基于纹理分解的变换域JND模型及图像编码方法

doi:10.3969/j.issn.1000-436x.2014.06.024

摘要/Abstract

摘要：

为了提高变换域JND模型的精度，在计算对比度掩盖因子时只对纹理分量滤波并判断区域类型的方法避免了JND低估的问题。将改进的JND模型用于图像编码，考虑到辅助信息对编码效率的影响，把经过调整后的JND模型结合到量化过程中，能去除更多的视觉冗余并保持兼容性。仿真结果表明，纹理分解的方法提高了JND阈值，改进的编码方法在相似的视觉质量下能节省更多的码率并且不需要增加额外的比特开销，该编码思路也适用于视频编码。

关键词: 视觉特性, 最小可觉察误差, 纹理分解, 图像编码

Abstract:

In order to improve the accuracy of the just noticeable difference (JND) model in transform domain, an en-hanced JND model with a new method for contrast masking factor estimation was proposed. The image was decomposed and the textural image was used for an accurate block classification, thus the accurate JND in DCT domain was obtained. The improved JND model was applied on the perceptual image coding. Considering the compatibility and the auxiliary information which would affect the encoding efficiency, the JND model was adjusted to the quantization process and re-moved more visual redundancy. Experimental results show that the proposed algorithm can improve the JND threshold;compared with JPEG standard, the perceptual coding method can save more bit rate and does not need extra bit for auxil-iary information at the similar visual quality. The proposed algorithm is also applicable to the perceptual video coding.

Key words: visual perception, just noticeable difference(JND), textural decomposition, image coding

郑明魁,苏凯雄,王卫星,兰诚栋,杨秀芝. 基于纹理分解的变换域JND模型及图像编码方法[J]. 通信学报, 2014, 35(6): 185-191.

Ming-kui ZHENG,Kai-xiong SU,Wei-xing WANG,Cheng-dong LAN,Xiu-zhi YANG. Improved sub-band JND model with textural decomposition and its application in perceptual image coding[J]. Journal on Communications, 2014, 35(6): 185-191.

图/表 11

图1

图2

图3

图4

表1

表2

图5

表3

表4

图6

表5

参考文献 17

[1]	YANY X K , LIN W S , LU Z K , et al. Just noticeable distortion model and its applications in video coding[J]. Signal Processing: Image Communication, 2005,20(7): 662-680.
[2]	MATTEO N , FERNANDO P . Advanced H.264/AVC-based perceptual video: coding: architecture, tools, and assessment[J]. IEEE Transaction on Circuits and Systems for Video Technology, 2011,21(6): 766-782.
[3]	PHIBANG N , BEGHDADI A , LUONG M . Perceptual watermarking robust to JPEG compression attack[A]. Proceedings of the 5th Interna-tional Symposium on Communications Control and Signal Process-ing[C]. Rome, Italy, 2012.1-4.
[4]	HSU M C , WU G L , CHIEN S Y . Combination of SSIM and JND with content-transition classification for image quality assessment[A]. Vis-ual Communications and Image Processing (VCIP)[C]. San Diego, USA, 2012.1-6.
[5]	LIU A , LIN W S , PAUL M , et al. Just noticeable difference for image with decomposition model for separating edge and texture regions[J]. IEEE Transaction on Circuits and Systems for video Technology, 2010,20(11):1648-1652.
[6]	WU J J , QI F , SHI G M . Self-similarity based structural regularity for just noticeable difference estimation[J]. Journal of Visual Communication and Image Representation, 2012,23(6):845-852.
[7]	AHUMADA A J , PETERSON A . Luminance-model-based DCT quantization for color image compression[A]. Hum. Vision Visual Process. Digital Display III[C]. 1992.365-374.
[8]	WATSON A B . DCTune: a technique for visual optimization of DCT quantization matrices for individual images[A]. Proceedings of Soci-ety for Information Display Digest of Technical[C]. San Diego, USA, 1993.946-949.
[9]	ZHANG X H , LIN W S , XUE P . Improved estimation for just- no-ticeable visual distortion[J]. Signal Processing, 1993,84(4): 795-808.
[10]	JIA Y , LIN W S , KASSIM A A . Estimating just noticeable distortion for video[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2006,16(7): 820-829.
[11]	WEI Z Y , NGAN K N . Spatio-temporal just noticeable distortion profile for grey scale image video in DCT domain[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2009,19(3):337-346.
[12]	MA L , NGAN K N . Adaptive block-size transform based just-noticeable difference model for images/videos[J]. Signal Processing: Image Communication, 2011,26(3):162-174.
[13]	LUO Z Y , SONG L , ZHENG S B , et al. H.264/AVC perceptual opti-mization coding based on JND-directed coefficient suppression[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2013,23(6):935-948.
[14]	YIN W , GOLDFARB D , OSHER S . A comparison of three total varia-tion based texture extraction models[J]. Journal of Visual Communication and Image Representation, 2007,18(3):240-252.
[15]	WANG C Y , LEE M , CHANG L W . Designing JPEG quantization tables based on human visual system[J]. Signal Processing: Image Communication, 2001,16(5):501-506.
[16]	潘榕，刘昱，侯正信等. 基于局部 DCT 系数的图像压缩感知编码与重构[J]. 自动化学报， 2011,37(6): 674-681. PAN R , LIU Y , HOU Z X , et al. Image coding and reconstruction via compressed sensing based on partial DCT coefficients[J]. Acta Auto-matica Sinica, 2011,37(6): 674-681.
[17]	WANG Z , BOVIK A C , SHEIKH H R , et al. Image quality assessment:from error measurement to structural similarity[J]. IEEE Transactions on Image Process, 2004,13(4):600-612.

