A guide to reading faces

What physiognomy is

The Encyclopedia of Korean Culture defines gwansang as a form of divination that reads a face to judge fortune and to ward off misfortune. The word divination matters. The encyclopedia itself files this under divination, and does not present it as a science for finding out a person's character or ability.

The same entry writes that Chinese physiognomy is thought to have entered Korea in the early seventh century, during the reign of Queen Seondeok of Silla. Thought is the encyclopedia's own word. It then traces a line through the Goryeo and Joseon periods in both monastic and popular practice.

The lineage of books turns on two treatises, the Mayi and the Damo, with others such as the Shuijing ji and the Shenxiang quanbian alongside them. This app quotes only the digital text of the Shenxiang quanbian and two encyclopedia entries. For the rest we either could not obtain machine readable text or the edition is still in copyright.

The frames the books use

The encyclopedia lists eleven frames for reading a face, including the five organs, the six treasuries, the three powers, the three sections, the five stars, the six luminaries, the five peaks, the four rivers, the twelve palaces and two sets of halls. Eleven is a lot. It also tells you that this is not one unified system but several strands handed down together.

This app covers three of them. The three sections divide the face into bands from top to bottom. The five peaks match the forehead, the nose, the chin and the two cheekbones to five mountains. The twelve palaces give a name to each of twelve places. We chose these three because their locations are definite enough to point at on a photograph.

The Shenxiang quanbian sets the section boundaries like this: from the hairline down to between the brows is the upper section, from between the brows down to the nose is the middle section, and from below the nose to the chin is the lower section. The same book quotes another work with slightly different boundaries, and this app uses the first definition.

The positions of the twelve palaces come from the opening sentence of each palace in the same book. The palace of destiny is between the brows, the palace of wealth is the nose, siblings are the brows, house and land are the eyes, children lie under the eyes, servants sit at the jaw, the spouse sits at the outer corner of the eye, illness sits at the root of the nose, movement lies by the ends of the brows, office is the middle of the forehead, fortune is the temple, and countenance is the peaks and the sections read together.

What the app actually does

When you choose a photo, the browser finds 478 points on the face. It then does two things with them. It locates the three section boundaries and works out what percentage of the face height each band takes, and it places the named spots on an outline drawing.

It also measures a few widths: between the cheekbones, across the jaw, between the eyes, one eye, the nostrils, the mouth and one brow. All of them are printed as a fraction of the width of the face. Printing millimetres would imply we know the distance to the lens, and we do not.

There is no reference line anywhere in this. No ideal proportion, no average face and no golden ratio exists in the code. That is why nothing here can say how far off you are, and why no score, grade or ranking can be produced. There is no screen that compares you with anyone.

If the points cannot be found, the app says so and stops. If more than one face is visible, it will not pick one for you. Squeezing a result out of a single photograph makes for a smoother screen, but it sells a certainty that is not there.

The program that reads faces is hosted on our own server. No outside script is loaded, and the browser security policy blocks any connection beyond our own address. The library we use contains code that sends usage statistics to Google, and that policy blocks it too.

Choosing a photo

A photo taken straight on reads best. A turned head changes the widths and a lowered head changes the heights. This app straightens a sideways tilt only, and cannot undo a turn or a nod.

Bright, evenly lit photos work better. A fringe over the forehead pushes the top point up into the hair. Thick spectacle frames hide the places around the eyes. A passport style photo, with the face large in frame, is steadier than one taken from far away.

Feed in a different photo of the same person and the numbers shift a little. That is normal. These numbers describe the photograph at least as much as the face.

How to read the result

Sentences in this app have one of two subjects. One is the books, and those sentences sit inside quotation marks with a source and the date we checked it. The other is the app, and those sentences say only where a place is and what a measurement came to. The app never says what kind of person you are.

What the books say mixes the auspicious and the ominous. We did not filter those passages out. Filtering would misrepresent the books, and rewriting them in our own words would make the judgement ours. Instead we simply built no place where those words get applied to your face.

Where sources disagree we left the disagreement standing. The encyclopedia and the Shenxiang quanbian assign the eastern and western peaks to opposite cheekbones. We could not establish which is right, and printing both seemed the honest way to handle it.

Honestly: when you work with faces you cannot fully avoid thoughts about appearance. The books themselves are full of good and bad, and a reader takes even a quotation personally. What we can do is keep the books as the speaker and build no score, no grade and no comparison at all.

Where people get stuck

If you cannot see the button for choosing a photo, the switch for reading a face photo is off. It starts off for everyone who arrives.

The first run takes a few seconds to a minute while the browser downloads the program that finds face points. It is about 15 megabytes and comes only from our server. After that it is fast.

If it says no face was found, the photo is probably dark, or the face is small, or it is turned to the side. If it says the program will not start at all, that browser does not support the format we need. The dictionary of places still works in that case.

There is no place to save a result. If you want to keep one, take a screenshot. This app leaves nothing on a server, and the only thing left in this browser is the marker that the switch was turned on.