Insiders Hub

AI based Smart Mirror for enhancing selfie experience

Tina Chandwani

Tina Chandwani

Mumbai, Maharashtra

1 0
  • 0 Collaborators

A smart mirror is an application that enhances the user-experience of taking pictures.It classifies the natural beauty of humans which is not limited to fairness. Our system will help categorize beauty with respect to the anatomy of the face and other attributes. ...learn more

Project status: Published/In Market

Artificial Intelligence

Docs/PDFs [1]Code Samples [1]Links [1]

Overview / Usage

This project aimed at designing a web-based application that can not only see the skin-deep reflection but also go beyond an ordinary mirror and can classify hair and skin type by just seeing the image. The application can be used to see the changes in a person like change in hair color, skin tone, and so on. Knowing your skin and hair type is essential as proper measures can be taken to avoid any further damage to hair and skin. The type of skin a person has depends on the amount of water and oil in the skin. If these are out of balance, a person may have to suffer.

In order to know a type of skin and hair with a click, we have built an application that can classify hair as thick, thin, medium, long, straight, so on and skin as fair, acne breakout, oily, sensitive, so on. The main features of Smart Mirror are 1) extensible: many such modules can be integrated. 2) changes in skin and hair can be recorded with a click .

Methodology / Approach

** Dataset**

A lot of datasets were searched but none of them had the attributes that met our proposed system’s requirements. Hence a custom Dataset was developed each for skin attributes and hair. The images were manually searched and downloaded from google.com and were manually mapped with their respective attributes. The dataset for skin had over 850 custom images with 21 attributes each. These attributes were: Acne Breakouts, Dark skin tone, Dry skin, Earings, East Asian, Fair Skin Tone, Heart-shaped face, Left side face, Medium skin tone, Necklace, Oily, Oval-shaped face Resistive skin, Right side face, Sensitive skin, South East Asian, Tattoos, Visible pores, Glasses, Goggles, Western. The dataset for hair had over 500 custom images with 11 attributes each. These attributes were: Hair color: black, brown, blond Hair length: medium, long, short Hair quality: Thick, thin/fine Hair type: Straight, wavy, curly.

Methodology

  1. Data in a structured format.
  2. Loading and Pre-processing of the data
  3. Defining the model’s architecture: This includes deciding the number of hidden layers, number of neurons in each layer, activation function, and so on.
  4. Train the model using Multi-Label Image Classification using Keras and TensorFlow module.
  5. Use the trained model to get predictions on new images.

Technologies Used

  • Tensorflow
  • Keras
  • Python
  • Jupyter Notebook
  • Google Drive
  • HTML
  • Javascript
  • Bootstrap

Documents and Presentations

SMART_MIRROR_Report.pdf

Repository

https://github.com/TinaChandwani/Smart-Mirror

Comments (0)