ArticlePDF Available

The Automatic Selection of Radial Distortion Models

Authors:

Abstract

Several kinds of distortions exist in imaging systems which in specific circumstances may affect an image's geometry without debilitating quality or diminishing the information existing in the image. The most important type is the radial distortion which represents high distortion accurately. Many lens distortion models exist combined with several variations where different techniques are used to calibrate each distortion model. This paper presents an algorithm to select automatically the best lens distortion model for four lenses of different focal length using different statistical information criterion without sacrificing a significantly lower error. The used method requires a simple chessboard pattern, which observed from different position and calibrated using Zhang method, to compute the complexity of the lens distortion model automatically. The result shows the 6 th order radial distortion model is the best model with the minimum error about-0.273 for lens of focal length 30.64 mm using MDL criteria, while at 4 th order the minimum error about-0.177, and at 2 nd order about-0.112.
International Journal of Computer Applications (0975 8887)
Volume 179 No.40, May 2018
9
The Automatic Selection of Radial Distortion Models
Rihab K. Hamad
Optoelectronics Department/
College of Engineering
University of Technology/ Iraq
Baidaa Hamed
Physics Science Department/
College of Science
Mustansiriyah University/ Iraq
H. A. Hassonny
Control and system
Engineering Department/
College of Engineering
University of Technology/ Iraq
ABSTRACT
Several kinds of distortions exist in imaging systems which in
specific circumstances may affect an image’s geometry
without debilitating quality or diminishing the information
existing in the image. The most important type is the radial
distortion which represents high distortion accurately. Many
lens distortion models exist combined with several variations
where different techniques are used to calibrate each
distortion model. This paper presents an algorithm to select
automatically the best lens distortion model for four lenses of
different focal length using different statistical information
criterion without sacrificing a significantly lower error. The
used method requires a simple chessboard pattern, which
observed from different position and calibrated using Zhang
method, to compute the complexity of the lens distortion
model automatically.
The result shows the 6th order radial distortion model is the
best model with the minimum error about - 0.273 for lens of
focal length 30.64 mm using MDL criteria, while at 4th order
the minimum error about -0.177, and at 2nd order about -
0.112.
Keywords
Camera Calibration, Radial Distortion, Lenses
1. INTRODUCTION
Camera lens distortion is significant in a medium to wide
angle lenses. At present many distortion models and executing
are available for evaluating camera lens distortion choosing
the right model and executing could provide accurate result.
The simpler methods which are used by the computer vision
community have been developed due to the complexity of the
evaluation trials and advances in the use of computers for
analytical analysis [1, 2 and 3]. Numerous PC vision
calculations critically depended upon the supposition of a
linear pinhole camera, especially structure from motion
algorithms. Some of the proposed distortion calibration
methods consider an automatic distortion model selection
method [4, 5 and 6]. There are several kinds of lens distortion
but the most effective kind is the radial distortion. Which
bends straight lines into circular arcs, and it’s important in
higher quality cameras that introduce error into 3 dimension
reconstruction processes [7, 8]. Radial distortion may appear
as a barrel distortion, which arising usually at short focal
lengths, or pincushion distortion, which arising usually at
longer focal lengths.
2. RADIAL DISTORTION
It’s the main kind of lens distortion which formed by defects
in lens shape and displays itself only as radial positional error.
While the other types of distortion are usually created by
indecorous lens and camera gathering, produce both radial
and tangential errors in point positions. Radial distortion is a
linear motion for the image point radially from or to the image
center, due to the fact that objects at several angular spaces
from the lens axis suffer many amplifications and its lack in a
straight line transmission. It is important to indicate that radial
distortion is exceedingly connected with focal length, even if
it’s not modeled within intrinsic parameters of the camera [9,
10].
The following polynomial equation symbolizes the radial
distortion [11]:
r f(r)= r (1+ k1r2+ k2r4+ ..) (1)
Where k1, k2 are the distortion coefficients and r2 = x2 + y2.
The most commonly radial distortion models that have been
used are still in the polynomial form of (1) until recently.
The previous polynomial radial distortion models are used
with n-orders act as standards for evaluating the performance
of the radial distortion models automatically which will be
presented in results section.
