E-mail Alert
RSS
-
摘要:
高动态范围成像通常需要利用多个不同曝光时间的低动态范围图像来合成高动态范围图像。在合成后图像噪声会进一步放大，可能导致最终的高动态范围图像视觉质量严重降低。针对合成图像需要保留低曝光图像中高亮区域的细节信息以及高曝光图像中低暗区域的细节信息，且图像噪声与亮度有关这一问题，本文提出一种基于亮度分区、噪声水平估计的高动态范围图像融合过程中的噪声抑制算法。首先，根据图像的亮度信息，确定低动态范围图像的不同亮度区域；然后对图像不同亮度区域，利用重叠块估计噪声水平，根据得到的噪声水平，指导图像的稀疏去噪；最后，对处理后的低动态范围图像，采用融合方法合成高动态范围图像。实验结果表明，所提出的算法能有效地抑制图像噪声，合成的高动态范围图像具有更好的视觉质量。
Abstract:A high dynamic range image is generated by using multiple low dynamic range images with different exposure times. After the generation, the image noise will be further amplified, resulting in a severe degradation in the visual quality of the final high dynamic range image. In view of the problem that the generated image needs to retain the detail information of the high-lighted area in the low-exposure image and the detail information of the low-dark area in the high-exposure image and the image noise is related to the luminance, a noise suppression algorithm based on luminance partitioning and noise level estimation in the process of high dynamic range image fusion is proposed in this paper. Firstly, according to the luminance information of the image, different luminance regions of the low dynamic range image are determined. And then the overlapped blocks are used to estimate the noise level of the different luminance regions of the image, so as to guide the sparse denoising of the image. Finally, a high dynamic range image is generated by the processed low dynamic range images. The experimental results show that the proposed algorithm can effectively suppress noise, and the generated high dynamic range image has better visual quality.
-
Overview: High dynamic range (HDR) images can accurately represent the dynamic range of real scenes and are therefore receiving widespread attention. At present, the methods for obtaining HDR images are mainly divided into two categories. The first type is to map the original low dynamic range (LDR) images to the irradiation field by estimating the camera response function, and perform fusion in the irradiation field to obtain a wide dynamic range image. The second is to fuse a multiple exposure image sequence directly in the pixel domain to produce an LDR image with HDR effect. However, these fusion methods do not consider the effect of noise in the images. When shooting a multi-exposure image sequence in an actual low light environment, the camera sets high sensitivity to prevent images blurring, which generates noise. After the fusion, the image noise will be further amplified, resulting in a severe degradation in the visual quality of the final HDR image. To solve this problem, according to the noise characteristics of multi-exposure images, this paper proposes a noise suppression algorithm in the high dynamic range image fusion process based on luminance partition and noise level estimation. Firstly, according to the luminance information of the images, each LDR image is decomposed into three areas of low-luminance, middle-luminance, and high-luminance. Noise levels are estimated by using overlapping blocks in three luminance areas. Since the noise mainly exists in the dark areas of the image, therefore, for the low-luminance area, " winner-take-all" strategy is used to process the overlapping blocks to obtain the noise level. For the middle-luminance and high-luminance areas, an average strategy is used to obtain the noise level. Then the noise level is used to guide the sparse denoising of the image. The denoised images of the three different luminance regions are combined into a single complete LDR image. Finally, the processed multiple exposure LDR images are fused into a single high-quality HDR image. Experimental results show that the proposed algorithm not only can effectively suppress the noise of the image, but also processes the image according to the noise level without blurring the details of the image and preserves the texture information of the original image well. In addition, using the existing objective metrics to evaluate the fusion images, the results show that the image quality obtained by this method is higher, and image detail information is retained while suppressing image noise.
