Binarization is a classic task of image processing. Binarization is often used to simplify data and speed up subsequent processing, which nowadays does not seem important. But in the analysis of porous materials, binarization is fundamental, since the data model here does not imply an intermediate state between an empty pore and an impermeable matrix. But the algorithm that works perfectly "out of the box", as usual, is not. There are algorithms with tuning parameters, there are wonderful neural network architectures. For them to work, they need to be set up / trained. What to do if in our task obtaining reference answers is very laborious? From this article, you can learn about one curious way to do without markup, as well as get acquainted with the world of computational tomography and adjacent areas.
, , .
. , , – , – .. . , ( , ).
, , , , , , . , , . ? .
() . . 1 , , , . . . – , , . , , - .
. 1. () () .
. , , 0 180 . 2D (. 2). .
. 2. .
, . ( ). , – . 3. , , .
. 3. – . , .
. . . , . , . .
– , . , . ( ) ( ) .
, , . , , , , , . , - .
, - – . . , - . , , . 1 , «ought to be enough for anybody». , . .. « » – , , (. 4). . , – , : – , . ( ), ( «» -, ).
. 4. . . .
, , . , , ., . – ( ), .. (. 5). : ? , , « » . , , .
. 5. .
, [1]. Document Image Binarization Context (DIBCO) 2017 : - . , ! , ( ), , , . , , . , , , . 6. , – . , ( ). , .
. 6. . (). ().
, , , . , , ( ), . , , , , , . .. (ground truth) – . . : , . ?
, , . . , – . – , , .
. 7. – , – , .
7. , , , , . , . , . ? , , , , . , , . – , .
, ( , ) : « » . . -, «», ( , ). , , «», . -, , . , , «», , , . , «», . , .
, – « » , . , : « » … . , . , [2]. , , , , . – .
, ( ) . : , .
SSIM , . SSIM , , .
. 8. – , – , .
, «» (, ), . , . 8 , . - , . – . . , , . , , , .
[1] Bezmaternykh PV, Ilin DA, Nikolaev DP. U-Net-bin: hacking the document image binarization contest. Computer Optics 2019; 43(5): 826-833. DOI: 10.18287/2412-6179-2019-43-5-826-833.
[2] .. , .. . . 2013; 63(3): 85-94.