Simple Camera Calibration For Light Measurements

Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

Simple Camera Calibration for Light Measurements

Joo C. Aparcio Fernandes (1), Jos A. B. Campos Neves, Manuel J. Seplveda Freitas (2), Joo Lus Afonso (3)
(1, 2, 3)

Dept. Industrial Electronics, Universidade do Minho, Campus de Azurm, Guimares, Portugal


[email protected] , [email protected] , [email protected], [email protected]

Abstract - The availability and decreasing prices of digital


photographic cameras using megapixel sensors could point to its
use as high quality light measuring devices if simple and effective
calibration procedures could be put to work.
This paper proposes such a system, using a common digital
camera combined with low cost reference color charts. The
procedure to obtain the response function of the camera, the
production of the color chart and its calibration is presented.
In this way, it is possible to obtain light measurements
otherwise only possible with sophisticated and expensive
equipments.

I.

INTRODUCTION

Fig. 1 - Reflective color chart reference

Digital photographic cameras are widely available today at


low prices due to their large scale production. Improved
systems are constantly introduced to the market at lower prices
than its predecessors. The light sensors these cameras use are
sophisticated light measuring devices, unavailable a few years
ago. However, these systems are aimed for photographic
purposes and their technical characteristics are deeply buried in
the assembled camera. The intended results are pictures and the
light information data is processed according to specifications
and methods of the equipment manufacturer that constitute his
trade secrets.
Also the scale used for digital images is a non-linear
relationship the gamma-law that further complicates the
process, mainly because it is often disregarded [1].
On the other hand, light measurements must deal with the
distribution of the electromagnetic spectrum as well as with the
intensities of the radiation. This implies the separation of the
spectral components into controlled intervals and power
measurement for each interval using a spectrometer a
sophisticate and expensive piece of equipment.
The need of a simple method to obtain equivalent results for
electrical lamp characterization lead to the method here
described.
The digital camera is used for the light power measurements.
For the spectral separation we use the classical reflective color
chart approach: the incident light is filtered by the different
color regions on the chart, in a fixed relationship with the
incident light.
To use the setup, a full calibration of both the camera and the
chart must be accomplished. A simple method is envisaged, in
order to allow for easy chart recalibration either for its

978-1-4244-1666-0/08/$25.00 '2008 IEEE

degradation in time or replacement for a new copy obtained


from the digital image of the chart in another photo printing
equipment, possibly not exactly matched to the prior one.
II. THE REFERENCE CHART
The reference chart uses reflection areas with different
reflection coefficients. In the example in Fig. 1, we placed six
grey squares from almost white to almost black in sequence at
the bottom [2]. This is the main area, from where the
calibration departs. The other color same-size squares are
intended to cover the color spectrum as much as possible: next
row comprises the basic colors components (red-green-blue
and cyan-magenta-yellow) and the other two are a mix
selection of colors. The borders are in black to avoid light to
fool the camera exposure measurement.
This image was put in a jpeg file and taken to a commercial
photo finishing machine produce several printed charts in
various sizes.
III. THE CAMERA
The chart is photographed in the standard approach: steady
mounted camera and chart, camera facing the chart, at a
distance to practically cover the useful picture area with the
chart image, lens optical axis orthogonal to the chart plane.
Any camera able to be focused to the chart and with manual
settings for the light quality and exposure can be used. We had
at hand a Nikon D1, a Coolpix 4200 and a Konica-Minolta
A100.

1025

The illumination was by daylight the standard natural


reference for the initial calibration.
Images with different exposures were taken, at 1/8, 1/4, 1/2,
1, 2 and 4 times the correct exposure (-3, -2, -1, 0, +1, +2 in
relative exposure values EV).
The images were saved in the usual jpeg file format at the
best quality and transferred to a computer with the image
analysis program.

100
90

Image Intensity

80

IV. IMAGE ANALYSIS


The images were filtered and average values of 5x5 areas
were taken at the centers of the respective square spots in Lab
mode (Photoshop). The results for the six grey areas are in
Table 1.
The succession of the values in each row relates to the light
intensity variation on the sensor due to the differences in
exposures. Differences in the columns relate to different
intensities due to different reflections for the spots in the same
image.

70
60
50
40
30
20
10
0
0

Log Exposure
Fig. 2 - Graphic for spot #3

TABLE I
LAB VALUES FOR THE GREY SPOTS

u = x G am m a

Exposure

(eq. 1).

Spot #
-2

-1

+1

+2

16

28

43

59

10

17

30

45

67

83

17

27

43

63

89

98

24

35

54

76

96

99

30

43

65

89

99

100

35

48

72

93

99

100

The difference between consecutive exposures is known the light amount was doubled for each consecutive exposure the result can be plotted in a log x scale, as in Fig. 2, were 1st
value in the xaxis corresponds to exposure -3 EV (1/8 of the
normal exposure), the 4th to 0 EV (normal exposure) and the
6th and last to +2 EV (4x normal exposure).

Practical values for Gamma are around 2.2. Coding the


electric signal by a similar function using the inverse of the
exponent, the net result is approximately linear.
The human sensor response is logarithmic and coding the
output in a log-scale, the power function translates into a linear
representation, and Gamma is the slope - just like in the H&D
representation.
Changing the vertical scale of the graph in Fig. 2 from linear
to logarithmic, the result is in Fig. 3.

