基于掩蓋效應(yīng)和梯度信息的無參考噪聲圖像質(zhì)量評(píng)價(jià)改進(jìn)算法
doi: 10.11999/JEIT180195 cstr: 32379.14.JEIT180195
-
重慶大學(xué)生物工程學(xué)院 ??重慶 ??400044
Improved No-reference Noisy Image Quality Assessment Based on Masking Effect and Gradient Information
-
Institute of Bioengineering, Chongqing University, Chongqing 400044, China
-
摘要:
針對(duì)目前大多數(shù)噪聲圖像質(zhì)量評(píng)價(jià)算法借助域變換或機(jī)器學(xué)習(xí)所帶來的運(yùn)算量大、訓(xùn)練過程繁復(fù)等弊端，以及依賴人工設(shè)置固定閾值存在普適性不佳的問題，該文改進(jìn)了一種基于掩蓋效應(yīng)的空域噪聲圖像質(zhì)量評(píng)價(jià)算法。首先依據(jù)Hosaka原理提出層遞進(jìn)的分塊規(guī)則，將圖像分成與其內(nèi)容頻率分布高低相符的不同尺寸的子塊并賦予相應(yīng)的掩蓋權(quán)值；然后通過提取像素點(diǎn)梯度信息，經(jīng)兩步檢噪實(shí)現(xiàn)子塊噪點(diǎn)甄別；再使用掩蓋權(quán)值對(duì)子塊噪聲污染指標(biāo)加權(quán)得到初步質(zhì)量評(píng)價(jià)結(jié)果；最終修正和歸一化后為整圖質(zhì)量評(píng)價(jià)結(jié)果——改進(jìn)的無參考峰值信噪比(MNRPSNR)。應(yīng)用該算法在LIVE和TID2008圖像質(zhì)量評(píng)價(jià)數(shù)據(jù)庫上對(duì)多種噪聲類型圖像進(jìn)行實(shí)驗(yàn)，結(jié)果顯示其較目前主流評(píng)價(jià)算法保有很強(qiáng)競爭力，對(duì)傳統(tǒng)算法改進(jìn)效果顯著，與人眼主觀感受一致性高，普適于多種噪聲類型。
-
關(guān)鍵詞:
- 無參考圖像質(zhì)量評(píng)價(jià) /
- 掩蓋效應(yīng) /
- 噪聲檢測 /
- 梯度信息
Abstract:Heavy computational burden, or complex training procedure and poor universality caused by the manual setting of the fixed thresholds are the main issues associated with most of the noise image quality evaluation algorithms using domain transformation or machine learning. As an attempt for solution, an improved spatial noisy image quality evaluation algorithm based on the masking effect is presented. Firstly, according to the layer-layer progressive rule based on Hosaka principle, an image is divided into sub-blocks with different sizes that match the frequency distribution of its content, and a masking weight is assigned to each sub-block correspondingly. Then the noise in the image is detected through the pixel gradient information extraction, via a two-step strategy. Following that, the preliminary evaluation value is obtained by using the masking weights to weight the noise pollution index of all the sub-blocks. Finally, the correction and normalization are carried out to generate the whole image quality evaluation parameter——i.e. Modified No-Reference Peak Signal to Noise Ratio (MNRPSNR). Such an algorithm is tested on LIVE and TID2008 image quality assessment database, covering a variety of noise types. The results indicate that compared with the current mainstream evaluation algorithms, it has strong competitiveness, and also has the significant effects in improving the traditional algorithm. Moreover, the high degree of consistency to the human subjective feelings and the applicability to multiple noise types are well demonstrated.
