Added "hair color intensity" as an indexing option for characters.

This is somewhat of a very long term project. Adding values for 29,000 characters will take time, and coming up with a standard means of measuring intensity will also take some time to figure out.

For now, feel free to take a stab wherever seems appropriate. We're using a percent scale, currently tuned from:

0%
10%
20%
30%
...
100%

following 10% increments. I can open it up to 1% increments later on.

Essentially, the goal is to be able to do, let's say, a Blue hair search, then order the characters by intensity. The ones at the top of page 1 will have the most intense hair color, and as we scroll down and go page by page, the hair color should continually get less intense.

Every character initially has an intensity of 0% ( or unset ).



One thing I'd like to do early on is establish 100% intensity examples, and 10% intensity examples.

Thoughts, ideas, pictures?

[ Edited Aug 22, 2011 ]

私は日本語を少し読むと書くをできます。あなたも？

I think it's an interesting idea. Guess I'll give it a try, but it'll really feel I'm muddling through it because I don't have any examples.

I do think it might come in handy when searching. I'm assuming that lesser intensity is both very dark and very light shades?

Kind of difficult to decide what should be the 100% mark. Maybe someone like:
- Makoto Kasai
- Jun Seibun

I think a good example of 10% would be:
- Quorelia Retie
- Honoka Yukishiro
Mainly due to the fact that I at least had to take a closer look to see if they were blue.

I personally think we should permanently affix either light or dark as highest intense and work from there, simply because it keeps things simpler by having a two dimensional chart(color, brightness) instead of a three dimensional chart (color, brightness and shade). Also, we face the additional problem of multicolor gradients, by which I mean hair that could be pink or red (or some other closely related colors) as people searching might think one is a "low intensity red" as opposed to a "high intensity pink".

I doubt the effort vs merit of this project.

As long as it's only used for ordering results and not filtering, I suppose it couldn't do any harm, but I can't help but feel your time is better spent doing something else.

I think it should not go any finer than 5% steps if implemented, because 1% would be too arguable and may be too hard to judge the difference.

But I can see that there is probably going to be a quality control issue without some sort of comparison chart (which may get large in resolution) as a reference.

Because this is based off of the images, the image may not provide correct data. Let's use Shinobu from Love Hina as an example. In some instances from the anime (screencaps) her hair might be about 75%, but if you look at images from the artbook, you could probably argue that it's 90%. And the image from the artbook may also be incorrectly processed (exposure, color, saturation, and/or gamma imbalance), which would lead to a false intensity/shade. So to get a good value, we would need some sort of "system" to solve this problem. We would need pictures that are "properly adjusted" (which we would need to also create a standard) before continuing on to judging the shade. Afterwards we would have to take the lightest possible shade that the character's hair appears (75% for Shinobu) and the darkest shade that the character's hair appears (90%) and use the average for the shade (which would b 82.5 and rounded up to 85%).

But also another problem exists, which is gradient hair (Potemayo and very few other games/animes). The characters with gradient hair in Potemayo is Guchuko, Mikan, Nene, and Potemayo. For the latter two, there is more of one color than the other, but for the former two, the two colors that make up the gradient are about equal. Luckily, the two colors that make up Mikan's hair is pretty close in intensity/shade, so it won't carry my point as well as Guchuko. Guchuko's hair is a lavender/magenta gradient, and the lavender is obviously lighter than the magenta. So how would we judge what her hair color intensity/shade is? Would we average both colors of hair together or use some of other sort of method? This may make judging that much more difficult when it becomes the situation.

Because of the standardization/quality control that it may require, I will agree with Kyareshi that the merit of adding this "system" may not be worth all the effort.

Thanks everyone for the feedback!


From what I can see, some simplification is needed to get it started.

I'm going to add 3 labels mild, medium, strong to the list.

20% = mild

50% = medium

80% = strong


This leaves room for fine tuning where needed, and makes the decisions a little easier initially.




I did have this as a light/dark flag originally, but there proved to be far too many situations 2 values were not enough to meet.

You raise a good point about gradient hair 13x31. That's going to be a problem we'll need a clear standard for how to handle. Or we could just mark those characters as having gradient hair. Wouldn't work for sorting, but would work for filtering. Hmm...


I can understand Kyareshi's point. I don't know if the effort will merit the end result. But I feel it's worth a try.


And to further touch of 13x31's point about different images of the same character: It's a challenge I may have to address by diverging ( correct word? ) the site along two paths.

Path 1: Characters
Path 2: Images

In order to meet our goals of helping people identify any image of any character, it may be necessary to start indexing the images themselves, and link those up to character profiles?



I think those are good examples LiCobra.
has a near solid color.


While she has a near absence of color:

私は日本語を少し読むと書くをできます。あなたも？

You're welcome for the feedback.

Perhaps mild should be renamed as weak to avoid any possible confusion.

##I think "divulge" technically works, but "divide" may work better.##

[ Edited Sep 7, 2011 ]

Rei In order to meet our goals of helping people identify any image of any character, it may be necessary to start indexing the images themselves, and link those up to character profiles?

This is a very interesting proposition. The way I see it, separating the properties of "characters" from those of "images" and linking them together will solve a lot of annoying little problems – it is, in fact, a large step toward further normalization of your database. However, I have some concerns over whether or not you have the work force to do that level of data capture.

Also, as a point of interest, this step would bring your database considerably closer to that of Danbooru in terms of functionality and purpose, albeit with the advantage of recording characters and franchises as entities more than simple tags.