3. DISTORTION MODEL SELECTION
When several competing models for a given system can
represent the distortion, the task of distortion model selection
would be good for choosing the finest model. Using the most
fitting and instructions model will give both better precision
and decreased computational model complexity. The model
with more degrees of freedom in most cases will fit the data
closer than other less complex models [5, 7]. Sum of common
and actual model selection criteria is commonly used in
different applications. These criterions shows a good perform
in computer vision applications.
In the following criteria the parameters of the equations as
follow: N refers to the number of samples, k is the parameters
number in the model, and SSE Refers to the sum-square-error
(SSE) computed as: SSE=
2
Where 2
= -` is
the difference between the measured and estimated image
points.
3.1 Akaike Information Criterion (AIC):
AIC is the first criteria presented in statistics literature for
model selection. It has the following form:
AIC = N ln (SSE N) + 2 k (2)
The model that can depict any future information adequately
with a similar distribution is correct and chosen by AIC.
In other words, AIC selects the model that reduces the error of
another perception. And it can compare very different models
[8].
3.2 Bayesian Information Criterion (BIC):
BIC is presented by Schwarz in 1978; the principle of this
criterion is as following:
International Journal of Computer Applications (0975 8887)
Volume 179 No.40, May 2018
10
Choose the model that maximizes the conditional probability
data. This data is constrained by some priori information
data. The BIC can take different structures depending on the
supposed priori data [8]. It has the following form:
BIC = N ln (SSE N) + k logN (3)
3.3 Consistent AIC (CAIC):
Bozdogan presented CAIC in 1987, it is an endeavor to
conquer the tendency of the AIC to overestimate the
complexity of the model [9]. It has the following form:
CAIC = N ln (SSE N) + k (logN + 1) (4)
3.4 Minimum Description Length (MDL):
MDL is presented by rissanen in 1978, the simplest
model that depicts the information adequately will be chosen
in MDL [10]. It has the following form:
MDL = N ln (SSE N) + 2 k logN (5)
the explained model selection criterions are shown in Table 1
Table 1. Model Selection Criterions
Model selection criterions
Formula
AIC
N ln (SSE N) + 2 k
BIC
N ln (SSE N) + k logN
CAIC
N ln (SSE N) + k (logN + 1)
MDL
N ln (SSE N) + 2 k logN
4. RESULTS
The presented algorithm shown in figure (1) consists of two
process calibration and model selection process. The
calibration was done for the three radial models (2nd, 4th and
6th order) and four lenses of different focal length 70 mm,
100mm, 150 mm, and190mm using Zhang method. A simple
chessboard pattern was observed from different position with
resolution of 1830×1330, and the model plane contained of
8×11=88 points corner. For each lens five different images
were used and calibrated. The equations (2, 3, 4 and 5) of the
criteria were applied to select the best model and show the
performance of the different radial distortion models. Figure
(2) shows one image for each lens.
Fig 1. The Presented Algorithm
Fig 2. Images captured with (a) 190mm, (b) 150mm, (c)
100mm, and (d) 70mm lens.
Tables (2, 3, and 4) show the calibration results for 2nd, 4th and
6th order radial distortion model after optimization for the
used lenses.
Table 2. Calibration result after optimization using 2nd
order distortion model for Several Lenses
different lenses
of focal length
(mm)
Radial distortion
coefficient
Pixel error
k1
u
v
30.64
0.25901
2.04594
1.91832
27.77
0.94332
3.46324
4.13295
26.31
0.52793
1.94436
2.96566
21.81
0.36694
4.43100
4.46265
Camera parameters
estimation
Fc, cc, alpha_c, kc, err
Calibration
Error analysis
Camera parameters
estimation
Fc, cc, alpha_c, kc,
err
Apply Statistic
Information Criteria
AIC, BIC, CAIC,
MDL
Select the model
with minimum score
Print Error
Model Selection Process
Calibration Process
International Journal of Computer Applications (0975 8887)
Volume 179 No.40, May 2018
11
Table 3: Calibration results after optimization using 4th
order distortion model for Several Lenses
different
lenses of
focal
length
(mm)
Radial distortion
coefficient
Pixel error
k1
k2
u
v
30.64
0.07210
-
1.49916
1.84158
2.07128
27.77
0.94949
-
1.86983
3.18991
3.97695
26.31
0.56427
8.14431
2.37465
2.23881
21.81
0.37225
0.31389
4.24701
4.44458
The complexity of 2nd order, 4th order and 6th order radial
distortion models are tested analytically using different model
selection criterions, to choose automatically the best distortion
model.
