227
edits
User:Dvdkhl/Ideas: Difference between revisions
→MangaDB File/Chapter/Volume Hashing
m (→Prerequisites) |
|||
| Line 206: | Line 206: | ||
* Image Hashing | * Image Hashing | ||
** http://www.phash.org/docs/pubs/thesis_zauner.pdf | ** http://www.phash.org/docs/pubs/thesis_zauner.pdf | ||
** Resizing based (not optimal) | ** Resizing based (not optimal?) | ||
*** http://www.hackerfactor.com/blog/index.php?/archives/432-Looks-Like-It.html | *** http://www.hackerfactor.com/blog/index.php?/archives/432-Looks-Like-It.html | ||
*** http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html | *** http://www.hackerfactor.com/blog/index.php?/archives/529-Kind-of-Like-That.html | ||
** Detect Text Bubbles | |||
** Simple OCR | |||
*** Recognising which character set is used (Latin, Asian) | |||
**** Infer which language is used | |||
=== Image Hash Algorithm === | |||
* Convert to grayscale (Black and white with thresholding?) | |||
* Rotate image so the height is always greater or equal to the width | |||
** This is done to reduce the ratio distortions | |||
* Resize to a fixed Dimension | |||
** Needs to ensure that the written text is still readable | |||
** Proposed Dimension: 1024x2048 (common ratios: 1:3 and 2:3 => 1.5:3 => 1:2) | |||
* Tile the image into 64x64 blocks | |||
** With this we get 512 blocks for all images | |||
* Apply 2D DCT for each block | |||
* Take only a specific range of coefficients from the 2D DCT Block | |||
* Produce hash from the selected ranges | |||
==== Grayscale Conversion ==== | |||
<pre>byte ToGrayScale(r, g, b) { | |||
var min = Math.Min(r, Math.Min(g, b)); | |||
var rgDiff = 1 + Math.Log(1 + Math.Abs(r - g), 256) * 2; | |||
var rbDiff = 1 + Math.Log(1 + Math.Abs(r - b), 256) * 2; | |||
var bgDiff = 1 + Math.Log(1 + Math.Abs(g - b), 256) * 2; | |||
var val = (byte)MathEx.Clamp(0, min * (rgDiff + rbDiff + bgDiff) / 3, 255); | |||
return val; | |||
}</pre> | |||
* Grayscale images are unaffected | |||
* Pixel showing colors are suppressed | |||
** Works as long as there are (black) contour borders | |||
** Might need border detection for contourless images | |||
== Remove/Change "hostile" judgemental anime/character tags? (Abandoned) == | == Remove/Change "hostile" judgemental anime/character tags? (Abandoned) == | ||
* Meh, just me overreacting a bit | * Meh, just me overreacting a bit | ||
** Approval rating seems to take care of it | ** Approval rating seems to take care of it | ||