基于深度跳跃级联的图像超分辨率重建

袁昆鹏; 席志红

doi:doi:10.3788/AOS201939.0715003

光学学报, 2019, 39 (7): 0715003, 网络出版: 2019-07-16

基于深度跳跃级联的图像超分辨率重建下载： 1494次

Image Super Resolution Based on Depth Jumping Cascade

袁昆鹏 ^*席志红

作者单位

哈尔滨工程大学信息与通信工程学院, 黑龙江哈尔滨 150001

图 & 表

图 1. SRCNN网络结构图

Fig. 1. Diagram of SRCNN network structure

下载图片查看原文

图 2. VDSR网络结构图

Fig. 2. Diagram of VDSR network structure

下载图片查看原文

图 3. 本文算法网络结构图

Fig. 3. Network structural diagram of proposed algorithm

下载图片查看原文

图 4. 跳跃级联块

Fig. 4. Jumping cascade block

下载图片查看原文

图 5. 图像butterfly放大4倍后的可视化结果。(a)低层特征;(b)跳跃级联特征;(c)重建特征

Fig. 5. Visualization results of picture butterfly after 4-time magnification. (a) Low-level feature; (b) hopping cascade feature; (c) reconstructing feature

下载图片查看原文

图 6. Set5测试集下不同层数PSNR均值随迭代次数增长的曲线

Fig. 6. Relationship between average PSNR of different layers with number of iterations under Set5 test set

下载图片查看原文

图 7. Set5测试集下不同网络结构的迭代次数与PSNR均值的关系曲线

Fig. 7. Relationship between iteration times and average PSNR of different network structures under Set5 test set

下载图片查看原文

图 8. Set5测试集下不同激活函数PSNR均值随迭代次数的变化曲线

Fig. 8. Relationship between average PSNR of different activation functions with number of iterations under Set5 test set

下载图片查看原文

图 9. Set5测试集下不同滤波器数目条件下的PSNR均值随迭代次数的变化曲线

Fig. 9. Relationship between average PSNR of different filter numbers with number of iterations under Set5 test set

下载图片查看原文

图 10. Set5测试集下不同方法的运行时间与PSNR均值的关系曲线

Fig. 10. Relationship between run time and average PSNR of different methods under Set5 test set

下载图片查看原文

图 11. Butterfly在不同算法下的效果对比图

Fig. 11. Comparison of butterfly images obtained by different algorithms

下载图片查看原文

图 12. PPT在不同算法下的效果对比图

Fig. 12. Comparison of PPT obtained by different algorithms

下载图片查看原文

图 13. Man在不同算法的对比图

Fig. 13. Comparison of man images obtained by different algorithms

下载图片查看原文

表 1在测试集Set5, Set14, BSD100下比较不同方法的PSNR均值

Table1. Average PSNR of test sets Set5, Set14, and BSD100 under different algorithms

Dataset	Scale	Bicubic	SRCNN^[12]	ESPCN^[14]	FSRCNN^[13]	VDSR^[15]	DCSR
	2	33.68	36.19	36.38	36.45	37.34	37.70
Set5	3	30.45	32.46	32.71	32.59	33.47	34.13
	4	28.46	30.15	30.29	30.42	30.78	31.87
	2	30.21	32.1	32.2	32.21	32.82	33.26
Set14	3	27.51	28.99	29.12	29.12	29.51	29.97
	4	25.98	27.23	27.17	27.43	27.62	28.27
	2	29.43	30.88	30.93	31.24	31.51	31.81
BSD100	3	27.08	28.06	28.16	28.25	28.43	28.79
	4	25.84	26.63	26.59	26.85	26.87	27.28

查看原文

表 2在测试集Set5, Set14, BSD100下比较不同方法的SSIM均值

Table2. Average SSIM of test sets Set5,Set14, and BSD100 under different algorithms

Dataset	Scale	Bicubic	SRCNN^[12]	ESPCN^[14]	FSRCNN^[13]	VDSR^[15]	DCSR
	2	0.9306	0.9551	0.9568	0.9567	0.9580	0.9636
Set5	3	0.8686	0.9110	0.9150	0.9122	0.9188	0.9321
	4	0.8102	0.8621	0.8629	0.8659	0.8750	0.8979
	2	0.8693	0.9576	0.9597	0.9634	0.9104	0.9644
Set14	3	0.7744	0.8836	0.8873	0.8923	0.8271	0.8981
	4	0.7023	0.8207	0.8227	0.8273	0.7592	0.8446
	2	0.8440	0.8801	0.8831	0.8867	0.8924	0.8964
BSD100	3	0.7401	0.7755	0.7811	0.7803	0.7928	0.7987
	4	0.6697	0.692	0.6943	0.7009	0.7186	0.7203

查看原文

袁昆鹏, 席志红. 基于深度跳跃级联的图像超分辨率重建[J]. 光学学报, 2019, 39(7): 0715003. Kunpeng Yuan, Zhihong Xi. Image Super Resolution Based on Depth Jumping Cascade[J]. Acta Optica Sinica, 2019, 39(7): 0715003.

