Fast Impulsive Noise Removal Source IEEE Transactions on

Fast Impulsive Noise Removal Source: IEEE Transactions on Image Processing, Vol. 10, No. 1, JAN. 2001, pp. 173 -179 Author: Piotr S. Windyga Speaker: Chi-Lung Chiang Date : 2001/03/08 多媒體系統及網路安全實驗室 1

Motivation: Standard median filter needs sorting data. =>long processing time. Other faster method? =>Peak-and-Valley filter. 多媒體系統及網路安全實驗室 2

Impulsive noise: with probability p. with probability 1 -p. N(i, j): Noise, which uniformly distributed in [0, 255]. O(i, j): Original image pixel. 多媒體系統及網路安全實驗室 3

Proposed method (1 -D Example): Peak cutting: P’(i)=max(P(i-1), P(i+1)) if P(i)>P(i-1) and P(i)>P(i+1) 多媒體系統及網路安全實驗室 4

cont. Valley filling: P’(i)=min(P(i-1), P(i+1)) if P(i)<P(i-1) and P(i)<P(i+1) 多媒體系統及網路安全實驗室 5

Algorithm (cutting-otherwise-filling) P’(i)=max(P(i-1), P(i+1)) if P(i)>P(i-1) and P(i)>P(i+1) P’(i)=min(P(i-1), P(i+1)) if P(i)<P(i-1) and P(i)<P(i+1) P’(i)= P(i) else 多媒體系統及網路安全實驗室 6

2 -D Example Horizontal direction, cutting-otherwise-filling 多媒體系統及網路安全實驗室 7

cont. Vertical direction, cutting-otherwise-filling 多媒體系統及網路安全實驗室 8

Test results Hamburg Taxi 15% corrupted 多媒體系統及網路安全實驗室 9

cont. Median Peak-and-Valley 多媒體系統及網路安全實驗室 10

Checkerboard ANGIOGRAPHIC 多媒體系統及網路安全實驗室 11

Checkerboard Filter %noise eliminated %noise attenuated %image spoiled Average attenuation Average spoiling Time (sec. ) PSNR (d. B) Proposed 94. 38 5. 01 0. 42 72. 62 8. 23 4 39. 29 Median 95. 01 4. 58 1. 67 77. 55 10. 86 9 38. 47 Hamburg Taxi Filter %noise eliminated %noise attenuated %image spoiled Average attenuation Average spoiling Time (sec. ) PSNR (d. B) Proposed 12. 94 83. 82 46. 74 77. 22 3. 76 3 31. 93 Median 15. 37 82. 97 63. 31 77. 58 4. 31 8 30. 51 Angiographic Filter %noise eliminated %noise attenuated %image spoiled Average attenuation Average spoiling Time (sec. ) PSNR (d. B) Proposed 94. 38 5. 01 0. 42 72. 62 8. 23 4 39. 29 Median 95. 01 4. 58 1. 67 77. 55 10. 86 9 38. 47 多媒體系統及網路安全實驗室 12