Table 4. Calibration result after optimization using 6th
order distortion model for Several Lenses
differen
t
lenses
of focal
length
(mm)
Radial distortion
coefficient
Pixel error
k1
k2
k3
u
v
30.64
0.2656
7
-
0.12787
0.0442
8
1.9329
7
1.9052
7
27.77
0.1702
1
-
0.09652
0.0297
0
2.0263
8
2.1845
5
26.31
0.6462
8
19.2709
9
-
0.0064
1.9503
8
2.4210
6
21.81
0.1376
0
0.45825
0.0422
1
2.4322
1
2.3575
2
Table 5. Complexity of four lenses using 2nd order radial
distortion model
Lenses of
focal length
(mm)
AIC
BIC
CAIC
MDL
30.64
4.839
0.936
3.936
-0.112
27.77
8.091
4.187
7.187
3.139
26.31
5.909
2.006
5.005
0.957
21.81
8.879
4.975
7.975
3.927
Table 6. Complexity of four lenses using 4th order radial
distortion model
Lenses of focal
length (mm)
AIC
BIC
CAIC
MDL
30.64
4.775
0.871
3.871
-0.177
27.77
7.799
3.895
6.895
2.847
26.31
5.597
1.694
4.694
0.646
21.81
8.764
4.860
7.860
3.812
Table 7. Complexity of four lenses using 6th order radial
distortion model
Lenses of focal
length (mm)
AIC
BIC
CAIC
MDL
30.64
4.677
0.774
3.774
-0.273
27.77
5.141
1.238
4.238
0.190
26.31
5.327
1.424
4.424
0.376
21.81
5.785
1.882
4.882
0.834
MDL is utilized over the other entire criteria in comparing the
radial distortion models and picking the best one, which it’s a
good estimate of the least complex model that will give a
model which model the distortion competently. This is
compatible with Broaddus opinion in [10], where MDL choose
the complexity equal to or less than that of the other criteria
without sacrificing a significantly lower error.
Fig 3. the complexity using MDL criterion of the 2nd and
6th order radial distortion model for four different lenses
(1) 30.64mm, (2) 27.77mm, (3) 26.31mm and (4) 21.81mm
5. CONCLUSION
A complete automatic selection for the camera lens distortion
was presented for the use with wide angle camera. Some of
information criteria have successful results in choosing the
distortion model complexity in automatic way.
Results show that the 6th order radial distortions model it’s the
best model to use which contain minimum error than 4th and 2nd
orders. The model complexity decreases gradually depending
on the focal length of the lens, as clarified in figure (3).
6. REFERENCES
[1]. J. Heikkila and O. Silven, 1997, A Four-Step Camera
Calibration Procedure with Implicit Image Correction,”
In Proceedings of IEEE Conference on Computer Vision
and Pattern Recognition.
[2]. R. Lenz and R. Tsai, 1987, Techniques for Calibration of
the Scale Factor and Image Center for High Accuracy 3D
Machine Vision Metrology," In Proceedings of the IEEE
International Conference on Robotics and Automation,
Raleigh NC.
[3]. Tsai, R., 1987, A versatile camera calibration technique
for high-accuracy 3D machine vision metrology using
off-the-shelf TV cameras and lenses. IEEE Journal on
Robotics and Automation.
[4]. El-Melegy MT, Farag AA, 2003, Nonmetric Lens
Distortion Calibration: Closed-form Solutions, Robust
Estimation and Model Selection, InICCV.
[5]. Abdul, Rihab K. Hamad Sinan M., and Sattar Razi J. Al-
azawi, 2017, Calibration of Convex Lenses with 2nd
International Journal of Computer Applications (0975 8887)
Volume 179 No.40, May 2018
12
Order Radial Distortion Model, international journal of
computation and applied science.
