1. Sejong Univ. CH3. Area Processes Convolutions Blurring Sharpening Averaging vs. Median Filtering

2. Sejong Univ. - 2 -  Area process: Weighted sum of neighboring pixels  Weight is called as mask, filter, kernel, or window  i.e. Area process = convolution with the mask  If mask is separable, faster processing is possible  Ex) per pixel (Multiplication & addition 9 times) -> (3*2 = 6 times) = & Convolutions 121 000 -2 1 0 121

3. Sejong Univ. - 3 -  Ex) Embossing: p71 Fig.3.5, Photoshop: filter>stylize>emboss  Convolving with the following directional mask & +128 Convolutions 00 000 001 001 000 00 100 000 00

4. Sejong Univ. - 4 -  What about processing the boundary pixels?  Zero padding: traditional method, no good  Symmetric expansion is better. Convolutions

5. Sejong Univ. - 5 - Blurring  Remove noises, low-pass filter  Averaging filter: p75 Fig. 3.7 Photoshop: filter>blur>average

6. Sejong Univ. - 6 - Blurring  Gaussian smoothing filter: Improves contrast Photoshop: filter>blur>Gaussian

7. Sejong Univ. - 7 -  Inverse process of blurring, sharpening the image p.77 Fig. 3.9, Photoshop: filter>sharpen Sharpening

8. Sejong Univ. - 8 -  High-pass filter p.78 Fig. 3.19, Photoshop: filter>etc>highpass  High-boost filter (Output) = α(Input) – Low-pass filtered value Ex) When low-pass filter is averaging filter, w = 9α -1 Sharpening

9. Sejong Univ. - 9 - Averaging vs. Median Filtering  Median value in the window Photoshop: filter>noise>median  Compare the performance of median and averaging filter for scratch noise & Gaussian noise, respectively.

10. Sejong Univ. - 10 -  Implement 1) Embossing, 2) Averaging (3*3 mask), 3) Gaussian Smoothing, 4) Median, 5) Sharpening, 6) High-pass filter by using the given masks. 2. Compare Averaging Filter & Median Filter for lena-scratch- noise.raw & lena-gaussian-noise.raw images.

11. Sejong Univ. - 11 -  You need to complete 5 computer programs.  One point per a program is given.  All the programs are very similar to the programs given in experiment 4-6.  In order to get the point, you need to show the result within the given time for each program.  It is highly recommended to bring your own source codes for experiment 4-6.