-
-
图 1 高动态范围图像融合过程中的噪声抑制算法框图
Figure 1. Framework of noise suppression algorithm in the process of high dynamic range image
图 2 亮度分区图。(a)不同曝光的LDR图像；(b)低亮度图像；(c)中间亮度图像；(d)高亮度图像
Figure 2. Images of luminance partition. (a) Different exposure LDR images; (b) Low luminance images; (c) Middle luminance images; (d) High luminance images
图 3 不同曝光图像。(a)低曝光图像；(b)中间曝光图像；(c)高曝光图像
Figure 3. Different exposure images. (a) Low exposure image; (b) Middle exposure image; (c) High exposure image
图 4 噪声水平指导去噪对比实验。(a)多曝光图像序列；(b)噪声水平估计；(c)无噪声水平估计
Figure 4. Comparisonexperiment of noise level guide denoising. (a) Multi-exposure image sequence; (b) Noise level estimation; (c) No noise level estimation
表 1 噪声水平估计
Table 1. Results of noise level estimation
图像 区间 真实噪声水平 估计噪声水平 低曝光图像 低亮度区间 7.65 7.25 中间亮度区间 5.1 4.73 高亮度区间 0.00 0.00 中间曝光图像 低亮度区间 7.65 7.26 中间亮度区间 5.1 5.19 高亮度区间 2.25 2.31 高曝光图像 低亮度区间 0.00 0.00 中间亮度区间 5.1 4.87 高亮度区间 2.25 2.48 
下载: 导出CSV
表 2 融合图像Qabf值对比
Table 2. Qabf value comparison of fusion images
融合方法 噪声抑制 Lamp Cafe Window Arch Debevec[4] 本文算法 0.4646 0.5158 0.4136 / 不处理 0.3657 0.4957 0.3826 / Li[9] 本文算法 0.5580 0.6868 0.6646 / 不处理 0.4601 0.6294 0.5993 / Liu[10] 本文算法 0.5821 0.6868 0.6545 / 不处理 0.4845 0.6357 0.6008 / Mertens[11] 本文算法 0.5803 0.6587 0.6514 / 不处理 0.4842 0.6123 0.5914 / Oh[2] 本文算法 / / / 0.4835 不处理 / / / 0.4244 Sen[3] 本文算法 / / / 0.5031 不处理 / / / 0.4524
下载: 导出CSV
-
[1] Chiang J C, Kao P H, Chen Y S, et al. High-dynamic-range image generation and coding for multi-exposure multi-view images[J]. Circuits, Systems, and Signal Processing, 2017, 36(7): 2786–2814. doi: 10.1007/s00034-016-0437-x
[2] Oh T H, Lee J Y, Tai Y W, et al. Robust high dynamic range imaging by rank minimization[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(6): 1219–1232. doi: 10.1109/TPAMI.2014.2361338
[3] Sen P, Kalantari N K, Yaesoubi M, et al. Robust patch–based hdr reconstruction of dynamic scenes[J]. ACM Transactions on Graphics, 2012, 31(6): 1–11. doi: 10.1145/2366145.2366222.
[4] Debevec P E, Malik J. Recovering high dynamic range radiance maps from photographs[C]//Proceedings of the 24th Annual Conference on Computer Graphics and Interactive Techniques, New York: ACM/Press, 1997: 369–378.
[5] Ma K, Yeganeh H, Zeng K, et al. High dynamic range image compression by optimizing tone mapped image quality index[J]. IEEE Transactions on Image Processing, 2015, 24(10): 3086–3097. doi: 10.1109/TIP.2015.2436340
[6] Li Y Z, Sharan L, Adelson E H. Compressing and companding high dynamic range images with subband architectures[J]. ACM Transactions on Graphics, 2005, 24(3): 836–844. doi: 10.1145/1073204
[7] 芦碧波, 李玉静, 王玉琨, 等.亮度分区自适应对数色调映射算法[J].计算机应用研究, 2018, 35(9). http://www.arocmag.com/article/02-2018-09-046.html. http://www.arocmag.com/article/02-2018-09-046.html
Lu B B, Li Y J, Wang Y K, et al. Adaptive logarithmic tone mapping arithmetic with luminance regionalization[J]. Application Research of Computers, 2018, 35(9). http://www.arocmag.com/article/02-2018-09-046.html. http://www.arocmag.com/article/02-2018-09-046.html
[8] Ma K, Li H, Yong H W, et al. Robust multi-exposure image fusion: a structural patch decomposition approach[J]. IEEE Transactions on Image Processing, 2017, 26(5): 2519–2532. doi: 10.1109/TIP.2017.2671921
