Analysis of CCD used in ES-1000

ICX054AK?

As I searched CCD information on WWW, I was able to find a page of data sheets of CCD area sensors on the WWW server of Sony America. Among various CCDs shown here, ICX054AK is 1/3 inch large and has 270K pixels which matches with the catalogue spec of ES-1000/DC-20. So I wrote a decoder expecting that this CCD is used in ES-1000/DC-20. As it worked properly, I assume here that ICX054AK is the CCD used.

I would sincerely welcome objections to this assumption if you will name the alternative CCD :-).

Summary of Specs

Optical size: 1/3 inch
Chip size: 6.0mm(H) x 4.96mm(V)
Unit cell size: 9.6micron(H) x 7.5micron(V)
Number of effective pixels: 510(H) x 492(V) (approx. 250K)
Number of total pixels: 537(H) x 505(V) (approx. 270K)
Ye, Cy, Mg, G complementary color mosaic filter

Detailed specs are in the data sheet in PDF format.

Output format

Cy	Ye	Cy	Ye
G	Mg	G	Mg
Cy	Ye	Cy	Ye
Mg	G	Mg	G

Physically, this CCD consists of the color filter array of Ye (yellow), Cy (cyan), Mg (magenta), and G (green). They are aligned as shown left.

Cy+G Ye+Mg Cy+G Ye+Mg
Cy+Mg Ye+G Cy+Mg Ye+G
But we can't obtain data of each pixel separately; actual output data is calculated by adding two adjacent lines vertically as shown left.

2G+B 2R+G+B 2G+B 2R+G+B
R+G+2B R+2G R+G+2B R+2G
We can represent this output in RGB as shown left by substituting Ye = R + G, Cy = G + B, and Mg = R + B.

This format looks rather complicated; the point is that we can obtain 2R+3G+2B (which can be used as luminosity) by adding any two horizontally adjacent signals.

How to decode this output

After trials and errors, I found that following principles should be kept:

To obtain RGB value, three of the four CCD signals (i.e., 2G+B, 2R+G+B, R+G+2B, and R+2G) are needed.
Each pixel has only one signal; other two signals should be approximated from signals on pixels up, down, left, and right.
Transition of colours between adjacent pixels is continuous.
Luminosity, which can be obtained relatively correctly, is fixed first; colours are calculated based on it it.

Based on these principles, cmttoppm decodes CCD data as follows (a sample data is attached in which signal 1 is 2R+G+B, signal 2 is 2G+B, and signal 3 is R+2G):

2G+B=45

2R+G+B=108

2G+B=53

CCD data itself is used as signal 1.
Calculate signal 2 by horizontal interpolation.
- Use average of signal 2 of the pixel left and that of pixel right as the initial value of signal 2 of this pixel.
- Use average of signal1 / signal 2 of the pixel left and that of pixel right as signal 1 / signal 2 of this pixel, and derive signal 2 of this pixel from that value. This operation is repeated several times.
  
  2G+B=45
  2R+G+B=90.0
  (2R+G+B)/(2G+B)=2.00 2R+G+B=108
  2G+B=52.9
  (2R+G+B)/(2G+B)=2.04 2G+B=53
  2R+G+B=110.0
  (2R+G+B)/(2G+B)=2.08

Calculate signal 3 by vertical interpolation.

Use average of signal3 / luminosity of the pixel up and that of pixel down as signal 3 / luminosity of this pixel, and derive signal 3 of this pixel from that value.

R+2G=45
R+G+2B=90.0
(R+2G)/((R+2G)+(R+G+2B))=0.33

2R+G+B=108
2G+B=52.9

R+2G=46
R+G+2B=72.0
(R+2G)/((R+2G)+(R+G+2B))=0.39

R+2G=45
R+G+2B=90.0
(R+2G)/((R+2G)+(R+G+2B))=0.33

2R+G+B=108
2G+B=52.9
R+2G=57.9
(R+2G)/(2R+G+B)+(2G+B))=0.36

R+2G=46
R+G+2B=72.0
(R+2G)/((R+2G)+(R+G+2B))=0.39

Generate R, G, and B from signal 1, 2, and 3.

2R+G+B=108
2G+B=52.9
R+2G=57.9
->
R=33.6
G=12.1
B=28.7
Perform several postprocessing operations including Gamma correction.

Critics and improvements to this method is welcome.

Back to ES-1000 page

Please send questions and comments about this page to hideki@teikan.net.