logo

SCIENTIA SINICA Informationis, Volume 49, Issue 6: 708(2019) https://doi.org/10.1360/N112017-00293

Image NSST-HMT model with associated multi-state coefficients

More info
  • ReceivedMay 14, 2018
  • AcceptedSep 5, 2018
  • PublishedJun 6, 2019

Abstract

In recent years, due to its anisotropy, multi-directional capture characteristics, and translation invariance, the non-subsampled shearlet transform (NSST) hasplayed an important stabilizing role inthe process of image restoration. In this study, weanalyze an image's NSST coefficients, including the relationship betweencoefficients in the same subband, the relationship between“father-son” coefficients, and the relationship between “brotherhood”coefficients in different subbands. The results reveal that thecoefficients in the NSST subbands are sparse, and both “father-sonrelationship” and “brotherhood relationship” coefficients exhibitaggregation and transitivity. On this basis, a hidden Markov tree (HMT)model with associated multi-state coefficients (M-NSST-HMT) isproposed. This model estimates the reconstructed coefficientsusing “father-son relationship” and “brotherhood relationship”of the NSST subband coefficients as joint states of guiding thecoefficients' transfer between subbands. In addition, the model integrates thereconstructed coefficients using the mutual information betweenthese two associated states. Finally, the proposed model is appliedto image denoising with favorable results. The results indicatethat the proposed model can reveal the relationship ofcoefficients in NSST subbands and improve the prediction accuracy ofcoefficients more effectively than the traditional HMT model.


Funded by

国家自然科学基金(41671439,61402214)

辽宁省高等学校创新团队支持计划(LT2017013)

大连市青年科技之星项目支持计划项目(2015R069)

南京大学计算机软件新技术国家重点实验室开放课题项目(KFKT2018B07)


References

[1] Crouse M S, Nowak R D, Baraniuk R G. Wavelet-based statistical signal processing using hidden Markov models. IEEE Trans Signal Process, 1998, 46: 886-902 CrossRef ADS Google Scholar

[2] Portilla J, Strela V, Wainwright M J. Image denoising using scale mixtures of gaussians in the wavelet domain. IEEE Trans Image Process, 2003, 12: 1338-1351 CrossRef PubMed ADS Google Scholar

[3] Jin H Y, Jiao L C, Liu F. SAR image de-noising based on curvelet domain hidden markov tree models. Chinese Journal of Computers, 2007, 30: 491--497. Google Scholar

[4] Hou B, Xu J, Liu F, et al. Image Segmentation Using Second Generation Bandelet-domain Hidden Markov Tree Models. Acta Automatica Sinica, 2009, 35: 498--504. Google Scholar

[5] Po D D Y, Do M N. Directional multiscale modeling of images using the contourlet transform. IEEE Trans Image Process, 2006, 15: 1610-1620 CrossRef ADS Google Scholar

[6] Wang X H, Chen M Y, Song C M. Contourlet HMT model with directional feature. Sci China Inf Sci, 2012, 55: 1563-1578 CrossRef Google Scholar

[7] Wang X H, Ni P G, Su X, et al. The nonsubsampled Contourlet HMT model. Sci Sin Inform, 2013, 43: 1431--1444. Google Scholar

[8] Chen P, Zhang Y, Jia Z. Remote Sensing Image Change Detection Based on NSCT-HMT Model and Its Application.. Sensors, 2017, 17: 1295-1310 CrossRef PubMed Google Scholar

[9] Shahdoosti H R, Hazavei S M. Image denoising in dual contourlet domain using hidden Markov tree models. Digital Signal Processing, 2017, 67: 17-29 CrossRef Google Scholar

[10] Easley G, Labate D, Lim W Q. Sparse directional image representations using the discrete shearlet transform. Appl Comput Harmonic Anal, 2008, 25: 25-46 CrossRef Google Scholar

[11] Guo K, Labate D. Optimally Sparse Multidimensional Representation Using Shearlets. SIAM J Math Anal, 2007, 39: 298-318 CrossRef Google Scholar

[12] Kutyniok G, Labate D. Resolution of the Wavefront Set Using Continuous Shearlets. Trans American Math Soc, 2009, 361: 2719--2754. Google Scholar

[13] Kutyniok G, Labate D. Shearlets: Multiscale Analysis for Multivariate Data. Basel: Birkhauser, 2012. Google Scholar

[14] Easley G R, Labate D. Image Processing Using Shearlets. Boston: Birkhauser, 2012. 283--325. Google Scholar

[15] Bosse S, Chen Q, Siekmann M, et al. Shearlet-based reduced reference image quality assessment. In: Proceedings of IEEE International Conference on Image Processing, 2016. 2052--2056. Google Scholar

[16] Duval-Poo M A, Noceti N, Odone F. Scale Invariant and Noise Robust Interest Points With Shearlets. IEEE Trans Image Process, 2017, 26: 2853-2867 CrossRef PubMed ADS arXiv Google Scholar

[17] Hou B, Zhang X, Bu X. SAR Image Despeckling Based on Nonsubsampled Shearlet Transform. IEEE J Sel Top Appl Earth Observations Remote Sens, 2012, 5: 809-823 CrossRef ADS Google Scholar

[18] Jafari S, Ghofrani S. Using two coefficients modeling of nonsubsampled Shearlet transform for despeckling. J Appl Remote Sens, 2016, 10: 015002 CrossRef ADS Google Scholar

[19] Liu X, Mei W, Du H. Structure tensor and nonsubsampled shearlet transform based algorithm for CT and MRI image fusion. Neurocomputing, 2017, 235: 131-139 CrossRef Google Scholar

[20] Dempster A P, Laird N M, Rubin D B. Maximum Likelihood from Incomplete Data Via the EM Algorithm (with discussion). J Royal Stat Soc, 1977, 39: 1--38. Google Scholar

[21] Olea R A, Pawlowsky-Glahn V. Kolmogorov-Smirnov test for spatially correlated data. Stoch Environ Res Risk Assess, 2009, 23: 749-757 CrossRef Google Scholar

[22] Maes F, Collignon A, Vandermeulen D. Multimodality image registration by maximization of mutual information.. IEEE Trans Med Imag, 1997, 16: 187-198 CrossRef PubMed Google Scholar

[23] Lu Q W, Chen W F. Image segmentation based on mutual information. Chin J Comput, 2006, 29: 296--301. Google Scholar

[24] Irum I, Shahid M A, Sharif M, et al. A Review of Image Denoising Methods. J Eng Sci Tech Rev, 2015, 8: 41--48. Google Scholar

[25] Gupta K, Gupta S K. Image Denoising Techniques- A Review paper. Int J Innovative Tech Exploring Eng, 2013, 2: 2278--3075. Google Scholar

[26] Shen H, Li X, Cheng Q, et al. Missing information reconstruction of remote sensing data: a technical review. IEEE Geosci Remote Sens Mag, 2015, 3: 61--85. Google Scholar

[27] Meng X, Li J, Shen H. Pansharpening with a Guided Filter Based on Three-Layer Decomposition.. Sensors, 2016, 16: 1068-1082 CrossRef PubMed Google Scholar

[28] Shen H, Meng X, Zhang L. An Integrated Framework for the Spatio-Temporal-Spectral Fusion of Remote Sensing Images. IEEE Trans Geosci Remote Sens, 2016, 54: 7135-7148 CrossRef ADS Google Scholar

[29] Li J, Yuan Q, Shen H. Noise Removal From Hyperspectral Image With Joint Spectral-Spatial Distributed Sparse Representation. IEEE Trans Geosci Remote Sens, 2016, 54: 5425-5439 CrossRef ADS Google Scholar

[30] Kumar G A, Kusagur A. Evaluation of image denoising techniques a performance perspective. In: Proceedings of International Conference on Signal Processing, 2016. 1836--1839. Google Scholar

[31] Shao Y H, Deng N Y, Chen W J. A proximal classifier with consistency. Knowledge-Based Syst, 2013, 49: 171-178 CrossRef Google Scholar

[32] Huang Y F. Based on support vector machines image denoising and image quality assessment research. Dissertation for Master Degree. Shanxi: North University of China, 2013. Google Scholar

[33] Gasparini M. Markov chain Monte Carlo in practice. Technometrics, 1999, 39: 338--338. Google Scholar

[34] Dabov K, Foi A, Egiazarian K. Image denoising with block-matching and 3D filtering. Proc SPIE-IS&T Electronic Imaging, 2006, 6064: 354--365. Google Scholar

[35] Pierazzo N, Morel J M, Facciolo G. Multi-scale DCT denoising. 2017, 7: 288--308. Google Scholar

