Camera Response Function – MIMO aware
Renan Georgio Carvalho Cuzzuol
Dep. Of Control and Automation
Engineering
Instituto Federal do Espirito Santo
Vitoria, Brazil
renangeorgio@[Link]
Abstract—A few methods have proposed to estimate the II. PREVIOUS WORK
CRF, however most of them tend to fail due to the fact that,
this methods do not take into account the interrelation of the
In [14], Mann and Picard the idea was to use a gamma-
parameters and the spatial non-uniformities of the sensor. For curve as good representation for the CRF and assuming that
instance, the slope dependency of the CRF with respect of the exposure ratios between images were known. In [15],
camera gain and a particular exposure speed, and the Mitsunaga and Nayar proposed a method based on least
discrepancies produced by non-uniformities, like the photo square minimization to estimate the CRF. In [1], a
response nonuniformity. With that in mind this work aim to representation for CRF, referred to as EMoR. This
adress this issue by an improvement in the radiometric representation was derived from a database of real-world
calibration algorithm for chartless calibration. The keys to this camera response functions. In [16, 17] a method that
improvement are, first, using the camera gain information with estimated the CRF from the brightness transfer function
the exposure stack, to find the slope change in the CRF with (BTF) was proposed. In [18] a method based on the rank
different gain values. Second, not treat the different color minimization on the observation matrix that was constructed
channels as if they were independent regarding the image from the recovered BTFs of the exposure stack. In [19], a
irradiance. Third, register the different responses for the same method to estimate the CRF from an image sequence using a
value of scene radiance along the sensor axis. The result of this nonparametric linear least square estimation.
work is an alternative for computer vision applications that
need camera models more robust.
Keywords—radiometric calibration, camera response III. METHOD
function, brightness transfer function, exposure time. A stack os imagens is captured with the camera
remaining motionless throughout the process, for a especific
I. INTRODUCTION
exposure ratio and fixed gain. The absence of movement is
With the increasing number of computer vision validated by the absolute difference between two consecutive
applications, the requirements for a good visual system have images in the stack. After the first stack be captured, the
become more sophisticated. Many of these applications process is repited, only this time the value of the exposure
require precise measurements of scene radiance to recover ratio is changed.
the scene properties [1]. Examples of applications that
explicitly use scene radiance measurements are color This process is repited util the curves obtained for the
constancy [2, 3], stereo [4, 5, 6], construction of linear high different stacks by least-square method start to show elevated
dynamic range images [7, 8, 9], shape from shading [10], degree of correlation. When this point is reached, the value
estimation of reflectance and illumination from shape and os the gain is changed, and the process restart.
brightness [11], recovery of BRDF from images [12], and After this process is finished, the raw imagens from the
surface reconstruction using Helmholtz stereopsis [13]. stacks, start to be analyzed, and pixels to closed of the limits
The transformation that is used to connect the image are discarded to limite the influence of the sensor
brightness recorded by the visuaal system with the scene compreesed dynamic range.
radiance is refered as, Camera Response Function (CRF), When this process is ended, least-square method is
that is obtained through a process called, Radiometric applied, and a set of curves is obtained. With this set of
Calibration. The goal of this work is to provide an accurate curves is possible to represent all possible situations in a real
and robust model of the mapping from scene radiance to scene.
image brightness.
A lot of methods have been used in the past for achieving
this task. However, most of them fail when they do not take
into account the interrelation of the camera parameters and
possible non-uniformities introduced by the sensor. So, in
this work is proposed an alternative for computer vision
applications that need robust camera models.
©2021 IEEE
In In Proceedings of IS&T, 46th Annual Conference, pages 422–428,
1995.
REFERENCES [11] B.K.P. Horn and M.J. Brooks. Shape from Shading. MIT Press, 1989.
[12] Q.T. Luong, P. Fua, and Y. Leclerc. The Radiometry of Multiple
Images. IEEE Trans. on PAMI, 24(1):19–33, January 2002.
[1] Michael D. Grossberg and Shree K. Nayar , “What is the Space of
Camera Response Functions? ” , May 2014. [13] P.E. Debevec. Rendering Synthetic Objects into Real Scenes. Proc. of
ACM SIGGRAPH, 1998:189–198, July 1998.
[2] ] S. Mann and R.W. Picard, “On being undigital with digital cameras:
Extending dynamic range by combining differently exposed pictures,” [14] T. Zickler, P.N. Belhumeur, and D.J. Kriegman. Helmholtz
Proc. IST, pp.442–448, 1995.. Stereopsis: Exploiting Reciprocity for Surface Reconstruction. In
Proc. of the ECCV, page III: 869 ff., 2002.
[3] G.D. Finlayson, S.D. Hordley, and P.M. Hubel. Color by Correlation:
A Simple, Unifying Framework for Color Constancy. IEEE Trans. on [15] P.E. Debevec and J. Malik, “Recovering high dynamic range radiance
PAMI, 23(11):1209–1221, November 2001. maps from photographs,” Proc. 24th Annual Conference on Computer
Graphics and Interactive Techniques, SIGGRAPH ’97, New York,
[4] E.H. Land and J.J. McCann. Lightness and Retinex Theory. JOSA, NY, USA, pp.369–378, ACM Press/Addison-Wesley Publishing Co.,
61(1):1–11, 1971. 1997.
[5] R. Basri and D.W. Jacobs. Photometric Stereo with General, [16] S. Mann and R. Mann, “Quantigraphic imaging: Estimating the
Unknown Lighting. In Proc. of CVPR, pages II:374–381, 2001. camera response and exposures from differently exposed images,”
[6] S.K. Nayar, K. Ikeuchi, and T. Kanade. Shape from Interreflections. Proc. 2001 IEEE Computer Society Conference on Comput. Vis.
IJCV, 6(3):173–195, August 1991. Pattern Recognit., CVPR 2001, pp.842–849, 2001.
[7] R.J. Woodham. Photometric Method for Determining Surface [17] S.J. Kim, D. Gallup, J.M. Frahm, and M. Pollefeys, “Joint radiometric
Orientation from Multiple Images. Optical Engineering, 19(1):139– calibration and feature tracking system with an application to stereo,”
144, January 1980. Comput. Vis. Image Understand., vol.114, no.5, pp.574– 582, 2010.
[8] T. Mitsunaga and S. K. Nayar. Radiometric Self Calibration. In Proc. Special issue on Intelligent Vision Systems.
of CVPR, volume 2, pages 374–380, June 1999. [18] ] S.J. Kim and M. Pollefeys, “Robust radiometric calibration and
[9] P. E. Debevec and J. Malik. Recovering High Dynamic Range vignetting correction,” IEEE Trans. Pattern Anal. Mach. Intell.,
Radiance Maps from Photographs. In Proc. of ACM SIGGRAPH, vol.30, no.4, pp.562–576, April 2008.
pages 369–378, 1997. [19] M. Grossberg and S.K. Nayar, “Determining the camera response
[10] S. Mann and R. Picard. Being ’undigital’ with digital cameras: from images: what is knowable?,” IEEE Trans. Pattern Anal. Mach.
Extending dynamic range by combining differently exposed pictures. Intell., vol.25, no.11, pp.1455–1467, Nov. 2003.