- Problem Statement
- Executive Summary
- Data Sources
- Data Dictionary
- Conclusions and Recommendations
- Requirements
Online shopping is a huge market. For individuals who wear make up, store closures have been a huge problem with testing and matching foundation shades. What if you don't need to go inside of a store to find a foundation that matches your skin tone? The purpose of this application is to match an individual to a shade of foundation sold by sephora or ulta (two of the biggest cosmetic stores within the United States) through image recognition.
This application utilizes various technologies listed below. It begins by connecting to the individuals computer webcam and then snapping a picture of the individuals face. If webcam is not assesible , an individual can upload an image in the form of .jpg or png. It then recognizes which regions in the picture are indicative of skin color and extracts the pixel-based color and finds the dominant colors. It does this through unsupervised learning including K-Means Clustering Algorithm. After extracting the colors in the form of RGB, the values are then compared to a dataset with over 3,000 foundation shades and it will recommend a foundation that is most similar. The recommender is using cosine similarity of the hex values of the skin compared to the dataset of foundations.
allShades.csv: | Description of the dataallshades_updated.csv: An Updated Version of the dataset with the converted HEX values to RGB
| Feature | Type | Dataset | Description |
|---|---|---|---|
| Foundation | object | allShades_updated | The Name of The Brand and the Product and The Product Number |
| URL | object | allShades_updated | The URL to the Product online |
| Shade | object | allShades_updated | The Name of Product Shade |
| HEX | object | allShades_updated | The Tuple of the converted HEX values |
| Hue | float | allShades_updated | The Hue Value of the shade |
| lightness | float | allShades_updated | The Lightness Value of the shade |
| red | int | allShades_updated | The red value of the RGB from the converted HEX Code |
| green | int | allShades_updated | The green value of the RGB from the converted HEX Code |
| blue | int | allShades_updated | The blue value of the RGB from the converted HEX Code |
This includes the following:
- python 3.6 or later
- pandas
- numpy
- open cv
- colors
- Scikit learn
- Counter
- imutils
- pprint
- matplotlib
- pickle
- joblib
In conlcusion our algorithm is able to extract the image and determine the skin color, and make a recommendation based on cosine similarities. The next steps include building out the application in Flask. Also would increase the sample size of images on the algorithm.