Result

Following are some demosaicing results:

1. Point Array (With good optics setting)

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

2. Point Array (With Poor optics setting)

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

In the point array pattern, high quality linear interpolation (HQL) is much better than others. This is because HQL uses the information of all three channels to estimate the missing color pixels, and in such a gray level image RGB channels are highly correlated to each other.

3. Mackay Pattern (With Good optics setting)

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

4. Mackay Pattern (With Good optics setting but with some noise)

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

5. Mackay Pattern (With Poor optics setting)

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

Now we can see some difference between each algorithm. HQL still gives a better result in gray image. Two-line interpolation (TL) now has some notible noisy spot in its result. It's interesting that edge-directed interpolation (ED) also has some noisy spots. It is because in the core of ED's algorithm it requires an inverse matrix which may not exist. In that case matlab will return a poor approximation of the inverse matrix. I believe this is the main reason of these obvious noisy spots.

6. Frequecy Orientation Pattern

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

For color images, click on the image to see a large (512x512) one.

7. Lenna

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

8. Peppers

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

9. Baboon

Origin Bilinear Interpolation Two Line Interpolation Data Dependent Triangulation Edge Directed Interpolation High Quality Linear Interpolation

PSNR

Bilinear Two Line DDT ED HQL
pointArrayGoodOptics 36.3648 35.7747 35.6585 35.4222 41.1929
pointArrayPoorOptics 36.9486 35.5251 36.0582 35.9671 43.3026
Mackay-GoodOptics 35.1965 30.3028 34.5957 30.4705 39.0364
Mackay-GoodOptics-Noisy 34.4876 30.0661 33.9442 28.8066 38.6460
Mackay-PoorOptics 36.6779 31.0248 36.0480 30.8752 39.8976
FREQ-ORIENT 25.6803 20.7551 24.3690 15.0310 23.4313
Lenna 29.0966 26.7559 29.5427 21.3632 23.8906
peppers 28.5986 25.2555 27.1646 22.2679 25.7284
babbon 21.1712 19.6614 20.7621 17.3188 20.8455

< Algorithm > < Conclusion >
< Home >