基于深度跳跃级联的图像超分辨率重建下载： 1494次

图 1. SRCNN网络结构图

Fig. 1. Diagram of SRCNN network structure

图 2. VDSR网络结构图

Fig. 2. Diagram of VDSR network structure

图 3. 本文算法网络结构图

Fig. 3. Network structural diagram of proposed algorithm

图 4. 跳跃级联块

Fig. 4. Jumping cascade block

图 5. 图像butterfly放大4倍后的可视化结果。(a)低层特征;(b)跳跃级联特征;(c)重建特征

Fig. 5. Visualization results of picture butterfly after 4-time magnification. (a) Low-level feature; (b) hopping cascade feature; (c) reconstructing feature

图 6. Set5测试集下不同层数PSNR均值随迭代次数增长的曲线

Fig. 6. Relationship between average PSNR of different layers with number of iterations under Set5 test set

图 7. Set5测试集下不同网络结构的迭代次数与PSNR均值的关系曲线

Fig. 7. Relationship between iteration times and average PSNR of different network structures under Set5 test set

图 8. Set5测试集下不同激活函数PSNR均值随迭代次数的变化曲线

Fig. 8. Relationship between average PSNR of different activation functions with number of iterations under Set5 test set

图 9. Set5测试集下不同滤波器数目条件下的PSNR均值随迭代次数的变化曲线

Fig. 9. Relationship between average PSNR of different filter numbers with number of iterations under Set5 test set

图 10. Set5测试集下不同方法的运行时间与PSNR均值的关系曲线

Fig. 10. Relationship between run time and average PSNR of different methods under Set5 test set

图 11. Butterfly在不同算法下的效果对比图

Fig. 11. Comparison of butterfly images obtained by different algorithms

图 12. PPT在不同算法下的效果对比图

Fig. 12. Comparison of PPT obtained by different algorithms

图 13. Man在不同算法的对比图

Fig. 13. Comparison of man images obtained by different algorithms

表 1在测试集Set5, Set14, BSD100下比较不同方法的PSNR均值

Table1. Average PSNR of test sets Set5, Set14, and BSD100 under different algorithms

表 2在测试集Set5, Set14, BSD100下比较不同方法的SSIM均值

Table2. Average SSIM of test sets Set5,Set14, and BSD100 under different algorithms

关于本站 Cookie 的使用提示

全站搜索

基于深度跳跃级联的图像超分辨率重建 下载： 1494次

图 1. SRCNN网络结构图

Fig. 1. Diagram of SRCNN network structure

图 2. VDSR网络结构图

Fig. 2. Diagram of VDSR network structure

图 3. 本文算法网络结构图

Fig. 3. Network structural diagram of proposed algorithm

图 4. 跳跃级联块

Fig. 4. Jumping cascade block

图 5. 图像butterfly放大4倍后的可视化结果。(a)低层特征;(b)跳跃级联特征;(c)重建特征

Fig. 5. Visualization results of picture butterfly after 4-time magnification. (a) Low-level feature; (b) hopping cascade feature; (c) reconstructing feature

图 6. Set5测试集下不同层数PSNR均值随迭代次数增长的曲线

Fig. 6. Relationship between average PSNR of different layers with number of iterations under Set5 test set

图 7. Set5测试集下不同网络结构的迭代次数与PSNR均值的关系曲线

Fig. 7. Relationship between iteration times and average PSNR of different network structures under Set5 test set

图 8. Set5测试集下不同激活函数PSNR均值随迭代次数的变化曲线

Fig. 8. Relationship between average PSNR of different activation functions with number of iterations under Set5 test set

图 9. Set5测试集下不同滤波器数目条件下的PSNR均值随迭代次数的变化曲线

Fig. 9. Relationship between average PSNR of different filter numbers with number of iterations under Set5 test set

图 10. Set5测试集下不同方法的运行时间与PSNR均值的关系曲线

Fig. 10. Relationship between run time and average PSNR of different methods under Set5 test set

图 11. Butterfly在不同算法下的效果对比图

Fig. 11. Comparison of butterfly images obtained by different algorithms

图 12. PPT在不同算法下的效果对比图

Fig. 12. Comparison of PPT obtained by different algorithms

图 13. Man在不同算法的对比图

Fig. 13. Comparison of man images obtained by different algorithms

表 1在测试集Set5, Set14, BSD100下比较不同方法的PSNR均值

Table1. Average PSNR of test sets Set5, Set14, and BSD100 under different algorithms

表 2在测试集Set5, Set14, BSD100下比较不同方法的SSIM均值

Table2. Average SSIM of test sets Set5,Set14, and BSD100 under different algorithms

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索

基于深度跳跃级联的图像超分辨率重建下载： 1494次