Detail Preserving Filter

A Seminar On

“A Detail Preserving Filter for Impulse Noise Detection & Removal” By

Vikas K. Bhangdiya (2008MEC004)

Supervisor Dr. S. V. Bonde

Outline 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11.

Introduction Impulse Noise Model Need of DPF Detail Preserving Filter (DPF) Impulse Noise Detection Edge detection Filtering Schemes Blur Metric Results Conclusions References

2

Introduction • Noise • Linear Filter: - A neighborhood averaging mechanism to remove impulse noise and tend to destroy all high frequency details like edges, lines and other fine image details.

• Non-linear Filter: - It also operates on neighborhoods, however it operations based directly on the values of the neighborhood under consideration, and they don't use explicitly use coefficients.

• Blur Metric: - It is based on the analysis of the spread of edges in an image.

3

Impulse Noise Model  N ij with probabilty p  X ij     I ij with probabilty 1  p  Where i = 1,2,……..s1

j = 1,2,……..s2

I ij : -Original Image N ij : -Noisy Image X ij : -Observation Image 4

Need of DPF 

Classical Filter: - All input samples are unconditionally affected by the filtering process.



Selective Filter: - First checked pixel is corrupted by an impulse. If so, replace it by a value estimated from its neighbors in the window; otherwise pass it to the output unprocessed.

5

Selective Filter Noisy Image

Is pixel Noisy Yes Detail Preserving filter (DPF)

6

Detail Preserving Filter

Impulse Noise detection As a first step in DPF, Adaptive Median Filter based impulse detector is applied for finding the position of impulses.

 X ij | p  q  i  p  m 

W pq ( X )  



 X ij | q  m  j  q  m 

Where p, q : - index of the current pixel. It could be observed that the corrupted pixels belong to the set {Wmin,Wmax}, where Wmin is the minimal pixel value in the defined window and Wmax is the maximum pixel value.

A pixel may be corrupted and assigned to a flag matrix ‘N’ as

 1 if  ( X ij  M ij ) & X ij   Wmin , Wmax  

N (i, j )  

 0

else

Where X : - Corrupted Images M: - Filtered images This impulse detection scheme detects impulse noise even at higher corruption levels setting the flag matrix N(I,j) values as 1 wherever noise exists.

9

 Adaptive Median Filter is a good method for removing random-valued impulse noise.  We take large thresholds so it will only select pixels that are most likely to be noisy, then we restore them.  Subsequent iterations, we decrease the thresholds to include more noise pixel. Since the edges and the details are preserved by the regularization successfully in each iteration, the restored image will not be distorted by this method.

10

Edge detection •

To isolate the noisy pixels present on the edge. Extracting edges from the corrupted image is a difficult task without having the prior knowledge of the edge information.

•

So median filtering is applied on the corrupted image f’(x,y), Canny edge detector is applied on the median filtered output m(i,j).

•

Canny detects true edges at higher level corruption also.

11

e(i, j )   (G I ) Where  G : - Gaussian function of standard deviation I(i,j): - obtained from the median filtered output m(i,j). e(i,j): - Edge matrix will have the value 1 if there is an edge pixel and value 0 for pixels not on edge. •

Initially edge is detected from the median filtered output, and then edge is detected from the iterative filtered result.

12

Categorization of Noisy Pixel  1 if e(i, j )  1 & N (i, j )  1 N e (i, j )    0 Otherwise  

 1 if e(i, j )  0 & N (i, j )  1 N e ' (i, j )    0 Otherwise  

Where N (i, j ) : -Noise Matrix e(i , j ) : -Edge Matrix N e (i, j ) : -Noise on edge N e ' (i, j ) : -Noise not on edge

13

Filtering Schemes • Noise indicated on edge pixels by Ne(i,j). • Noise indicated not on edge indicated by Ne’(i,j).  Each noisy edge pixel is replaced by taking the median of closest non noisy edge pixel present along the direction of the edge.  The direction of edge is found by using the connectivity of the edge pixels.

• where Z is a noisy edge pixel. Shaded regions indicate the direction of edge, Here noisy edge pixels indicated by the 1’s and 0’s represent either noise free or non-edge pixels.

15

• The proposed method tends to replace ‘Z’ by median of nearest non noisy pixels contained in the vector ‘Y’ along the direction of the edge.

S1i , j   med Y

i . j  s ,t

 Nei , j  1

S 2i , j   med ( X )i. j  s , t  Ne 'i , j  1

S (i, j )  S1(i, j ) U S 2(i, j ) S1i , j : -Filtered output of the noise other than on edge. S 2i , j : -Filtered output of the noisy edge pixels.

S (i, j ) : -Final Filter Image

V : -Vector containing non noisy pixels present in its neighborhood.

16

Blur Metric • An image appears blurred when its high spatial frequency values in the spectrum are attenuated. • Motion blur, Out of focus blur, etc. • A no-reference blur measurement technique. We assume no knowledge of the original image, and do not make any assumptions on the type of content or the blurring process.

17

18

Fig: - One row of the blurred image. The detected edges are indicated by the dashed lines, and local minima and maxima around the edge by dotted lines. The edge width at P1 is P2 − P2. 19

Results

Blur Metric= 2.5166, Blur Metric= 2.4269 SSIM=0.8691 SSIM=0.8923

Figure: - Comparison of filtered output 20

Figure: - Comparison of Iterative Results

21

Conclusion The proposed Filtering technique applies Iterative, selective and directional filtering on the corrupted image to reduce the blur. The results shows that this method removes impulse noise, also simultaneously preserves edges at higher levels of noise as is evident from comparison with existing filters.

22

References [1] S. Md. Mansoor Roomi , T. Pandy Maheswari , V. Abhai Kumar “A Detail Preserving Filter for Impulse Noise Detection and Removal” ICGST-GVIP Journal, Volume 7, Issue 3, November 2007. [2] Rafael Gonzalez Richard Woods , Digital Image Processing, Pearson Publications.  [3] Raymont H. Chan, “An Iterative procedure for removing random- valued impulse noise," IEEESignal Process. Lett., vol. no. 11, pp. Dec 2004. [4] P. Marziliano, F. Dufaux, S. Winkler, T. Ebrahimi,“A noreference perceptual Blur metric”, in: Proceedings of the International Conference on ImageProcessing, Vol. 3, Rochester, NY, 2002, pp. 57–60. [5] Kh. Manglem Singh and Prabin K. Bora, “Features Preserving Filters Using Fuzzy Kohonen Clustering Network in Detection of Impulse 23 Noise“  

Thank You …!!! 24

25

Adaptive Median Filter

26