20-06-2011, 10:58 AM
[attachment=14127]
Towards Optimal Color Image Coding using Demosaicing
Abstract—We introduce an optimal approach to color image compression using demosaicing. Accordingly, we propose an efficient algorithm for color image coding. In the encoder, a
mosaic of the primary colors is encoded instead of the full color image. This mosaic is considered as four different color channels that are compressed using subband transform coders. In the decoder, a demosaicing algorithm is applied to the decompressed channels to reconstruct the full image. We use optimized color transforms both in the encoding stage and in the decoding
stage. Our experimental results show that the proposed method outperforms currently available techniques with regard to visual and quantitative measures and is efficient for color image coding.
I. INTRODUCTION:
In this work we introduce a new compression approach to color images based on the naturally high inter-color correlations of the RGB primaries. Various methods have
been proposed in order to reduce the amount of data that is actually encoded, such as transforming the RGB primaries into a new color space such as YUV, and then encoding the new
color components at different rates according to their energy concentration or visual significance.
Here, however, we present a new image compression approach based on image demosaicing and a coding algorithm introduced recently [6]. A Bayer pattern [7] of a given color image
is used. The Bayer pattern (Fig. 1) consists of a mosaic of the pixels of the primary colors, so that the green occupies half of the pattern while the red and the blue occupy a quarter
each. Interpolation is a fundamental problem in image processing to re-sample the image size [1]. Due to the
physical limitation of imaging hardware, image interpolation techniques are often employed to
reconstruct a higher resolution (HR) image from its low resolution (LR) counterpart. Image
interpolation is widely used in digital photographs, medical imaging and remote sensing, etc, and
many interpolation algorithms have been proposed, including the simple but fast linear interpolators
[2-5] and those more complex non-linear interpolators [6-15].
Most of the existing interpolation schemes assume that the original image is noise free. This
assumption, however, is invalid in practice because noise will be inevitably introduced in the image
acquisition process. Usually denoising and interpolation are treated as two different problems and they
are performed separately. However, this may not be able to yield satisfying result because the
denoising process may destroy the edge structure and introduce artifacts, which can be further
amplified in the interpolation stage. With the prevalence of inexpensive and relatively low resolution
digital imaging devices (e.g. webcam, camera phone), demands for high-quality image denoising and
interpolation algorithms are being increased. Hence new interpolation schemes for noisy images need
to be developed for better suppressing the noise-caused artifacts and preserving the edge structures.
Instead of performing denoising and interpolation separately, it is possible to perform
interpolation and denoising jointly to reduce the artifacts introduced in the denoising process. Actually,
both denoising and interpolation can be viewed as an estimation problem. Denoising is to estimate the
original pixels from the noisy measurements, while interpolation is to estimate the missing sample
from its local neighbors. The idea of joint denoising and interpolation has been exploited in [16-17].
Hirakawa et al [16] proposed an algorithm of joint denoising and color demosaicking (which is a
special case of interpolation) to reconstruct the full color image from the color filter array image. The
total least square technique was used to estimate the noiseless and missing color components. In [17],
Zhang et al developed a directional estimation and wavelet based denoising scheme for joint denoising
and color demosaicking. A well designed scheme of interpolating noisy images can generate fewer
artifacts and preserve better edge structures compared with schemes that perform denoising
interpolation separately.
Edge preservation is a crucial issue in both denoising and interpolation. A general principle is to
smooth the noise and interpolate the missing samples along the edge direction, instead of across the
edge direction. As a directional image decomposition tool, wavelet transform has been widely used in
image denoising [18-23]. Another important class of denoising techniques is partial differential
equation (PDE) based anisotropic diffusion (AD). The representative anisotropic filters include the
models proposed by Perona and Malik [24] and Alvarez et al [25]. The total variation minimization
method was proposed by Rudin et al [26] to minimize the energy function related to nonlinear
anisotropic diffusion. An iterated total variation refinement scheme was developed in [27], and the
iterative regularization method was generalized to nonlinear inverse scale space and applied to wavelet
based denoising [28-29]. Recently, regularization and PDE-based techniques for solving inverse
problems have been attracting much research attention [30-31, 35-36]. They are able to perform both
zooming and denoising within the same framework.
The directional information also plays a key role in interpolation. The traditional linear
interpolation methods, such as bi-linear and bi-cubic interpolation, are simple and fast but they do not
work well in edge preservation due to the ignorance of local directional information. Most of the later
developed interpolation techniques aim at maintaining the edge sharpness. Jensen and Anastassiou [6]
detected edges and fitted them by some predefined templates to improve the visual perception of
enlarged images. The image interpolator by Carrato and Tenze [7] first replicates the pixels and then
corrects them by using some preset 3×3 edge patterns and optimizing the parameters in the operator.
Muresan [10] detected the edge in diagonal and non-diagonal directions and then recovered the
missing samples along the detected direction by using 1-D polynomial interpolation. In [8], Li et al
estimated the local covariance matrix, which can reflect the local directional information, from the LR
image, and then used it to calculate the interpolation parameters. Chang et al [11] adaptively estimated
the singularity of wavelet coefficients to predict the unknown image edge information for image
zooming. The EASE scheme in [14] tries to correct the interpolation error generated by bilinear
interpolator to enhance the edge structure. Other edge preserving interpolators can be found in [13, 15].