-
表 1 數(shù)據(jù)庫信息及實(shí)驗(yàn)使用子集
數(shù)據(jù)庫 國家/機(jī)構(gòu) 參考圖像數(shù)量 主觀評(píng)價(jià)指標(biāo) 所選失真類型 損傷層級(jí) LIVE 美國/德克薩斯州立大學(xué) 29 DMOS 白噪聲(WN) 6 TID2008 烏克蘭/國家航空航天大學(xué)
意大利/羅馬大學(xué)
芬蘭/坦佩雷理工大學(xué)25 MOS 加性高斯噪聲(AGN) 5 顏色通道加性噪聲(ANCC) 空間相關(guān)噪聲(SCN) 掩蔽噪聲(MN) 高頻噪聲(HFN) 脈沖噪聲(IMN) 
下載: 導(dǎo)出CSV
表 2 改進(jìn)算法與不同檢噪閾值下傳統(tǒng)NRPSNR算法對(duì)monarch圖像組評(píng)價(jià)結(jié)果
DMOS NRPSNR MNRPSNR Nth=10 Nth=50 Nth=100 圖2(a1) 0.000000 58.35138 69.74312 79.08684 90.0779 圖2(a2) 23.94275 50.45518 69.31391 79.51719 77.9929 圖2(a3) 28.44905 47.50428 69.01371 79.70202 76.7756 圖2(a4) 41.16959 39.03095 49.28878 65.95135 68.3129 圖2(a5) 49.08675 36.47912 43.03078 52.72840 65.3847 圖2(a6) 65.73029 33.06348 36.52051 41.17893 60.7793
下載: 導(dǎo)出CSV
表 3 MNRPSNR與相關(guān)算法特征及在LIVE數(shù)據(jù)庫測試性能指標(biāo)
算法名稱 是否有參考圖像 是否需要訓(xùn)練 是否需要域變換 性能指標(biāo) PLCC SROCC RMSE PSNR 是 否 否 0.9050 0.9010 8.4500 SSIM 是 否 否 0.9700 0.9690 3.9540 BIQI 否 是 小波 0.9538 0.9510 8.4094 LBIQ 否 是 小波 0.9761 0.9700 7.9100 DIIVINE 否 是 小波 0.9880 0.9840 4.3100 BLIINDS 否 是 離散余弦 0.9140 0.8900 11.2700 BLIINDS-II 否 是 離散余弦 0.9799 0.9691 N/A NIQE 否 是 否 0.9773 0.9662 N/A BRISQUE 否 是 否 0.9851 0.9786 N/A NRPSNR 否 否 否 0.8681 0.8900 10.9133 MNRPSNR 否 否 否 0.9745 0.9813 4.9369
下載: 導(dǎo)出CSV
表 4 TID2008數(shù)據(jù)庫測試PLCC指標(biāo)比對(duì)
VSNR IFC NQM UQI NRPSNR MNRPSNR AGN 0.7513 0.6147 0.7397 0.5407 0.6467 0.7922 ANCC 0.7489 0.5628 0.6935 0.4930 0.0402 0.7291 SCN 0.7700 0.6567 0.7757 0.5589 0.1624 0.5808 MN 0.7799 0.7309 0.7575 0.7515 0.7903 0.5164 HFN 0.8861 0.7199 0.9134 0.7059 0.9283 0.9005 IMN 0.6244 0.4950 0.7492 0.4829 0.6403 0.8214
下載: 導(dǎo)出CSV
表 5 TID2008數(shù)據(jù)庫測試SROCC指標(biāo)比對(duì)
VSNR IFC NQM UQI NRPSNR MNRPSNR AGN 0.7745 0.6204 0.7592 0.5335 0.6276 0.7900 ANCC 0.7725 0.5921 0.7200 0.4798 0.0869 0.7115 SCN 0.7860 0.6403 0.7910 0.5472 0.0491 0.5786 MN 0.7555 0.7374 0.7624 0.7292 0.8018 0.5214 HFN 0.8870 0.7488 0.8952 0.6863 0.9039 0.8852 IMN 0.6460 0.5378 0.7666 0.4951 0.6495 0.8300
下載: 導(dǎo)出CSV
表 6 TID2008數(shù)據(jù)庫測試RMSE指標(biāo)比對(duì)
VSNR IFC NQM UQI NRPSNR MNRPSNR AGN 0.4005 0.4783 0.4131 0.5112 0.4682 0.3746 ANCC 0.3646 0.4448 0.3942 0.4844 0.5594 0.3832 SCN 0.3878 0.4655 0.3924 0.5055 0.6134 0.5060 MN 0.3745 0.3844 0.3890 0.3955 0.3673 0.5133 HFN 0.4259 0.6069 0.3691 0.6671 0.3563 0.4161 IMN 0.4022 0.4366 0.3402 0.4483 0.3933 0.2948
下載: 導(dǎo)出CSV
表 7 MNRPSNR與相關(guān)算法在LIVE數(shù)據(jù)庫上運(yùn)行時(shí)間(s)
算法名稱 DIIVINE BLIINDS-II NRPSNR MNRPSNR 平均單幅耗時(shí) 149 70 3.45 10.10
下載: 導(dǎo)出CSV
-
HADIZADEH H and VAN BAJIC I. Full-reference objective quality assessment of tone-mapped images[J]. IEEE Transactions on Multimedia, 2018, 20(2): 392–404. doi: 10.1109/TMM.2017.2740023 MA Jian, AN Ping, SHEN Liquan, et al. Reduced-reference stereoscopic image quality assessment using natural scene statistics and structural degradation[J]. IEEE Access, 2018, 6: 2768–2780. doi: 10.1109/ACCESS.2017.2785282 FANG Yuming, YAN Jiebin, LI Leida, et