Based on your colors - who you are - we determine the colors that are good for you - or bad for you.
We use a selfie to determine your key colors - from your skin, your hair, and your eyes.
Using Fashion Science we then build your personalized color wheel - showing what colors work best for you.
BestColors is a set of apps (iOS and Android) and an API service to support:
- Capturing your selfie and transforming it into your best colors.
- Evaluating any image, say of clothes in a store as to their fit to your best colors
- Allowing the evaluation of specific colors, say from an online store as to their fit to your best colors. This is through an API.
This is an open source project. Collaborators and researchers are welcome.
The goal is to identify the color of clothes that match best to you. With the right colors the claim is this makes you more attractive than wearing neutral or bad colors.
Historically the approach taken is seasonal colors, as illustrated by the popular book in the 1970's 'Color Me Beautiful'. This segments humans into four seasons - winter, spring, summer and fall - each season having specific colors that are appropriate for people in that season. Further refinement has taken the four seasons approach into more segments, still based on seasonality - 12 and 16 segments are common.
From our experimental data testing this approach - we have analyzed well over 100,000 selfies from a global sample - it just doesn't hold water that people segment this way. It turns out that humans range over this space in a continuous fashion rather than supporting a statistical bucketing. And anecdotally this was confirmed - "oh, I'm not really a Spring - in some cases I'm a Summer", ...
So the challenge was - how to create a personalized color wheel?
We define the physical 'you' as your selfie - from which we identify your skin, hair, and eyes. Those colors define the current 'you'.
At that point we use color harmonies to determine the range of colors that work with 'you'. These color harmonies include:
- monochromatic
- complementary
- analogous. (A fun tool to explore color harmonies is the Adobe Kuhler tool - https://color.adobe.com/
Note that all harmonies are based on the source colors. All colors in the color wheel can be traced back to the selfie.
This works great in a perfect world, with perfect lighting, high quality cameras and color correction approaches such as a gray card. Unfortunately that isn't the normal world we live in. Take a closer look at the selfies of your friends, say in Facebook. Your eye is good at figuring out what 'skin' should be - but many selfies are far off what a 'human' should have for a skin tone.
Key issues we've seen, in well over 100,000 selfies so far include:
- Poor lighting, generally too dark
- Shadows from bright lighting, e.g., hiding the eyes
- Other poor composition including bright backgrounds that blend into the hair, and others
- Unknown errors generated by the smart phone camera image processing system (e.g., focusing on something other than the face for a selfie.)
An approach discovered here is a method to dynamically determine the appropriate adjustments on a selfie to bring colors back to the real world. This is an AI algorithm (using a CNN) that is patent pending.
Note that evaluating a real world image, e.g., a store against a personalized color wheel involves using a non-corrected image. This is a problem. Currently we expand the error bars for matching, i.e., allowing more permissable colors.
Evaluating colors against a catalog, where the colors are explicit and 'correct' works very well.
- What about bright yellow for me? (or Green, Blue, etc.) Go for it. By it's nature this approach (currently) won't find bright colors for you outside of your source colors (monochromatic harmony) - your source colors are generally low in saturation - or close to the center of the color wheel. This is an open topic of research.
- What about highlights in my hair? Colored contacts? Strong make-up? These colorations are used to define 'you', so your color wheel - your preferred colors will reflect the new 'you'.
- What about patterns in fabric? Patterns in fabrics are designed so that the individual colors work well together. Your color wheel will usually support all or none of the colors. If a partial match occurs, then we've got to go with the science - it's not best for you.
Collaborators are invited. There are many areas to be explored, ranging from the technical (code/hosting) to AI to fashion science and applicability.
Thanks to the team so far: Peter, John, Cindi, and Bruce.