International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 550

ISSN 2229-5518

Multi-channel Image Deblurring using

Blind Deconvolution

Gururaju 1, Siddaraju K. 2

1. Research Scholar

2. Research Scholar and Assistant. Professor of Computer Science, .Maharani’s Science College for

Women, Mysore-570005, Karnataka, India

Abstract_The proposed restoration involves a DCT domain zonal filtering pre-processing followed by a post- processing step of time domain deconvolution. The technique applies to situations on which the scene consists of a finite support object against a uniformly black, gray or white and color backgrounds. Preliminary simulations in noise- free and noisy cases are conducted. The results are compared with standard Median filter for image denoising and the proposed blind restoration scheme is shown to exhibit improvement in SNR.

1 INTRODUCTION:

——————————  ——————————

2. FLOWCHART FOR THE ITERATIVE

MULTICHANNEL BLIND IMAGE RESTORATION:

The process of simultaneously estimating the PSF and restoring an unknown image using partial or no information about the imaging system is known as Blind Image restoration. For the linear degradation model when the noise item is neglected, it is specially referred to as blind deconvolution. There exit several motivating factors for the use of blind deconvolution in image processing applications. In many situations, it is difficult to accurately measure the degradation using calibration or on-line identification techniques; in addition, it is costly, dangerous, or physically impossible to obtain a priori information about the scene to be imaged. For example, in remote sensing and space imaging, fluctuations in the PSF are difficult to characterize as a random process, and there is difficulty in statistically modeling the original image. In addition, the use of adaptive optics system is often too expensive for some observation facilities, and the potential for phase error exists with cheaper partially compensating systems. Thus, technique such as blind deconvolution will be required.

Fig: 1 Iterative Multi-channel blind image restoration

2.1 Iterative Multi-channel Blind image

Restoration

Computer simulation results are shown in the section 2.2 for multi-channel blind restoration using blind deconvolution. Classical Lena and Barbara colored images shown in fig 3 were chosen as test images. Three blurred channels of the same
original image are considered. The images are

International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 551

ISSN 2229-5518

blurred wsith motion blur at an angle of 20,30,40 degrees. The blurred images are passed through a deblurring filter iterative blind deconvolution. In case of blind deconvolution the blur function or the Point Spread Function (PSF) is unknown or very little statistical information is available about the original image. An initial guess on the PSF is made the guessed PAF is convolved with the degraded image in order to obtain an estimate of the true image. The Improvement in Signal to Noise Ratio (ISNR) and Mean Square Error (MSE) is found. The experiment is repeated until we get a appropriate guess of PSF for which the ISNR is maximum and MSE is minimum. This final PSF is then deconvolved with the degraded image in order to obtain the estimate of true image. These steps are repeated for all the channels in order to obtain the restored images; from the multiple restored images a single output image is obtained which satisfies most of the assumption made about the true image.

2.2 Simulation Results:

1. Single-Input Multi-Output:- Lena Image

Fig.2 a: Original Lena image with SNR- 25 dB for

1st Channel, SNR -20 dB for 2nd Channel and
SNR=15dB for 3rd channel

II. Multi-Input Single-Output:- Lena

Image

Fig.2b: Restored Lena with ISNRI=5.6260dB for 1st channel, ISNR2=5.2743dB for 2nd channel, ISNR3=5.0615dB for 3rd channel and MSE=189

I. Single-Input Multi-Output:- Barbara

Image

Fig 3a: Original Barbara image with SNR -25dB for

1st channel, SNR-20 dB for 2nd channel, SNR=15dB
for 3rd channel

International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 552

ISSN 2229-5518

II. Multi-Input Single-Output:- Barbara

Image

Fig3b: Restored Barbara with ISNR1=2.3492dB for

1st channel, ISNR2=2.6013 dB for 2nd channel, ISNR=2.7203dB for 3rd channel and MSE=393.91

3. Multichannel Image Denoising Using Zonal

Filter

The presence of noise is visually annoying and makes it more difficult to perform tasks such as image restoration, segmentation, recognition, and scene interpretation. It then becomes desirable to construct a good estimate of the original image from the blurred noisy observations. Initially Median filters were considered to remove the noise present in the image. Median filters effectively remove the noise by substituting the pixel position with the median of the input data. Intuitively the filtering should be applied only to the corrupted pixels while leaving those uncorrupted ones intact.
We have proposed a new noise removal technique called zonal coding or zonal filter for multichannel color images. A zonal mask is applied to the low frequency components and higher frequency components are discarded. Thus the noise, which is present in the image, is filtered or removed. Experimental results show that Zonal filters have proven to achieve a good performance
over Median filters.

3.1 Flow chart Multichannel images using zonal filter

Fig4: Multi-Channel restoration using zonal filter

Computer simulation results for multichannel colored image denoising using zonal filter are shown in section 3.2. The classical Lena and Barbara colored imaged shown in fig 4.3 were chosen as test images. Initially the experiments were carried with median filters and the results were compared with zonal filters.
Three different noisy versions of the test image were considered which were corrupted by AWGN (SNR-25dB,SNR-20dB and SNR-15dB). Then the degraded or the noisy images were compressed into a JPEG format. Zonal mask shown in the fig
4.2a was applied i.e., 8x8 block DCT was considered and the mask was applied only to the lower frequency components ant the higher frequency components were discarded as most of the noise resides in the higher frequency components. Thus the noise, which is present in the image, is removed to greater extent. This process is applied to all the blocks of the entire image. The resultant images were subjected to inverse DCT and the images obtained were the estimate of the true image. Finally from the multiple-channels a single output image is obtained which satisfies most of the assumption made about the true image.

International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 553

ISSN 2229-5518

3.2 Simulation Results:

1. Single-Input Multi-Output:- Lena

Image (Median Filter)

Fig5a: original Lena using Median filter with SNR-

25dB for 1st channel, SNR-20dB for 2nd channel and
SNR-15dB for 3rd channel

II Multi-Input Single-output:- Lena Image

(Median Filter)

Fig5b: Restored Lena with ISNR1=2,456dB for 1st channel, ISNR2=3.5840dB for 2nd channel, ISNR3=4.0231dB for 3rd channel and MSE=280

I. Single-Input Multi-Output:-Lena image (Zonal Filter0.25 BR)

Fig 6a: Original Lena using zonal filter with SNR-

25dB for 1st channel, SNR-20dB for 2nd channel and
SNR-15dB for 3rd channel at 0.25 Bit Rate

II. Multi-Input Single-Output:- Lena Image

(Zonal filter0.25 BR)

Fig6b: Restored Lena with ISNR1=4.9935dB for 1st channel, ISNR2=5.5840dB for 2nd channel, ISNR3=6.0231dB for 3rd channel and MSE=199

International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 554

ISSN 2229-5518

I. Single-Input Multi-Output:- Barbara

Image (Median Filter)

Fig 7a: Original Barbara using Median filter with SNR-25dB for 1st channel, SNR-20dB for 2nd channel and SNR-15dB for 3rd channel

I. Multi-Input Single-Output:- Barbara

Image (Median Filter)

I. Single-Input Multi-Output:- Barbara Image

(Zonal Filter)

Fig 8a: original Barbara using zonal filter with SNR-25dB for 1st channel, SNR-20dB for 2nd channel and SNR-15dB for 3rd channel for 0.25 Bit Rate

I. Multi-Input Single-Output:- Barbara Image

(Zonal Filter)

Fig7b: Restored Barbara with ISNR1=26.41dB for

1st Channel, ISNR2=3.5840dB for 2nd channel, ISNR3=3.9912dB for 3rd channel and MSE=480

Fig. 8b: Restored Barbara with ISNR1=3.5497dB for

1st channel, ISNR2=4.4923 for 2nd channel, ISNR3=5.9945 for 3rd channel and MSE=420.54

International Journal of Scientific & Engineering Research, Volume 6, Issue 3, March-2015 555

ISSN 2229-5518

4. Multichannel Restoration of Blurred Noisy

Images using Zonal Filter

The proposed Multi-Input Single-Output model of section 4.6 is used for restoring blurred noisy images. Computer simulation results for multichannel restoration of blurred noisy colored image using zonal filter is shown in section 4.1. The classical Lena and Barbara colored images shown in fig 4.3 were chosen as test images. Initially the experiments were carried with median filters and the results were compared with zonal filters.
Three different Blurred Noisy versions of the test image were considered which were corrupted by AWGN and motion blur (BSNR1-
60dB for 1st channel, BSNR2-40dB for 2nd channel and BSNR3-50dB for 3rd channel), which is represented by.

Where ‘g’ denote the observed image, degraded by a motion blur with a done dimensional kernel m=(m1,…., mK) at an angle a, ‘f’ the original image degraded according to Eq(5.1). The motion blur in each channel is assumed at an angle of 20,30 and
40 degrees. The blurred nosy images were
compressed into a JPEG format. Zonal mask was applied to the blurred nosy versions in order to remove the noise contained in it i.e., 8x8 block DCT was considered and the mask was applied only to the lower frequency components and the higher frequency components were discarded as most of the noise resides in the higher frequency components. Then inverse DCT is applied to obtain the noise free image. In order to remove the blur, the deblurring filter i.e., Blind deconvolution was applied, this process was repeated to three different channels. Finally from the three multi- channels a single restored image is obtained by linear combination, which fulfills most of the assumptions made about the true image. Thus the
proposed restoration model involves a pre-
processing DCT domain zonal followed by a post- processing step of time domain deconvolution.
Experimental results show that zonal filters have proven achieve a greater noise reduction when compared to other noise reduction technique such as median filter.

4.1 Simulation results:

I. Single-Input Multi-Output:- Lena

Image (Median Filter)

Fig9a: original Lena using Median filter with BSNR1-60dB for 1st channel, BSNR-2-40dB for 2nd channel and BSNR3-50dB for 3rd channel.