al. No reference quality assessment for screen content images with both local and global feature representation[J]. IEEE Transactions on Image Processing, 2018, 27(4): 1600–1610. doi: 10.1109/TIP.2017.2781307 WANG Zhou, BOVIK A C, SHEIKH H R, et al. Image quality assessment: from error visibility to structural similarity[J]. IEEE Transactions on Image Processing, 2004, 13(4): 600–612. doi: 10.1109/TIP.2003.819861 YE Peng and DAVID D. No-reference image quality assessment based on visual codebook[C]. IEEE International Conference on Image Processing, Brussels, Belgium 2011: 11–14. WAN Wenfei, WU Jinjian, XIE Xuemei, et al. A novel just noticeable difference model via orientation regularity in DCT domain[J]. IEEE Access, 2017, 5: 22953–22964. doi: 10.1109/ACCESS.2017.2699858 MA Lin, WANG Xu, LIU Qiong, et al. Reorganized DCT-based image representation for reduced reference stereoscopic image quality assessment[J]. Neurocomputing, 2016, 215: 21–31. doi: 10.1016/j.neucom.2015.06.116 SAAD M A, BOVIK A C, and CHARRIER C. A DCT statistics-based blind image quality index[J]. IEEE Signal Processing Letters, 2010, 17(6): 583–586. doi: 10.1109/LSP.2010.2045550 SAAD M A, BOVIK A C, and CHARRIER C. Blind image quality assessment: A natural scene statistics approach in the DCT domain[J]. IEEE Transactions on Image Processing, 2012, 22(8): 3339–3352. doi: 10.1109/TIP.2012.2191563 CHANDLER D M and HEMAMI S S. VSNR: A wavelet-based visual signal-to-noise ratio for natural images[J]. IEEE Transactions on Image Processing, 2007, 16(9): 2284–2298. doi: 10.1109/TIP.2007.901820 QIN Ming, Lü Xiaoxin, CHEN Xiaohui, et al. Hybrid NSS features for no-reference image quality assessment[J]. IET Image Processing, 2017, 11(6): 443–449. doi: 10.1049/iet-ipr.2016.0411 LI Leida, YAN Ya, LU Zhaolin, et al. No-reference quality assessment of deblurred images based on natural scene statistics[J]. IEEE Access, 2017, 5: 2163–2171. doi: 10.1109/ACCESS.2017.2661858 YANG Guangyi, LIAO Yue, ZHANG Qingyi, et al. No-reference quality assessment of noise-distorted images based on frequency mapping[J]. IEEE Access, 2017, 5: 23146–23156. doi: 10.1109/ACCESS.2017.2764126 ABDEL-HAMID L, EI-RAFEI A, and MICHELSON G. No-reference quality index for color retinal images[J]. Computers in Biology and Medicine, 2017, 90: 68–75. doi: 10.1016/j.compbiomed.2017.09.012 MA Kede, LIU Wentao, ZHANG Kai, et al. End-to-end blind image quality assessment using deep neural networks[J]. IEEE Transactions on Image Processing, 2018, 27(3): 1202–1213. doi: 10.1109/TIP.2017.2774045 LIU Tsungjung and LIU Kuanhsien. No-reference image quality assessment by wide-perceptual-domain scorer ensemble method[J]. IEEE Transaction on Image Processing, 2018, 27(3): 1138–1151. doi: 10.1109/TIP.2017.2771422 MITTAL A, MOORTHY A K, and BOVIK A C. No-reference image quality assessment in the spatial domain[J]. IEEE Transactions on Image Processing, 2012, 21(12): 4659–4708. doi: 10.1109/TIP.2012.2214050 MOORTHY A K and BOVIK A C. Blind image quality assessment: From natural scene statistics to perceptual quality[J]. IEEE Transactions on Image Processing, 2011, 20(12): 3350–3364. doi: 10.1109/TIP.2011.2147325 LIU Anmin and LIN Weisi. Image quality assessment based on gradient similarity[J]. IEEE Transactions on Image Processing, 2012, 21(4): 1500–1512. doi: 10.1109/TIP.2011.2175935 王正友, 肖文. 