[9] Li S T, Kang X D, Hu J W. Image fusion with guided filtering[J]. IEEE Transactions on Image Processing, 2013, 22(7): 2864–2875. doi: 10.1109/TIP.2013.2244222
[10] Liu Y, Wang Z F. Dense SIFT for ghost-free multi-exposure fusion[J]. Journal of Visual Communication and Image Representation, 2015, 31: 208–224. doi: 10.1016/j.jvcir.2015.06.021
[11] Mertens T, Kautz J, Van Reeth F. Exposure fusion: a simple and practical alternative to high dynamic range photography[J]. Computer Graphics Forum, 2009, 28(1): 161–171. doi: 10.1111/cgf.2009.28.issue-1
[12] OğuzAkyüz A, Reinhard E. Noise reduction in high dynamic range imaging[J]. Journal of Visual Communication and Image Representation, 2007, 18(5): 366–376. doi: 10.1016/j.jvcir.2007.04.001
[13] 孙德全, 张军, 李晓峰.一种去除HDR图像噪声的新方法[J].中国图象图形学报, 2010, 15(2): 229–235. doi: 10.11834/jig.20100207
Sun Q D, Zhang J, Li X F. A novel noise removal algorithm for HDR image[J]. Journal of Image and Graphics, 2010, 15(2): 229–235. doi: 10.11834/jig.20100207
[14] Min T H, Park R H, Chang S. Noise reduction in high dynamic range images[J]. Signal, Image and Video Processing, 2011, 5(3): 315–328. doi: 10.1007/s11760-010-0203-7
[15] Yao W, Li Z G, Rahardja S. Intensity mapping function based weighted frame averaging for high dynamic range imaging[C]// Proceedings of 2011 6th IEEE Conference on Industrial Electronics and Applications. Beijing, China: IEEE, 2011: 1574–1577.
[16] Yao W, Li Z G, Rahardja S. Noise reduction for differently exposed images[C]//Proceedings of 2012 IEEE International Conference on Acoustics, Speech and Signal Processing. Kyoto, Japan: IEEE, 2012: 917–920.
[17] Lee D K, Park R H, Chang S. Ghost and noise removal in exposure fusion for high dynamic range imaging[J]. International Journal of Computer Graphics & Animation, 2014, 4(4): 1–18. http://cn.bing.com/academic/profile?id=5994e7dc7dcdc3a9d335fa8042fc5e80&encoded=0&v=paper_preview&mkt=zh-cn
[18] Liu X H, Tanaka M, Okutomi M. Single-image noise level estimation for blind denoising[J]. IEEE Transactions on Image Processing, 2013, 22(12): 5226–5237. doi: 10.1109/TIP.2013.2283400
[19] Dong W S, Zhang L, Shi G M, et al. Nonlocally centralized sparse representation for image restoration[J]. IEEE Transactions on Image Processing, 2013, 22(4): 1620–1630. doi: 10.1109/TIP.2012.2235847
[20] Buades A, Coll B, Morel J M. A review of image denoising algorithms, with a new one[J]. Multiscale Modeling & Simulation, 2010, 4(2): 490–530. http://cn.bing.com/academic/profile?id=c86d0fdb889019a7519c4db4d7de59be&encoded=0&v=paper_preview&mkt=zh-cn
[21] Dabov K, Foi A, Katkovnik V, et al. Image denoising by sparse 3-D transform-domain collaborative filtering[J]. IEEE Transactions on Image Processing, 2007, 16(8): 2080–2095. http://cn.bing.com/academic/profile?id=1c4ea0c173496cbb6529eb749e748145&encoded=0&v=paper_preview&mkt=zh-cn
[22] Chen Y J, Pock T. Trainable nonlinear reaction diffusion: a flexible framework for fast and effective image restoration[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1256–1272. doi: 10.1109/TPAMI.2016.2596743
[23] Zhang K, Zuo W M, Chen Y J, et al. Beyond a gaussian denoiser: residual learning of deep CNN for image denoising[J]. IEEE Transactions on Image Processing, 2017, 26(7): 3142–3155. doi: 10.1109/TIP.2017.2662206
-
点击扫一扫
图(9)
表(2)
计量
- 文章访问数: 8232
- PDF下载数: 3561
- 施引文献: 0
