.

5
Implementation

• Matlab code was developed to average or sum a set
  number of frames from an AVI file.
• Code was also developed to align multiple frames
  for maximum correlation.

6
Results

• One frame form the source AVI.

7
Results

• Same frame processed with a 3x3 averaging filter.

8
Results

• Same frame processed with a 3x3 median filter.

9
Results

• Result of averaging 8 frames.

10
Results

• Result of averaging 80 frames.

11
Results

• Average of 80 frames brightened by 2X.

12
Results

• Average of 80 frames brightened by 4X.

13
Results

• One unprocessed frame.

14
Results

• Average of 8 frames without alignment.

15
Results

• Average of 8 frames after alignment.

16
Conclusions

• Averaging frames can producer an image with less
  noise.
• Summing can produce a brighter image.

17
References

• D. Rutter, Webcam shootout D-Link DSB-C300
  versus Logitech QuickCam Express, Dans Data,
  May 2000. Online. Available http//www.dansdata
  .com/wccomp.htm Accessed April 28, 2006.