23-06-2012, 01:16 PM
A New Color Filter Array With Optimal Properties for Noiseless and Noisy Color Image Acquisition
[attachment=25237]
INTRODUCTION
THE growing popularity of digital photography demands
every attempt of improvement in terms of quality and
speed of the features provided in digital cameras. The heart of
a digital still or video camera is its sensor, a 2-D array of photosites
that measure the amount of light absorbed during the exposure
time. The color information is obtained by means of a
color filter array (CFA) overlaid on the sensor, such that each
photosite is covered by a color filter sensitive to only a portion
of the visible light spectrum [7], [8]. From the mosaicked image
acquired by the camera, some processing is required to recover
a full color image with three components per pixel, carrying information
in the red ®, green (G) and blue (B) spectral bands to
which the human visual system (HVS) is sensitive. This reconstruction
operation is called demosaicking, see, e.g., [7]–[10]
and references therein.
CFA DESIGN AS A CONSTRAINED OPTIMIZATION
PROBLEM AND ITS SOLUTION
Class of CFAs Robust to Aliasing
We construct, step by step, the structure of a CFA with desirable
properties, enforced as design criteria in the frequency
domain. Thus, we obtain a class of CFAs parameterized by a
small set of variables. We discuss the optimization of these remaining The requirements we adopt are the following:
Condition on the Luminance: For the CFA to be physically
realizable, the values have to lie in [0, 1]. This implies
that the luminance channel appears in the baseband of .
DEMOSAICKING STRATEGY AND ITS PROPERTIES
A. Demosaicking by Frequency Selection in the Noiseless Case
The proposed CFA has a natural and simple demosaicking
algorithm associated to it, inspired by its characteristics in the
Fourier domain. This demosaicking process amounts to separate
the frequency content of the mosaicked image into the luminance
and chrominance channels of the reconstructed image.
Demosaicking by frequency selection was first explained by
Dubois, for the Bayer CFA [25]. Let us recall his method, with
our notations.
CONCLUSION
In this paper, we redefined the problem of CFA design as the
maximization of the energy of the color scene encoded in the
mosaicked image, through the choice of the gains of the CFA in
an orthonormal luminance and chrominance basis. In fact, these
gains are the inverse of the noise amplification factors in the luminance
and chrominance channels of the demosaicked image.
We derived the analytical solution to the optimization of these
gains, under the constraint that the chrominance is modulated
far away from the luminance in the Fourier domain, for robustness
to aliasing. The proposed CFA has six colors and a periodic
pattern of size 2 3. A sensor equipped with the proposed CFA
instead of the standard Bayer CFA should provide images with
higher perceived resolution (because the anti-alias filter can be
removed from the sensor) and better quality (the lower level of
noise allows the use of a less destructive denoising method).
[attachment=25237]
INTRODUCTION
THE growing popularity of digital photography demands
every attempt of improvement in terms of quality and
speed of the features provided in digital cameras. The heart of
a digital still or video camera is its sensor, a 2-D array of photosites
that measure the amount of light absorbed during the exposure
time. The color information is obtained by means of a
color filter array (CFA) overlaid on the sensor, such that each
photosite is covered by a color filter sensitive to only a portion
of the visible light spectrum [7], [8]. From the mosaicked image
acquired by the camera, some processing is required to recover
a full color image with three components per pixel, carrying information
in the red ®, green (G) and blue (B) spectral bands to
which the human visual system (HVS) is sensitive. This reconstruction
operation is called demosaicking, see, e.g., [7]–[10]
and references therein.
CFA DESIGN AS A CONSTRAINED OPTIMIZATION
PROBLEM AND ITS SOLUTION
Class of CFAs Robust to Aliasing
We construct, step by step, the structure of a CFA with desirable
properties, enforced as design criteria in the frequency
domain. Thus, we obtain a class of CFAs parameterized by a
small set of variables. We discuss the optimization of these remaining The requirements we adopt are the following:
Condition on the Luminance: For the CFA to be physically
realizable, the values have to lie in [0, 1]. This implies
that the luminance channel appears in the baseband of .
DEMOSAICKING STRATEGY AND ITS PROPERTIES
A. Demosaicking by Frequency Selection in the Noiseless Case
The proposed CFA has a natural and simple demosaicking
algorithm associated to it, inspired by its characteristics in the
Fourier domain. This demosaicking process amounts to separate
the frequency content of the mosaicked image into the luminance
and chrominance channels of the reconstructed image.
Demosaicking by frequency selection was first explained by
Dubois, for the Bayer CFA [25]. Let us recall his method, with
our notations.
CONCLUSION
In this paper, we redefined the problem of CFA design as the
maximization of the energy of the color scene encoded in the
mosaicked image, through the choice of the gains of the CFA in
an orthonormal luminance and chrominance basis. In fact, these
gains are the inverse of the noise amplification factors in the luminance
and chrominance channels of the demosaicked image.
We derived the analytical solution to the optimization of these
gains, under the constraint that the chrominance is modulated
far away from the luminance in the Fourier domain, for robustness
to aliasing. The proposed CFA has six colors and a periodic
pattern of size 2 3. A sensor equipped with the proposed CFA
instead of the standard Bayer CFA should provide images with
higher perceived resolution (because the anti-alias filter can be
removed from the sensor) and better quality (the lower level of
noise allows the use of a less destructive denoising method).