[6]. Kinoshita, Keisuke, and Michael Lindenbaum, 1999,
Camera model selection based on geometric AIC,
Computer Vision and Pattern Recognition. Proceedings
IEEE Conference.
[7]. Zhang, Zhengyou, 1999, Flexible camera calibration by
viewing a plane from unknown orientations, The
Proceedings of the Seventh IEEE International
Conference.
[8]. Strand R, Hayman E., 2005, Correcting Radial Distortion
by Circle Fitting, InBMVC.
[9]. Wang, A., Qiu, T., and Shao, L., 2009, A simple method
of radial distortion correction with center of distortion
estimation. Journal of Mathematical Imaging and Vision.
[10]. Ma, Lili, YangQuan Chen, and Kevin L. Moore, 2004,
Rational radial distortion
[11]. Choi, Kai San, Edmund Y. Lam, and Kenneth KY
Wong, 2006, Automatic source camera identification
using the intrinsic lens radial distortion. Optics express.
IJCATM : www.ijcaonline.org
... It requires several images of a chessboard pattern to be captured from a different position with a known dimension. The algorithm figures the camera alignment parameters utilizing the connection between the checkerboard corners in a camera arrange framework and a world coordinate system joined to the checkerboard plane [5,6]. A lot of experience after Conrady and Brown is made to treat the eliminating of distortion from images [7,8]. ...
Article
Full-text available
Lens distortion is one of the highest elements affecting camera calibration. In this paper, a new improved algorithm for calculating the inverse of the best radial distortion model is presented. The presented method provides a model of reverse radial distortion. The method involves lens of focal length (190mm) with a Sony G camera, and calibrating them using Zhang method which entails a chessboard pattern to be detected from several positions, and an exact formula to calculate the inverse radial distortion. The results show that the highest order radial distortion is the better model to be reverse, it was shown that the 8th order radial distortion is the best model which contain a lowest value of distortion can be acquired after a number of iterations about 1.009842932 × 10-24, and this is the best value can be acquired without losing the accuracy, which produces low distortion performs as pincushion distortion. The new improved algorithm displays a good performance in obtaining least value for distortion coefficient.
Article
Full-text available
This paper presents a new simple method to determine the distortion function of camera systems suffering from radial lens distortion. Neither information about the intrinsic camera parameters nor 3D-point correspondences are required. It is based on single image and uses the constraint, that straight lines in the 3D world project to circular arcs in the image plane, under the single parameter Division Model. Most of former approaches to correct the radial distortion are based on the collinearity of undistorted points. The proposed method in this paper, however, is based on the conclusion that distorted points are concyclic and uses directly the distorted points not undistorted points, therefore it should be more robust. It also computes the centre of radial distortion, which is important in obtaining optimal results. The results of experimental measurements on synthetic and real data are presented and discussed.
Article
Full-text available
Source camera identification refers to the task of matching digital images with the cameras that are responsible for producing these images. This is an important task in image forensics, which in turn is a critical procedure in law enforcement. Unfortunately, few digital cameras are equipped with the capability of producing watermarks for this purpose. In this paper, we demonstrate that it is possible to achieve a high rate of accuracy in the identification by noting the intrinsic lens radial distortion of each camera. To reduce manufacturing cost, the majority of digital cameras are equipped with lenses having rather spherical surfaces, whose inherent radial distortions serve as unique fingerprints in the images. We extract, for each image, parameters from aberration measurements, which are then used to train and test a support vector machine classifier. We conduct extensive experiments to evaluate the success rate of a source camera identification with five cameras. The results show that this is a viable approach with high accuracy. Additionally, we also present results on how the error rates may change with images captured using various optical zoom levels, as zooming is commonly available in digital cameras.