测试图像	PSNR对比/dB		主观质量对比	复杂度对比/s
测试图像	文献[11]模型	本文模型	主观质量对比	文献[11]模型本文模型
Lena	30.638	30.010	0.1	0.828	0.953
Barbara	30.749	30.165	?0.1	0.453	0.531
hedgebw	29.656	29.202	0	0.437	0.531
Lady	31.016	30.555	0.1	0.422	0.547
Camerama	30.136	29.827	0	0.406	0.516
Goldhill	30.668	30.042	?0.1	1.906	2.313
Peppers	31.106	30.551	?0.1	1.891	2.234
Baboon	29.083	28.832	0	1.937	2.485
Boat	30.546	30.175	0	1.859	2.297
Indian	30.301	29.754	0	1.922	2.5
平均	30.390	29.911	?0.01	1.206	1.491
增加值		0.479			0.285

主观评分	描述
?3	右图比左图质量差很多
?2		?1	右图比左图质量只差一点
0	0
1	右图比左图质量只好一点
2	右图比左图质量好
3	右图比左图质量好很多

测试图像	算法码率/(比特·像素^?1)			码率减少量
测试图像	JPEG编码	文献[15]	本文方法	JPEG相比	文献[15]相比
Lena	0.953	0.911	0.823	13.6%	9.7%
Barbara	0.934	0.907	0.804	13.9%	11.4%
hedgebw	1.270	1.245	1.095	13.8%	12.0%
Lady	0.718	0.625	0.620	13.6%	0.8%
Camerama	0.876	0.890	0.751	14.3%	15.6%
Goldhill	0.835	0.781	0.698	16.4%	10.6%
Peppers	0.654	0.556	0.546	16.5%	1.8%
Baboon	1.390	1.507	1.191	14.3%	21.0%
Boat	0.744	0.707	0.637	14.4%	9.9%
Indian	0.944	0.926	0.793	16.0%	14.4%
平均				14.7%	10.7%

	客观质量SSIM对比			主观质量对比
测试图像
测试图像	JPEG编码	文献[15]	本文方法	JPEG编码	文献[15]
Lena	0.922	0.920	0.910	0.1	0.1
Barbara	0.896	0.913	0.891	0.1	0.1
hedgebw	0.924	0.923	0.909	0	0
Lady	0.935	0.925	0.926	0.1	0.2
Camerama	0.911	0.920	0.900	0.2	0.2
Goldhill	0.894	0.892	0.877	0	-0.1
Peppers	0.885	0.874	0.875	0	-0.1
Baboon	0.879	0.902	0.862	?0.1	0
Boat	0.930	0.929	0.920	?0.1	0
Indian	0.890	0.887	0.875	?0.1	0.1
平均	0.907	0.908	0.894	0.02	0.03

		算法复杂度/s
测试图像
测试图像
	JPEG编码	文献[15]	本文方法
Lena	1.719	1.954	2.797
Barbara	1.609	1.828	2.234
hedgebw	2.235	2.469	2.718
Lady	1.203	1.25	1.876
Cameraman	1.484	1.703	2.078
Goldhill	9.47	9.595	11.251
Peppers	6.828	6.265	8.922
Baboon	19.188	22.829	18.845
Boat	7.875	8.376	9.907
Indian	10.782	11.203	12.156
平均	6.239	6.747	7.278
本文增加值	16.7%	7.9%