  • Table 1   TheKolmogorv-Smirov of GMM fitting for test images in Figure
    2*Test imageThe 1st direction subband The 1st direction subband The 1st direction subband
    in the 1st layer in the 2nd layer in the 3rd layer
    Test-image1 0.0154 0.0211 0.0366
    Test-image2 0.0126 0.0125 0.0399
    Bridge 0.0173 0.0275 0.0497
  • Table 2   The parameters of Test-image1's M-NSST-HMT model
    2* $K=1$ $K=2$
    2* $K=1$ $K=2$
    State1 State2State1 State2 State1 State2State1 State2
    $P_{1,k,\rho}$0.35 0.65 0.30 0.70 $P_{1,k,c}$0.70 0.30 0.28 0.72
    2*$A_{2,2k-1,\rho}$ State10.380.94 0.71 0.01
    2*$A_{2,2k-1,c}$State10.97 0.02 0.20 0.85
    State20.62 0.06 0.29 0.99 State20.03 0.98 0.80 0.15
    2*$A_{2,2k,\rho}$ State10.630.04 0.24 0.99
    2*$A_{2,2k,c}$ State10.89 0.13 0.01 0.93
    State20.37 0.96 0.76 0.01 State20.11 0.87 0.99 0.07
    2*$A_{3,2k-1,\rho}$ State11.000.01 0.22 0.90
    2*$A_{3,2k-1,c}$State10.96 0.01 0.00 1.00
    State20.00 0.99 0.78 0.10 State20.04 0.99 1.00 0.00
    2*$A_{3,2k,\rho}$ State11.000.01 0.25 0.91
    2*$A_{3,2k,c}$State10.08 0.66 0.00 0.99
    State20.00 0.99 0.75 0.09 State20.92 0.34 1.00 0.01
    2*$A_{3,2k+3,\rho}$ State10.990.00 0.01 0.39
    2*$A_{3,2k+3,c}$State10.01 0.95 0.07 0.80
    State20.01 1.00 0.99 0.61 State20.99 0.05 0.93 0.20
    2*$A_{3,2k+4,\rho}$ State10.990.00 0.01 0.34
    2*$A_{3,2k+4,c}$State10.00 0.99 0.02 1.00
    State20.01 1.00 0.99 0.66 State21.00 0.01 0.98 0.00
    $\sigma_{1,k,\rho}$12.69 6.14 12.06 5.95 $\sigma_{1,k,c}$4.23 1.81 4.28 1.67
    $\sigma_{2,2k-1,\rho}$9.08 4.82 9.56 4.21$\sigma_{2,2k-1,c}$4.63 2.55 3.70 2.92
    $\sigma_{2,k,\rho}$9.10 4.90 8.90 4.09 $\sigma_{2,k,c}$4.49 2.57 3.87 2.90
    $\sigma_{3,2k-1,\rho}$5.74 2.30 4.86 2.22$\sigma_{3,2k-1,c}$4.53 2.03 5.44 1.92
    $\sigma_{3,2k,\rho}$5.74 2.30 4.92 2.23 $\sigma_{3,2k,c}$4.75 2.24 5.33 2.06
    $\sigma_{3,2k+3,\rho}$5.63 2.26 6.48 2.53$\sigma_{3,2k+3,c}$5.29 2.23 4.56 2.10
    $\sigma_{3,2k+4,\rho}$5.63 2.23 6.93 2.62$\sigma_{3,2k+4,c}$5.32 2.06 4.57 1.88
  • Table 3   The PSNR value ofdifferent denoising models
    ImageNoise PSNR (dB)
    level BM3D MSDCT BLS-GSMNSCT-HMT NSST-HMT Proposed
    6*Test-image1 25 26.49 25.94 26.38 25.64 25.59 27.07
    30 25.77 25.20 25.63 25.30 25.24 26.63
    35 25.19 24.61 25.03 24.94 24.89 26.32
    40 24.65 24.13 24.53 24.56 24.58 25.71
    45 24.22 23.72 24.11 24.17 24.21 25.16
    50 23.89 23.38 23.74 23.78 23.82 24.54
    6*Test-image2 25 25.59 24.93 25.48 24.72 24.70 26.15
    30 24.95 24.29 24.81 24.49 24.50 25.75
    35 24.47 23.83 24.31 24.24 24.26 25.30
    40 24.06 23.48 23.92 23.94 24.00 24.72
    45 23.68 23.20 23.61 23.61 23.65 24.41
    50 23.44 22.97 23.34 23.39 23.46 24.07
    6*Test-image3 25 24.77 24.12 24.37 23.21 23.25 24.92
    30 23.84 23.13 23.45 22.89 22.86 24.39
    35 23.08 22.35 22.74 22.54 22.43 23.83
    40 22.37 21.74 22.17 22.18 21.97 22.78
    45 21.95 21.25 21.70 21.81 21.72 22.47
    50 21.57 20.86 21.31 21.44 21.35 22.04
    6*Bridge 25 26.23 25.69 26.03 24.53 24.67 26.50
    30 25.46 24.92 25.25 24.24 24.38 26.07
    35 24.86 24.33 24.64 23.91 24.03 25.48
    40 24.30 23.86 24.15 23.56 23.68 25.97
    45 23.90 23.48 23.93 23.20 23.29 24.53
    50 23.59 23.15 23.38 22.83 22.94 24.08
    6*Boat 25 29.91 29.15 29.27 26.97 27.31 29.96
    30 29.12 28.29 28.43 26.52 26.81 29.42
    35 28.43 27.58 27.72 26.03 26.23 28.75
    40 27.74 26.99 27.12 25.53 25.60 28.32
    45 27.24 26.49 26.60 25.03 25.24 27.72
    50 26.78 26.05 26.15 24.53 24.60 26.94
    6*Hill 25 29.85 29.20 29.27 27.60 27.43 29.71
    30 29.16 28.51 28.55 27.33 27.18 29.34
    35 28.56 27.94 27.96 27.02 26.91 28.88
    40 27.98 27.46 27.47 26.68 26.62 28.22
    45 27.57 27.05 27.05 26.31 26.26 27.89
    50 27.18 26.69 25.92 25.92 25.94 27.48

Copyright 2019 Science China Press Co., Ltd. 《中国科学》杂志社有限责任公司 版权所有

京ICP备18024590号-1