Towards Optimal Color Image Coding using Demosaicing
Abstract—We introduce an optimal approach to color image compression using demosaicing. Accordingly, we propose an efficient algorithm for color image coding. In the encoder, a
mosaic of the primary colors is encoded instead of the full color image. This mosaic is considered as four different color channels that are compressed using subband transform coders. In the decoder, a demosaicing algorithm is applied to the decompressed channels to reconstruct the full image. We use optimized color transforms both in the encoding stage and in the decoding
stage. Our experimental results show that the proposed method outperforms currently available techniques with regard to visual and quantitative measures and is efficient for color image coding.
I. INTRODUCTION:
In this work we introduce a new compression approach to color images based on the naturally high inter-color correlations of the RGB primaries. Various methods have
been proposed in order to reduce the amount of data that is actually encoded, such as transforming the RGB primaries into a new color space such as YUV, and then encoding the new
color components at different rates according to their energy concentration or visual significance.
Here, however, we present a new image compression approach based on image demosaicing and a coding algorithm introduced recently [6]. A Bayer pattern [7] of a given color image
is used. The Bayer pattern (Fig. 1) consists of a mosaic of the pixels of the primary colors, so that the green occupies half of the pattern while the red and the blue occupy a quarter
each. Interpolation is a fundamental problem in image processing to re-sample the image size [1]. Due to the
physical limitation of imaging hardware, image interpolation techniques are often employed to
reconstruct a higher resolution (HR) image from its low resolution (LR) counterpart. Image
interpolation is widely used in digital photographs, medical imaging and remote sensing, etc, and
many interpolation algorithms have been proposed, including the simple but fast linear interpolators
[2-5] and those more complex non-linear interpolators [6-15].
Most of the existing interpolation schemes assume that the original image is noise free. This
assumption, however, is invalid in practice because noise will be inevitably introduced in the image
acquisition process. Usually denoising and interpolation are treated as two different problems and they
are performed separately. However, this may not be able to yield satisfying result because the
denoising process may destroy the edge structure and introduce artifacts, which can be further
amplified in the interpolation stage. With the prevalence of inexpensive and relatively low resolution
digital imaging devices (e.g. webcam, camera phone), demands for high-quality image denoising and
interpolation algorithms are being increased. Hence new interpolation schemes for noisy images need
to be developed for better suppressing the noise-caused artifacts and preserving the edge structures.
Instead of performing denoising and interpolation separately, it is possible to perform
interpolation and denoising jointly to reduce the artifacts introduced in the denoising process. Actually,
both denoising and interpolation can be viewed as an estimation problem. Denoising is to estimate the
original pixels from the noisy measurements, while interpolation is to estimate the missing sample
from its local neighbors. The idea of joint denoising and interpolation has been exploited in [16-17].
Hirakawa et al [16] proposed an algorithm of joint denoising and color demosaicking (which is a
special case of interpolation) to reconstruct the full color image from the color filter array image. The
total least square technique was used to estimate the noiseless and missing color components. In [17],
Zhang et al developed a directional estimation and wavelet based denoising scheme for joint denoising
and color demosaicking. A well designed scheme of interpolating noisy images can generate fewer
artifacts and preserve better edge structures compared with schemes that perform denoising
interpolation separately.
Edge preservation is a crucial issue in both denoising and interpolation. A general principle is to
smooth the noise and interpolate the missing samples along the edge direction, instead of across the
edge direction. As a directional image decomposition tool, wavelet transform has been widely used in
image denoising [18-23]. Another important class of denoising techniques is partial differential
equation (PDE) based anisotropic diffusion (AD). The representative anisotropic filters include the
models proposed by Perona and Malik [24] and Alvarez et al [25]. The total variation minimization
method was proposed by Rudin et al [26] to minimize the energy function related to nonlinear
anisotropic diffusion. An iterated total variation refinement scheme was developed in [27], and the
iterative regularization method was generalized to nonlinear inverse scale space and applied to wavelet
based denoising [28-29]. Recently, regularization and PDE-based techniques for solving inverse
problems have been attracting much research attention [30-31, 35-36]. They are able to perform both
zooming and denoising within the same framework.
The directional information also plays a key role in interpolation. The traditional linear
interpolation methods, such as bi-linear and bi-cubic interpolation, are simple and fast but they do not
work well in edge preservation due to the ignorance of local directional information. Most of the later
developed interpolation techniques aim at maintaining the edge sharpness. Jensen and Anastassiou [6]
detected edges and fitted them by some predefined templates to improve the visual perception of
enlarged images. The image interpolator by Carrato and Tenze [7] first replicates the pixels and then
corrects them by using some preset 3×3 edge patterns and optimizing the parameters in the operator.
Muresan [10] detected the edge in diagonal and non-diagonal directions and then recovered the
missing samples along the detected direction by using 1-D polynomial interpolation. In [8], Li et al
estimated the local covariance matrix, which can reflect the local directional information, from the LR
image, and then used it to calculate the interpolation parameters. Chang et al [11] adaptively estimated
the singularity of wavelet coefficients to predict the unknown image edge information for image
zooming. The EASE scheme in [14] tries to correct the interpolation error generated by bilinear
interpolator to enhance the edge structure. Other edge preserving interpolators can be found in [13, 15].