Color Processing
Everybody knows, that existing digital cameras do not provide true, real-life colors.
Color accuracy depends on color capturing system parameters and color processing algorithms. We have replaced the original image processing, provided by camera makers, with our own algorithm (KWE). On this page you can see pictures made with Canon PowerShot G2, Minolta DiMAGE 7, and Sony F828.
You can see, that our image-processing algorithms bring more realistic colors, natural structure of the objects and overall real-life appearance.
All work was done by our software program automatically. No manual correction was made.
Canon PowerShot G2
|
|
|
Analyzing pictures made with Canon camera it is easy to notice, that Canon increases an image contrast, intensifying blue and green colors, shifting violet into the blue zone and dark brown into the violet zone. To show how KWE Processing works with real-life-objects, KWE Color Processing was applied to RAW data, received from the Canon PowerShot G2, so you can compare original processing versus alternative one.
It has to be mentioned that the flower has been shot was violet, not blue.
|
| Canon's processing |
|
KWE processing |
Minolta DiMAGE 7
On these images made with Minolta DiMAGE 7 camera you can compare Minolta's processing versus KWE processing.
With KWE processing yellow flowers keep their yellow color on the image and green leafs has the same hue as they had in a real life.
|
|
|
|
| Minolta's processing |
|
KWE processing |
Sony DSC F828
|
We applied our Color Processing to RAW data, received from a Sony F828, so you can compare a Sony's processing versus alternative processing. Notice the green marker, which looks like blue after Sony's processing.
|
|
|
 |
| Sony's processing |
|
KWE processing |
Quantitative comparison of camera makers Color Processing
New approach to the color theory, developed at KWE International, Inc. and applied to RAW image data processing, provides an opportunity to reduce the color recognition error in several times.
The Color Recognition Root-Mean-Square Error (RMSE), computed for some cameras, is shown in the table below.
The data for calculating the RMSE was obtained in experiment with Pantone colors.
| Camera |
Sensor |
RMSE of Camera Maker's processing |
RMSE of KWE processing |
| Canon PowerShot G2 |
4 Mpixels,
3 colors |
16.8 |
4.24 |
| Canon PowerShot G5 |
5 Mpixels,
3 colors |
11.4 |
3.54 |
| Minolta DiMAGE 7 |
5 Mpixels,
3 colors |
21.5 |
4.43 |
| Sony DSC F828 |
8 Mpixels,
4 colors |
12.7 |
2.95 |
Experiment with Pantone Colors
The camera calibration experiment was conducted for consumer digital cameras Canon PowerShot G2, Canon PowerShot G5, Minolta DiMAGE 7, and Sony DSC F828.
 Eight Pantone cards containing fifty-six colors were used. By illuminating Pantone cards with Illuminant D65, a set of reflected light sources was created. After correcting for the unevenness of the light distribution over the object and the unevenness of the matrix sensitivity, the source might be described by a set of sRGB numbers given on a Pantone Card for every color. Color Capturing Error was considered as the difference between data provided for the color by corresponding Pantone Card and the average data obtained as a result of RAW image data processing. To compute the RMSE, sRGB data was converted to Lab format, as it had been recommended by CIE (Commission Internationale de l'Eclairage).
For easier comparison of the results of the experiment, sRGB digital data was transformed into graphical form, where the original color, Camera Maker's color, and KWE's color were placed side by side.
|
Each cell in the tables, referred by the following links: Canon PowerShot G2, Minolta DiMAGE 7, and Sony DSC F828, has been divided into three parts:
- An Original color was placed in the left half of each color-square;
- A color received after Camera Maker's processing was placed in the right top corner;
- A color received after KWE's processing was placed in the right bottom corner.
|
The method used has several limitations:
- Pantone sRGB tables include less than a half of all colors visible by a human.
- Pantone tables have hundreds of colors, however only 56 colors are selected for the experiment.
- Lights from different spectral characteristics can cause the same color sensation. Paint, used in the Pantone Tables, is only one of many possible realizations of a color.
Even though this method has limitations, the obtained results can be useful for quantitative assessment of digital image processing improvement.
If this experiment with Pantone is conducted not with D65, but with a narrow band source (an artificial color), or if the color base is enlarged beyond the bonds of sRGB, the error could be much bigger. In order for KWE specialists to lower error for all colors, including artificial colors and colors beyond sRGB range, and reach the theoretically best solution for each particular set of optics, filters and sensors, the spectral characteristics of the particular system are necessary.
Notes:
Published: 23 September 2004
© S.N. Berzyadin
References:
KWE International, Inc. site.
|