asp網(wǎng)站圖片不顯示附近的教育培訓(xùn)機(jī)構(gòu)有哪些
超分辨率(Super-resolution)概念理解:
wiki:超分辨率成像(SR-imaging)是提高成像系統(tǒng)分辨率的一類技術(shù)。光學(xué)SR技術(shù)可以超越系統(tǒng)的衍射極限,而幾何SR則可以提高數(shù)字成像傳感器的分辨率。超分辨率成像技術(shù)廣泛應(yīng)用于圖像處理和超分辨顯微術(shù)中。
形象理解:
給定圖像的像素越高,表示圖像質(zhì)量越接近于原始圖像。如果把低像素的圖片放大到一定程度,圖片會(huì)變得非常模糊,類似于馬賽克的情況。效果如下圖所示:
如果想讓上述圖像變得清新,這是需要的技術(shù)就是超分辨率重構(gòu)了。
超分辨率重構(gòu)現(xiàn)有方法:
在知乎上有看到一個(gè)帖子:圖像超分辨率重構(gòu)技術(shù)還有什么可研究嗎?以本人的理解,任何一項(xiàng)技術(shù)都有研究和提高的余地,技術(shù)進(jìn)步不就是這樣一點(diǎn)點(diǎn)來的嗎?總結(jié)一下現(xiàn)有的SR技術(shù)方法,并附找到的項(xiàng)目鏈接:
(1)稀疏編碼方法(Sparse Coding)
- Image super-resolution as sparse representation of raw image patches (CVPR2008)
- 基于原始圖像塊稀疏表示的圖像超分辨率
- Image super-resolution via sparse representation (TIP2010)
- Coupled dictionary training for image super-resolution (TIP2011)
- Single Image Super-Resolution from Transformed Self-Exemplars (CVPR2015)
- Naive Bayes Super-Resolution Forest (ICCV2015)
(4)基于金字塔算法
- http://vllab.ucmerced.edu/wlai24/LapSRN/
(5)深度學(xué)習(xí)方法(近幾年文章很多啊)
- Image Super-Resolution Using Deep Convolutional Networks (ECCV2014)
- Deep Networks for Image Super-Resolution with Sparse Prior (ICCV2015)
- Robust Single Image Super-Resolution via Deep Networks with Sparse Prior (TIP2016)
- Accurate Image Super-Resolution Using Very Deep Convolutional Networks (CVPR2016)
- Deeply-Recursive Convolutional Network for Image Super-Resolution (CVPR2016)
- Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network (CVPR2016)
- Deep Laplacian Pyramid Networks for Fast and Accurate Super-Resolution (CVPR 2017),
- Enhanced Deep Residual Networks for Single Image Super-Resolution (Winner of NTIRE2017 Super-Resolution Challenge)
關(guān)于深度學(xué)習(xí)在超分辨率重建中的應(yīng)用:https://zhuanlan.zhihu.com/p/25532538?utm_medium=social&utm_source=weibo
給出了幾種實(shí)現(xiàn)方法及介紹,github里面相應(yīng)的項(xiàng)目實(shí)現(xiàn)。另外還發(fā)現(xiàn)一篇有點(diǎn)尺度的文章《用GAN去除(愛情)動(dòng)作片中的馬賽克和衣服》,感興趣的請參見這里。
(6)Perceptual Loss and GAN(損失函數(shù)上改進(jìn))
- Perceptual Losses for Real-Time Style Transfer and Super-Resolution (ECCV2016)
- Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network (CVPR2017)
(7)Google基于哈希機(jī)制的實(shí)現(xiàn)
- 《RAISR: Rapid and Accurate Image Super Resolution》
分析:http://blog.csdn.net/jiangjieqazwsx/article/details/69055753
(8)視頻SR
- https://users.soe.ucsc.edu/~milanfar/software/superresolution.html
- Real-Time Video Super-Resolution with Spatio-Temporal Networks and Motion Compensation (CVPR2017)
小結(jié):SR使用稀疏編碼方法取得的方法已經(jīng)堪稱state-of-the-art級別,深度學(xué)習(xí)出現(xiàn)后又將效果進(jìn)一步提升。
11.15增補(bǔ):
今天看到一篇論文:《Super-Resolution From a Single Image 》(http://www.wisdom.weizmann.ac.il/~vision/SingleImageSR.html),http://cs.brown.edu/courses/csci1950-g/results/final/pachecoj/ ,
另外附幾個(gè)相關(guān)網(wǎng)頁:
https://people.mpi-inf.mpg.de/~kkim/supres/supres.htm
《Example-Based-Super-Resolution-Freeman》
18.1.3增補(bǔ):
神經(jīng)網(wǎng)絡(luò)實(shí)現(xiàn):
(1)《Accelerating the Super-Resolution Convolutional Neural Network》,使用matlab的實(shí)現(xiàn)。
(2)《Pixel Recursive Super Resolution》,項(xiàng)目實(shí)現(xiàn)鏈接。
參考:
https://en.wikipedia.org/wiki/Super-resolution_imaging
https://www.zhihu.com/question/38637977
https://github.com/huangzehao/Super-Resolution.Benckmark
https://zhuanlan.zhihu.com/p/25532538