by V. Bruni, D.De Canditiis and D. Vitulano

 

 

Motivation

 

  Because of masking , any kind of degradation on video sequences is not perceived in the same way in the whole image.

  For instance, noise is:

• visible on flat regions;

• negligible on textures.

  Two examples are shown in the Figures below:

 


Cameraman with additive Gaussian noise. Left: Original Image. Right: noisy version.

 

 

 


20-th frame of FlowerGarden with additive Gaussian noise. Left: Original Image. Right: noisy version.

 

 

 

  On the other hand, motion is not perceived in the same way in the whole sequence. It is:

• visible on edges;

• negligible on flat regions.

 

 

 

The Proposed Strategy: SUVHEP

(Speed Up of Video Enhancement based on Human Perception)

 

  A perception-based block classification can be then used:

  1-st Class: flat blocks - only denoising si required

  2-nd Class: edge blocks - motion estimation and denoising are required

  3-rd Class: textured blocks - no operations are required.

 

 

  A block scheme of the aforementioned classification is shown below:

 



 

 

  Starting from the classical SSIM [1,2]:

 



 

  one can get the two following hypothesis tests:

 

1st Test

 



 

 

2nd Test

 

then the block is textured

 

  Apart from their simplicity, a confidence level can be proven for them as well as a very low complexity:

 

opps

 

  operations per pixel, where N is the total number of pixels in the block B , while k is the number of its sub-blocks.

  Complexity can be further improved via a more effective version of SUVEHP: Fast SUVEHP (see [3] for details), that has the following complexity:

 

opps

 

 

 

 

Experimental Results

 

  Below there are three examples that show that the proposed classification though simple is still effective:

FlowerGarden.	Tennis.	Coastguard.
Classification map. Class 1,2 and 3 respectively are white, gray and black.	Classification map.	Classification map.

      The following results on some video sequences have been achieved by Fast SUVEHP:

Mobile sequence, CoastGuard sequence, FlowerGarden sequence.

  The subjective quality of the restored sequences is unchanged or sometimes better with a considerable computational effort saving.

    But also the objective quality (in terms of SNR and SSIM) is even better with the proposed approach, as proven by the following plots:

 



 

 

 

 

REFERENCES:
1. Z. Wang, A. C. Bovik, H. R. Sheikh, E. P. Simoncelli, Image Quality Assessment: From Error Visibility to Structural Similarity, IEEE Trans. on Image Processing, 13, pp. 600-612, April 2004.

2. Z. Wang, L. Lu, A. C. Bovik, Video Quality Assessment based on Structural Distortion Measurement, Signal Processing: Image Communication, 19(2), pp. 121-132, February, 2004.

3. V. Bruni, D. De Canditiis, D. Vitulano, Speed-Up of Video Enhancement based on Human Perception, submitted to Signal Image and Video Processsing.