Conference Paper
Full-text available
Proposes a flexible new technique to easily calibrate a camera. It only requires the camera to observe a planar pattern shown at a few (at least two) different orientations. Either the camera or the planar pattern can be freely moved. The motion need not be known. Radial lens distortion is modeled. The proposed procedure consists of a closed-form solution followed by a nonlinear refinement based on the maximum likelihood criterion. Both computer simulation and real data have been used to test the proposed technique, and very good results have been obtained. Compared with classical techniques which use expensive equipment, such as two or three orthogonal planes, the proposed technique is easy to use and flexible. It advances 3D computer vision one step from laboratory environments to real-world use. The corresponding software is available from the author's Web page (<http://research.microsoft.com/~zhang>)
Conference Paper
This paper describes techniques for calibrating certain intrinsic camera parameters for machine vision. The parameters to be calibrated are the horizontal scale factor, i.e. the factor that relates the sensor element spacing of a discrete array camera to the picture element spacing after sampling by the image acquisition circuitry, and the image center, i.e. the intersection of the optical axis with the camera sensor. The scale factor calibration uses a 1D-FFT and is accurate and efficient. It also permits the use of only one coplanar set of calibration points for general camera calibration. Three groups of techniques for center calibration are presented: Group I requires using a laser and a four-degree of freedom adjustment of its orientation, but is simplest in concept, and is accurate and reproducible. Group II is simple to perform, but is less accurate than the other two. The most general Group III is accurate and efficient, but requires accurate image feature extraction of calibration points with known 3D coordinates. A feasible setup is described. Results of real experiments are presented and compared with theoretical predictions. Accuracy and reproducibility of the calibrated parameters are reported, as well as the improvement in actual 3D measurement due to center calibration.
In geometrical camera calibration the objective is to determine a set of camera parameters that describe the mapping between 3-D reference coordinates and 2-D image coordinates. Various methods for camera calibration can be found from the literature. However surprisingly little attention has been paid to the whole calibration procedure, i.e., control point extraction from images, model fitting, image correction, and errors originating in these stages. The main interest has been in model fitting, although the other stages are also important. In this paper we present a four-step calibration procedure that is an extension to the two-step method. There is an additional step to compensate for distortion caused by circular features, and a step for correcting the distorted image coordinates. The image correction is performed with an empirical inverse model that accurately compensates for radial and tangential distortions. Finally, a linear method for solving the parameters of the inverse model is presented
Conference Paper
A powerful and popular approach for estimating radial lens distortion param- eters is to use the fact that lines which are straight in the scene should be imaged as straight lines under the pinhole camera model. This paper revis- its this problem using the division model to parameterise the lens distortion. This turns out to have significant advantages over the more conventional pa- rameterisation, especially for a single parameter model. In particular, we demonstrate that the locus of distorted points from a straight line is a circular arc. This allows distortion estimation to be reformulated as circle-fitting for which many algorithms are available. We compare a number of suboptimal methods offering closed-form solutions with an optimal, iterative technique which minimises a cost function on the actual image plane as opposed to ex- isting techniques which suffer from a bias due to the fact that they optimise a geometric cost function on the undistorted image plane.
Conference Paper
We address the problem of calibrating camera lens distortion, which can be significant in medium to wide angle lenses. While almost all existing nonmetric distortion calibration methods need user involvement in one form or another, we present an automatic approach based on the robust the-least-median-of-squares (LMedS) estimator. Our approach is thus less sensitive to erroneous input data such as image curves that are mistakenly considered as projections of 3D linear segments. Our approach uniquely uses fast, closed-form solutions to the distortion coefficients, which serve as an initial point for a nonlinear optimization algorithm to straighten imaged lines. Moreover we propose a method for distortion model selection based on geometrical inference. Successful experiments to evaluate the performance of this approach on synthetic and real data are reported.
Calibration of Convex Lenses with 2 nd
  • Rihab K Abdul
  • M Hamad Sinan
  • Sattar Razi
  • J Alazawi
Abdul, Rihab K. Hamad Sinan M., and Sattar Razi J. Alazawi, 2017, Calibration of Convex Lenses with 2 nd
Camera model selection based on geometric AIC, Computer Vision and Pattern Recognition
  • Keisuke Kinoshita
  • Michael Lindenbaum
Kinoshita, Keisuke, and Michael Lindenbaum, 1999, Camera model selection based on geometric AIC, Computer Vision and Pattern Recognition. Proceedings IEEE Conference.