V. RESPONSE CHARACTERIZATION
Early quantitative studies of the classical photographic
process were presented by Hurter and Driffield in 1890 in a
classic paper [4]. Their proposed method introduced the H&D
curve (also known as D-log E) relating optical density with the
logarithm of the exposure. The typical photochemical sensitive
material behavior produces an S-curve, with a linear central
part, surrounded by the toe and shoulder regions . The slope
of the function in the linear region is named gamma and
relates to the contrast of the process. Typical values are from
0.8 to 1.2 for negatives and 1.5 for transparencies.
With the advent of the electronic image systems for
television, the main output device was the cathode ray tube
(CRT). But it introduced one difficulty: it has a non-linear
response [1, 2, 5] empirically adjusted by a power-law as in
equation 1.

1026

100

Log Image Intensity

-3
1

10
0

Log Exposure
Fig. 3- Graphic for spot #3 using vertical Log scale

From the graphic, the relationship between the logarithms of


the exposure x axis and image intensity y axis is mostly
linear. It also presents a shoulder in the upper part (high
intensities) just like in the classical photographic film
approach. However, the low intensities do not present the
equivalent toe zone because there are no samples for very low
values.
VI. SENSOR RESPONSE CALIBRATION
Ploting the intensities variation acording to exposure for all
the spots, there should be a horizontal displacement between
the lines corresponding to the reflectance ratios between spots
(Fig. 4). The corresponding exposure ratios should match the
reflectances.
100

Spot #1
Spot #2
Spot #3
Spot #4
Spot #5
Spot #6

10

VII. FUTURE WORK

1
1

Probably, the cheapest cameras cannot provide the needed


accuracy in exposure control or the manual adjustment ranges
necessary for the method.
But, globally, an exposure shift between spots can be
computed. The ratios were 1.3, 1.4, 1.5, 1.8 and 2.4 for each
consecutive pair of spots.
Taking Spot #6 (the lightest) as reference, the ratios become
100, 77, 55, 37, 20 and 8.5%.
The global response of the sensor/camera system is plotted in
Fig. 5 for all the available data points, after taking in account
the calculated reflectance ratios of the 6 grey areas.
The global response in Fig. 5, although taken from a single
set of images, is already quite acceptable, with only minor
deviations from the expected smooth line. It can be further
optimized by averaging data from a larger set of images.
A problem yet to consider is the effect on the global response
that some in-camera image processing approaches may bring.
For instance there is a notice of recent cameras that include
some sort of dark area illumination compensation: the dark
areas are processed by the inbuilt algorithm in such a way as to
brighten up, trying the equivalent of a flash illumination. If this
is the case for the camera in use, this option should be disabled.

Fig. 4 - Graphic for all spots with vertical log scale

Plotted results show minor differences to the ideal situation.


This can be due mainly because of image values reading errors,
but also to camera exposure errors, noise problems or even
camera processing differences related to the non-standard
exposure choices.

This work is part of a larger project [3] aiming to study and


characterize the new breed of lamps, including the low-power
fluorescent compacts but also the led-type approaches that start
to appear. One of the objectives of this larger project is the
development of an automatic system able to analyze the images
and output the coefficients for the color chart and also the
spectral profile of the light emitted by the lamps.
The need for the chart calibration arises from the fact that
image colors fade in time. But with a calibration method, such
as the one proposed, even a partially faded chart can be used.
A single image of the chart taken in standard conditions
forced daylight type illumination settings, but not auto light
type adjustment will allow the direct camera calibration using
intensities of grey spots 1-6.
The corresponding light intensities ratios, computed from the
RGB components relative to a standard color light condition
such as daylight, translates into color components differences
to identify the light source quality.
The colored spots correspond to different light source
reflective filtering options and will allow further refinement of
the process.
ACKNOWLEDGMENT
The research work is being conducted with partial funding
from FCT (Portuguese Foundation for Science and
Technology).
REFERENCES
[1]

Fig. 5 - Log-log Camera/sensor response (relative values)

1027

J. C. A. Fernandes, "Linear versus Non-Linear Digital Image Processing


Methods," presented at Congresso Luso-Moambicano de Engenharia,
Maputo, Moambique, 2005.

[2]

[3]

J. C. Aparcio Fernandes , J. A. B. Campos Neves , Manuel J. S. Freitas ,


J. L. Afonso , "Calibration of Camera for Light Quality Measurements of
Low-Power Lamps ", 10th Portuguese-Spanish Conference In Electrical
Engineering - XCLEEE, 5th - 7th of July 2007, Funchal, Madeira,
Portugal
Research proposal - PTDC/EEA-ENE/72551/2006 - Power Consumption
and Spectral Radiance Characterization for Energy Saving Lamps, FCT,
Portugal, 2006

[4]
[5]

1028

Hurter, Ferdinand and Driffield, Vero Charles, Photochemical


Investigations and a New Method of Determination of the Sensitiveness
of Photographic Plates, J. Soc. Chem. Ind. May 31, 1890.
Poynton, C., The rehabilitation of gamma, in Rogowitz ,B. E. and T.
N. Pappas (eds.), Human Vision and Electronic Imaging III, Proceedings
of SPIE 3299, 232-249 ,Bellingham, Wash.: SPIE, 1998.

You might also like