基于掩蓋效應(yīng)的無參考數(shù)字圖像質(zhì)量評(píng)價(jià)[J]. 計(jì)算機(jī)應(yīng)用, 2006, 26(12): 2838–2840.WANG Zhengyou and XIAO Wen. No-reference digital image quality evaluation based on perceptual masking[J]. Computer Applications, 2006, 26(12): 2838–2840. 徐海勇, 郁梅, 駱挺, 等. 基于非負(fù)矩陣分解的彩色圖像質(zhì)量評(píng)價(jià)方法[J]. 電子與信息學(xué)報(bào), 2016, 38(3): 578–585. doi: 10.11999/JEIT150610XU Haiyong, YU Mei, LUO Ting, et al. A color image quality assessment method based on non-negative matrix factorization[J]. Journal of Electronics &Information Technology, 2016, 38(3): 578–585. doi: 10.11999/JEIT150610 蔣平, 張建州. 基于局部最大梯度的無參考圖像質(zhì)量評(píng)價(jià)[J]. 電子與信息學(xué)報(bào), 2015, 37(11): 2587–2593. doi: 10.11999/JEIT141447JIANG Ping and ZHANG Jianzhou. No-reference image quality assessment based on local maximum gradient[J]. Journal of Electronics &Information Technology, 2015, 37(11): 2587–2593. doi: 10.11999/JEIT141447 SHEIKH H R, WANG Zhou, and CORMACK L. LIVE image quality assessment database, release 2 [EB/OL]. http://live.ece.utexass.edu/research/quality, 2005. PONOMARENKO N. Tampere image database 2008 TID2008, version 1.0[EB/OL]. http://www.ponomarenko.info/index.html, 2009. Final Report from the Video Quality Experts Group on the Validation of Objective Models of Video Quality Assessment, Phase II VQEG [OL]. http://www.vqeg.org/, 2003. MOORTHY A K and BOVIK A C. A two-step framework for constructing blind image quality indices[J]. IEEE Signal Processing Letters, 2010, 17(5): 513–516. doi: 10.1109/LSP.2010.2043888 TANG Huixuan, JOSHI N, and KAPOOR A. Learning a blind measure of perceptual image quality[C]. IEEE Conference on Computer Vision and Pattern Recognition, Colorado, USA, 2011: 305–312. MITTAL A, SOUNDARARAJAN R, and BOVIK A C. Making a " completely blind” image quality analyzer[J]. IEEE Signal Processing Letters, 2013, 20(3): 209–212. doi: 10.1109/LSP.2012.2227726 SHEIKH H R, BOVIK A C, and DE VECIANA G. An information fidelity criterion for image quality assessment using natural scene statistics[J]. IEEE Transactions on Image Processing, 2005, 14(12): 2117–2128. doi: 10.1109/TIP.2005.859389 DAMERA-VENKATA N, KITE T D, GEISLER W S, et al. Image quality assessment based on a degradation model[J]. IEEE Transactions on Image Processing, 2000, 9(4): 636–650. doi: 10.1109/83.841940 WANG Zhou and BOVIK A C. A universal image quality index[J]. IEEE Signal Processing Letters, 2002, 9(3): 81–84. doi: 10.1109/97.995823 -
計(jì)量
- 文章訪問數(shù): 1483
- HTML全文瀏覽量: 764
- PDF下載量: 80
- 被引次數(shù): 0
下載:
