Implementation of Multi-Linear Gain Prior to Image Compression System...(IJIGSP-V7-N3-8)

Implementation of Multi-Linear Gain Prior to Image Compression System in Remote Sensing Electro-Optical Payloads

I.J. Image, Graphics and Signal Processing, 2015, 3, 51-57

Published Online February 2015 in MECS (/) DOI: 10.5815/ijigsp.2015.03.08

Implementation of Multi-Linear Gain Prior to Image Compression System in Remote Sensing

Electro-Optical Payloads

Ashok Kumar, Rajiv Kumaran

SFED/SEG/SEDA, Space Applications Centre, ISRO, Ahmedabad-India-380015

Email: ashokkumar@sac.isro.gov.in, rk_0474@sac.isro.gov.in

Abstract—Future high resolution instrument planned by compression ratio. However imaging data from remote ISRO for space remote sensing will lead to higher data sensing is also used for scientific analysis and lossy rates because of increase in resolution and dynamic range. compression techniques can have severe implications on Hence, image compression implementation becomes scientific information. mandatory. Presently designed compression technique In electro-optical imaging systems, system noise does not take account of imaging system noise like increases with input radiance. At higher radiance input, photon noise etc. This ignorance affects compression system noise is governed by photon noise and is of the system performance. As a solution, this paper proposes order of 4-10 counts (~1 to 4%) [5-6]. Generally, this MLG (Multi Linear Gain) operation prior to main photon noise is not considered in design of image compression system. With digital MLG operation, compression technique. By implementing digital Multi-captured image can be optimally adjusted to systems Linear Gain (MLG) prior to main compression system, noise. Proposed MLG operation improves compression captured image can be optimally adjusted to system noise, ratio. Simulation results show 15-30% improvement in which can help in improving image compression lossless compression ratio. However this improvement performance. depends on MLG gains and corner points which can be This paper investigates the potential of digital MLG driven by system SNR plot. MLG operation also helps in prior to image compression technique. JPEG2000 is improving SNR at lower radiance input, when lossy considered as main compression system for simulation JPEG2000 compression is used as main compression. Up purpose. Results show the MLG helps in achieving to 1-6 dB SNR improvement is observed in simulations. higher compression ratio in case of lossless compression. Proposed MLG implementation is of very low MLG also helps in achieving higher SNR for lower input complexity and planned to be used in future missions. radiance. This makes MLG implementation more useful, as previous payload‘s data analysis shows that around 60-Index Terms—Multi Linear Gain (MLG), image 95% of data belongs to illumination of <25% albedo. compression, JPEG, SNR, photon noise. For a typical 8-bit system, SNR analysis is shown in section-II. Detailed effect of wavelet decomposition in

compression system is also shown. MLG design is shown in section-III. Section-IV shows simulation results for I. INTRODUCTION

lossless compression ratio improvement. Section-V

In remote sensing electro-optical payloads, parameters shows SNR improvement for lossy compression. like spatial resolution, radiometric resolution, spectral

bands and swath determine the image data volume to be transmitted. Satellite power and transmission bandwidth

II. BACKGROUND

limit real time data transmission capability. To meet the requirement of transmission bandwidth and image data

A. SNR response of a typical remote sensing electro-volume, image compression becomes mandatory [1-2].

optical system

Various onboard image compression techniques have

For any electro-optical imaging system, noise been used in space missions [3].

Image compression techniques can be lossless or lossy. mechanism can be classified as [5, 7] These techniques were initially developed for digital still

A spatially determined so-called fixed pattern photography to reduce storage space. Lossy compression

noise (FPN) and algorithms are generally developed according to contrast

A stochastically random so-called temporal noise. sensitivity function of human visual system [4], as main

application of digital still photography is display, printing

The FPN often is corrected by a flat field correction. and feature extraction only. Lossy techniques emphasize

preservation of contrast only and thus provide higher The remaining random noise is dominated by

Implementation of Multi-Linear Gain Prior to Image Compression System in Remote Sensing Electro-Optical Payloads

A static noise floor for low image signals, which increases with temperature and exposure time, and A dynamic noise part, which is driven by the photon noise and which increases according to the square root of the signal level.

This dynamic part of noise dominates the image quality of our raw data. Furthermore, the origin sensor noise should be a white noise as the incident photons act stochastically independent of each other.

For any electro-optical imaging system, SNR can be approximated as

Signal

SNR

RadianceNoise

Radiance

MaxmumDetectableRadiance

Today in digital photography, JPEG2000 compression system is used mainly. Characteristically of this technique is extensively researched [7-12]. JPEG2000 is based on DWT transform based. One step wavelet decomposition example is shown in Fig. 3.

NoisePhoton2 NoiseDark2 NoiseReadout2 NoiseElectronics2

(1)

For a typical 8-bit imaging system (detector well capacity 20Ke), system noise with SNR is shown in Fig. 1

& 2. PSNR

is computed as 140. Other noises are considered as 0.5 counts.

Fig. 3. Wavelet decomposition example

In all of its four components, different levels/types of compression are applied. Visually it is difficult to analyze the implication produced by DWT method. Let‘s analyze DWT transform over 4X4 pixel dataset (Fig. 4).

Fig. 1. Noise plot for a typical imaging system

Fig. 2. Noise & SNR plot for a typical imaging system

Following are the few observations:

Sensor SNR increases monotonically with input signal.

SNR is governed by Photon noise at high illumination.

Maintaining the same accuracy throughout the dynamic range is of no use.

B. DWT decomposition in JPEG2000 compression

Fig. 4. Wavelet decomposition of 4X4 pixel datasets

It is quite evident that vertical frequency component cV‘ is same in both cases (independent of absolute count). However required radiometric accuracy at 50 count is much higher than the one required at 200 count. This required accuracy‘ information is not used in JPEG2000 compression system. Other transform based

Implementation of Multi-Linear Gain Prior to Image Compression System in Remote Sensing Electro-Optical Payloads

and prediction based compression system operates similarly.

III. PROPOSED MLG ALGORITHM

MLG (Multi-Linear Gain) is one data compression technique, which has been used in Resourcesat-2 AWiFS payload (ISRO-2011) to accommodate higher radiometric resolution (12-bit) data in to existing 10-bit format of Resourcesat-1 interface [13-14]. The imageries received clearly demonstrate a better overall image quality as ratio is noted. Same exercise is performed after MLG encoding.

Fig. 7. Simulation methodology

compared to Resourcesat-1 (10-bit radiometric resolution). For an imaging system shown in Fig. 2, MLG system is designed. Figure 5 shows designed MLG system response. For ADC count >58, MLG step size is kept as 2 ADC count. These corner points are derived based on system noise performance. Such MLG system can be approximated as lossless compression system. This MLG reduces 256 radiometric levels to 158 levels.

Fig. 5. Designed MLG response plot

It can be noted that MLG operation always adds quantization noise, hence minimal SNR degradation is expected at higher radiance. This is shown in Fig. 6.

Fig. 6. Effect of MLG on SNR

If MLG is performed prior to JPEG compression, MLG will reduce high spatial frequency components based on system noise on absolute count. This can help in achieving higher compression ratio and better SNR.

IV. IMPROVEMENT IN COMPRESSION RATIO WITH PROPOSED INCORPORATION OF MLG ALGORITHM A. Simulation Methodology

Simulation methodology is shown in Fig. 7. Lossless JPEG2000 compression is applied directly on 8-bit raw image (original) image and achieved best compression

Kakadu software [15] is used for JPEG2000 compression. Achieved compression ratios are compared. B. Simulation Image-sets

For simulation purpose, 11 different image-sets covering various contrast ranges and spatial resolutions are considered. Images 1 to 9 are of size 1000x1000 while images 10&11 are of 512x512. Few of these images are shown in Fig. 8, 9 & 10 with their histograms. Table 1 shows image parameters. Image mean and std-dev are computed for full frame. Line complexity is computed as an exercise to evaluate image complexity [13]. For these parameters, equations are shown below.

m

1n 1

Mean 1.1

Image(i,j) (2)

mni 0

j 0

m 1n 1Std dev

1.1 Image(i,j) Mean 2

(3) mni 0j 0

m 1n 2

Line_complexity

1m.1n

1 Image(i,j) Image(i,j 1)i 0j 0

(4)

Fig. 8. Simulation image-5 and its histogram

Implementation of Multi-Linear Gain Prior to Image Compression System in Remote Sensing Electro-Optical Payloads

Fig. 9. Simulation image-9 and its histogram

Fig. 10. Simulation image-10 and its histogram

Table 1. Image parameters of Simulation image-sets

C. Simulation Results

Fig. 11 shows that with MLG implementation higher compression ratio can be achieved. Reason behind this higher compression ratio is reduction of line complexity (shown in Fig. 12) due to MLG, as this leads to reduction in higher frequency components in image. Increase in compression ratio commensurate with data reduction by MLG operation. However the final improvement depends on contrast present in image.

Fig. 11. Achieved compression ratio comparison

Fig. 12. Effect of MLG on Line complexity

V. IMPROVEMENT IN SNR WITH PROPOSED INCORPORATION OF MLG ALGORITHM

A. Simulation Methodology

As count-wise SNR performance is important for remote sensing imaging application, full frame RMSE (Root Mean Square Error) computation is not helpful, thus not considered. Image-5 (1000X1000) & Image-9 (1000X1000) is considered for this simulation purpose. In this case, compression ratio is kept constant and at every original count SNR is computed [13,16]. Kakadu software v6.1 is used for this lossy JPEG2000 implementation [15]. Simulation methodology is shown in Fig. 13.

Fig. 13. Simulation methodology

For computing SNR at every count S‘, following

Implementation of Multi-Linear Gain Prior to Image Compression System in Remote Sensing Electro-Optical Payloads

equation is used. NoiseImaging-system is computed by Fig. 2. To increase accuracy of this measurement, only counts with at least 10 occurances are considered.

SNR(S) 20log10

S

.hist(S)

22

(S Recovered(i,j) NoiseImaging_system(S)) i,jif(Original_image(i,j) 'S')

(5)

B. Simulation Results

First of all, visual inspection is carried out in recovered images. Fig. 14 shows input image and Fig. 15 shows recovered image after 16:1 compression. Zoomed version is shown for better clarity. No artifacts are observed in recovered images. RMSE

and Peak error

comparison is also carried out, which shows no significant effect due to MLG operation.

Fig. 14. Input Image for simulation

Fig. 15. Recovered image comparison with 16:1 image compression

Fig. 16 shows SNR plot for image-5 with compression ratio 4:1. It is quite clear that MLG helps in retaining higher SNR for lower illumination. Significant SNR improvement (around 1-6 dB) was achieved because input image is optimally adjusted to input signal noise by MLG operation. Fig. 17 & 18 show SNR comparison with compression ratio 8:1 & 16:1 respectively. SNR plots are not constant for fixed JPEG compression ratio, as JPEG compression noise varies with image contrast. A minimal SNR

degradation is also observed at higher radiance.

Fig. 16. SNR comparison for Image 5 with compression of 4:1

Implementation of Multi-Linear Gain Prior to Image Compression System in Remote Sensing Electro-Optical Payloads

Fig. 17. SNR comparison for Image 5 with compression of 8:1

Fig. 18. SNR comparison for Image 5 with compression of 16:1

For image-9, SNR performance is shown in Fig. 19 with compression ratio 8:1. Around 2 dB SNR improvement is observed for lower illuminations.

Fig 19. SNR comparison for Image 9 with compression of 8:1

VI. CONCLUSION

The proposed MLG implementation prior to main compression system helps in improving compression performance. Simulation results show 15-30% improvement in compression ratio in case of lossless compression and 1-6 dB SNR improvement for lower illuminations in case of lossy compression. Proposed MLG technique can be optimally adjusted to system noise performance. Proposed technique is of very low complexity and can be implemented easily on low end FPGA. Proposed technique is a suitable candidate for onboard implementation. Simulation results can be further verified on more number of images before final implementation.

ACKNOWLEDGMENT

We gratefully acknowledge the constant encouragement and guidance received from Shri DRM

Samudraiah- Prof. Satish Dhawan Scientist, Shri Saji A Kuriakose– DD-SEDA and Shri A. S. Kiran Kumar-Director SAC. We are thankful to our colleagues of Payload Checkout Electronics Group for providing the images.

REFERENCES

[1] Ashok Kumar, Rajiv Kumaran, ―Improvement in DPCM

image comrpession technique‖, itSIP-2013, 18-19 Oct-2013, Mumbai-India, pp 280-284.

[2] Ashok Kumar, Rajiv Kumaran, ―A low complex ADPCM

image compression technique with higher compression ratio‖, IJCET, Vol 4 Issue 6, Nov-Dec (2013), pp 367-377. [3] Guoxia Yu, Tanya Vladimirova, ―Image Compression

systems on board satellites‖, Acta Astronautica, vol 64, pp 988-1005, 2009.

[4] Majid Rabbani, ―Digital Image Compression

Techniques‖,Eastman Kodak Company, Volume TT 7, Spie Optical Engineering Press, 1995.

[5] Theuwissen A., ―Solid-State Imaging with Charge-Coupled Devices‖, Kluwer Academic Publishers, 1995. [6] Filippov, A., ―How many bits are really needed in the

image pixels?‖. /articles/AT9913651997.html.

[7] Gregor Fischer, ―A Survey on Lossy Compression of

DSC Raw Data‖, SPIE proceeding Electronic Imaging 2008.

[8] F. Ebrahimi, M. Chamik, S. Winkler, ―JPEG vs.

JPEG2000: An objective comparison of image encoding quality‖, Proc. SPIE Applications of Digital Image Processing, vol. 5558, 300-308, 2004.

[9] U. Steingrimsson, K. Simon, ―Quality Assessment of the

JPEG 2000 Compression Standard‖, Proc. of the CGIV 2004 Aachen, 337-342, Germany, April 2004 , 2004.

[10] U. Steingrimsson, K. Simon, ―Perceptive Quality

Estimation: JPEG 2000 versus JPEG‖, Journal of Imaging Science and Technology, (47), 572-603, 2003.

[11] M.J. Weinberger, G. Seroussi, G. Sapiro, ―The LOCO-I

lossless image compression algorithm: principles and standardization into JPEG-LS‖, IEEE Transactions on Image Processing 9 (8) (2000) 1309–1324.

[12] W.B. Pennebaker, J.L. Mitchell, ―JPEG Still Image Data

Compression Standard‖, Chapman & Hall, New York, 1993.

[13] Ashok Kumar, ‖Low complex ADPCM image

compression technique‖, itSIP-2013, 18-19 Oct-2013, Mumbai-India, pp 276-279.

[14] Deviprasad: ―Indian Remote Sensing Satellites-

Resourcesat2 Mission Status‖, India Civil Commercial Imagery Evaluation Workshop, March 17, 2010. [15] User Manual, Kakadu software v6.1.

[16] Rafel C Gonzalez, ―Digital Image Processing using

MATLAB”.

Authors’ Profiles

Ashok Kumar, male, is a Scientist/Engineer at Space application Centre (ISRO). He completed his BE degree (Electronics and communication) from University of Rajasthan in 2007. He is currently involved in advance research and development activities for future electro-optical payloads at Sensor Front

End

Implementation of Multi-Linear Gain Prior to Image Compression System in Remote Sensing Electro-Optical Payloads

Electronics Division (Sensor Development Area). His research interest includes VLSI design, image processing, computation photography etc. He has been awarded ―ISRO team excellence award-2009‖ for ―Miniaturized camera electronics development. He has published 9 papers in various international/national conferences and peer reviewed journals.

Contacts:

4396, SFED/SEG/SEDA, Space Applications Centre, Jodhpur Tekra, Ahmedabad-380015 Phone- 91-79-26914396

Email: ashokkumar@sac.isro.gov.in

Rajiv Kumaran, male, is a Scientist/Engineer at Space application Centre (ISRO). He passed his BE (E&C) from GEC, Modasa. He joined Space Applications Center (SAC) in April 2000. Presently he is working on design and development of electronics for IRS payloads.

Contacts:

SFED / SEDA / Space Applications Center, Ahmedabad 380015

本文来源：https://www.bwwdw.com/article/kb21.html