Scribd
Upload
12 views6 pages

Scaler: 20225318 Woojae Jung / EE530 Project2

The document discusses 1D and 2D sampling rate conversion using sample-and-hold, linear, and cubic interpolation. It resizes the Barbara test image at ratios of 3/2 and 2/3 using the different interpolation methods. Cubic interpolation has the lowest mean squared error. The document also performs color demosaicing on Kodak test images using cubic and a proposed interpolation filter. While the proposed filter has higher mean squared error, it is expected to perform better than cubic interpolation, especially on the Kodak images.

Uploaded by

이강민
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
12 views6 pages

Scaler: 20225318 Woojae Jung / EE530 Project2

The document discusses 1D and 2D sampling rate conversion using sample-and-hold, linear, and cubic interpolation. It resizes the Barbara test image at ratios of 3/2 and 2/3 using the different interpolation methods. Cubic interpolation has the lowest mean squared error. The document also performs color demosaicing on Kodak test images using cubic and a proposed interpolation filter. While the proposed filter has higher mean squared error, it is expected to perform better than cubic interpolation, especially on the Kodak images.

Uploaded by

이강민
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 6

20225318 Woojae Jung / EE530 Project2

1. Scaler

1) 1D sampling rate conversion for x2 conversion

1 2

0.5 1

0 0
1 2 3 4 5 6 7 8 9 0 100 200 300 400 500 600

1 2

0.5 1

0 0
1 2 3 4 5 6 7 8 9 0 100 200 300 400 500 600

1 2

0.5 1

0
0
1 2 3 4 5 6 7 8 9 0 100 200 300 400 500 600

2) 1D sampling rate conversion for x3conversion

1 3

2
0.5
1

0 0
0 5 10 15 0 100 200 300 400 500 600

1 3

2
0.5
1

0 0
0 5 10 15 0 100 200 300 400 500 600

1 3

2
0.5

0
0
0 5 10 15 0 100 200 300 400 500 600
3) Download, read, typecast to double, and display barbara.png.

i) Scale the image by 3/2 using the sample-and-hold, linear and cubic interpolation.

Sample and Hold x3/2

All figures had been resized (smaller) in the pdf.


Linear x3/2

Cubic x3/2
ii) Scale the result by 2/3 using the sample-and-hold, linear and cubic interpolation.

Sample and Hold x2/3 Linear x2/3

Cubic x2/3

4) Compare the scaling performance of the sample-and-hold, linear and cubic interpolations.

As the Mean Squared Error is smallest at the Cubic interpolations, we can conclude that in the
monochrome image the cubic interpolation performed well compared to sample & hold and linear
interpolation.
2. Demosaicing

1) Download kodimg07r.png, kodimg07g.png, kodimg07b.png and kodim07.png and display them.

From the Left, Top: (R G) Bottom: (B Original)

2) Find the missing RGB pixel data by applying a cubic interpolation filter.
Interpolated Image
3) Find the missing RGB pixel data by applying a given interpolation filter.

Demosaiced Image

I divided the G pixels filter coefficients with 8 as the summation of coefficients are 8.

4) Compare the Demosaicing performance.

As the Mean Squared Error is smallest at the Cubic interpolations, I can see that proposed filter
model is not appropriate to use. But we all know that this proposed filter will show the better result
than cubic especially in the ‘kodim’ image. I think the MSE is higher in the proposed filter because I
except the first two, last two low and column because of indexing error. Finally, then I can conclude
with this filter showed better result than cubic interpolation filter because I showed similar result with
filtering a smaller number of pixels in original image.

You might also like