This article presents near-optimal guarantees for accurate and robust image recovery from under-sampled noisy measurements using total variation minimization. In particular, we show that from O(slog(N)) nonadaptive linear measurements, an image can be reconstructed to within the best s-term approximation of its gradient up to a logarithmic factor, and this factor can be removed by taking slightly more measurements. Along the way, we prove a strengthened Sobolev inequality for functions lying in the null space of suitably incoherent matrices.
Metadata: arXiv. Abstracts retain their original authors’ rights.
In the conversation
Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels
Specific modern technical issues with the original scan are:
Lena contains about 250,000 pixels, some 32 times smaller than a picture snapped with an iPhone 6. And then there’s a quality problem: The most commonly used version of the image is a scan of a printed page. The printing process doesn’t produce a continuous image, but rather a series of dots that trick your eye into seeing continuous tones and colors. Those dots, Acton says, mean that the scanned Lena image isn’t comparable to photos produced by modern digital cameras. and:
[..] Scott Acton, the current editor of IEEE Transactions [..] and two former editors revived the old call for a ban, and proposed to…
Read the full comment
the journal’s editorial board that IEEE Transactions institute a moratorium on Lena research. “In 2016,” he says, “demonstrating that something works on Lena isn’t really demonstrating that the technology works.” ~ https://www.theatlantic.com/technology/archive/2016/02/lena-...
the implication being that too many algorithms were trained on the Lenna image, to the point where the ASCII phrase LENA.JPG can now be expanded to the full original uncompressed image.
It, along with the Mandrill image, are quaint historic footnotes examples of types of images (drum scanned from magazine prints) that are no longer anything close to mainstream examples suitable for reference.
> it's surprisingly hard to find a good alternative test image.
https://arxiv.org/pdf/1202.6429v3.pdf
uses a licenced image of Italian heartthrob Fabio Lanzoni with skin tones and fine detailed blond hair.
I feel that whenever Lena is mentioned, we should also give a shoutout to the researchers who were fed up with her and decided to use Fabio instead.
https://arxiv.org/abs/1202.6429
What do you guys (yes, guys) think about making Fabio the standard image instead?
Two female image processing researchers recently used a picture of Fabio as a test image in a paper of theirs:
http://arxiv.org/abs/1202.6429
(Additional discussion at http://www.kitware.com/blog/home/post/246)
Personally, I'm all for the use of amusing (arousing?) test images, as long as they exhibit features that actually make for a good test image (a mix of flat and textured regions, sharp boundaries, etc.).
Recent work has demonstrated substantial gains on many NLP tasks and benchmarks by pre-training on a large corpus of text followed by fine-tuning on a specific task.
Despite their massive size, successful deep artificial neural networks can exhibit a remarkably small difference between training and test performance.