Camerabased projector calibration, investigation of the Bala method Espen Bårdsnes Mikalsen The Norwegian Color Research Laboratory Faculty of Computer.

Camerabased projector calibration, investigation of the Bala method

Espen Bårdsnes Mikalsen

The Norwegian Color Research Laboratory

Faculty of Computer Science and Media Technology

Gjøvik University College, Gjøvik, Norway

[email protected], http://www.colorlab.no,

Supervisors: Jon Yngve Hardeberg and Jean-Baptiste Thomas

Master thesis presentations, Gjøvik, 07.06.2007

2

Outline

Introduction Implementing the Bala method

Experimental setup

Results

Conclusion

3

The Bala method

Calibration method for projectors presented by Raja Bala and Karen Braun, Xerox co, 2006.

Focuses on tone response calibration, no 3 x 3 transform matrix.

Using a digital photo camera as a luminance measurement device by calibrating camera through visual luminance matching. Using the calibrated camera to tone response calibrate projector.

4

Research questions Q1: Verification of the Bala method

An evaluation of the calibration method suggested by Bala and Braun. Implementation and performance testing with a final numeric results and analysis as a measure of performance. Separate evaluation of methods parts to identify strengths and weaknesses in the approach.

Q2: Extensions to the Bala method

In the original paper where the calibration method was presented the authors purposed some extensions that could increase the methods performance. Through implemetation and analysis these will be evaluated. Extensions are visual matching of three luminance values per estimated curve instead of one, and separate correction of R, G and B color channel replacing uniform luminance correction.

5

Simple walkthrough of the Bala method

Step 1 - gather information

Step 2 - process information to estimate projector tone response and correction curve

6

Step 1 - Visual matching of luminance A binary rasterpattern consisting of 50% black and 50% white pixels. Adjust the

background colors luminance to perceptually match binary pattern. The adjusted luminance of background is the perceptually found 50% luminance value.

7

Display calibration target Luminance patch chart ranging from min to max luminance, with horizontal and

vertical duplications of matched 50% luminance for use with non-uniformity correction.

8

Step 1 – Capture image Capture an image of the projected chart with the uncalibrated camera.

9

Step 2 – Retrive data from captured image Rotate and crop image

10

Step 2 – Retrive data from captured image Read RGB data from image. Gives 24 individual sets of RGB. RGB sets are converted to a luminance value for each patch.

11

Step 2 – Perform non-uniformity correction

Devices like cameras and projectors often suffer from some kind of spatial non-uniformity. Non-uniformity is when a device responds spatially non-uniform to a uniform input. When capturing an image of a projection, the image will suffer from both non-uniformity of projector and of camera.

This method does not correct for non-uniformity in projection, but corrects for non-uniformity in data used for calibration of camera and projector.

Correction are based on calculating differances between the spread out duplications of the 50% luminance patch.

12

Step 2 – Estimating camera tone response curve Knowing the relationship between projected luminances and target

luminances ( min, max and 50% ) makes it possible to interpolate an estimation of the cameras tone response curve.

Description Luminance Captured camera value

Projector white Yw = 1 Ypatch white

Projector black Yb Ypatch black

Mid-gray ( 1 + Yb ) / 2 Ypatch 50% lum

Perfect black 0 0

13

Step 2 - Example of estimated camera TRC

14

15

Extensions to original method

Instead of only determining camera TRC based on the 50% luminance point. Add two new luminances ( 25% and 75% ) to help determine curve.

Instead of using same correction curve for R, G and B channel. Estimate curves separatly.

16

Experiment setup

Projectors Projectiondesign ActionOne DLP, 2003 model Panasonic AX-100 LCD, 2006 model

Cameras Nikon D200 DSLR FujiFilm s7000 compact digital camera

Spectroradiometer Minolta

Room conditions Dark room, only luminance from projection

17

Results – Visual matching Experiment set up to determine if visually matched luminance values deviate from

person to person and when repeating matching of same value. Matching done at 3 luminance levels for R, G, B and gray channel.

6 observers visually matching 12 luminances 3 times. A total of 216 values were matched.

Channel Mean Minimum Maximum % Deviance

Gray 25% 0,5114 0,4748 0,5262 5,14

Gray 50% 0,6959 0,6889 0,7020 1,31

Gray 75% 0,8554 0,8419 0,8706 2,87

Red 25% 0,5096 0,4850 0,5377 5,27

Red 50% 0,6957 0,6848 0,7076 2,28

Red 75% 0,8423 0,8141 0,8594 4,53

Green 25% 0,5099 0,4831 0,5237 4,06

Green 50% 0,6939 0,6894 0,7023 1,29

Green 75% 0,8521 0,8328 0,8656 3,28

Blue 25% 0,5302 0,5204 0,5622 4,18

Blue 50% 0,7169 0,7082 0,7308 2,26

Blue 75% 0,8628 0,8496 0,8748 2,52

18

Results - non-uniformity correction

19

Results – Camera TRC estimation (ActionOne - Nikon)

20

Results – Camera TRC estimation (Panasonic - Nikon)

21

Results – Camera TRC estimation (ActionOne - Nikon)

22

Results – Estimated projector TRC (ActionOne - Nikon)

23

Results – Estimated projector TRC (Panasonic - Nikon)

24

Results – Correction w/ correction curve (ActionOne-Nikon)

25

Results – Correction w/ separate RGB correction curves (ActionOne-Nikon)

26

Results – Correction w/ correction curve (ActionOne-Nikon)

27

Conclusion Q1: Verification of the Bala method

It has to some extent been proven that calibration of projectors using this method will result in a more exact reproduction of color then for example using standard sRGB gamma correction. Correction results are better for DLP then LCD projector, probably because of LCD conforms better to the sRGB gamma curve, and correction will therefore be less necessary.

Q2: Extensions to the Bala method

It has been proven that interpolating camera TRC with not only one visually matched point, but several will improve accuracy of camera TRC, and therefore also estimated projection TRC and correction curve. If retrieving separate camera TRC for R, G and B color channel has a positive effect on method performance has not yet been proven.