In this evolving digital society, portrait images are a booming part of everyone’s daily life. Whether it is for the social media or for the glamor industry, self-photographs have become an indispensable necessity. Multiple photo-editing software such as Adobe Photoshop are utilized to perform further editing and enhancement of one’s face. These makeup tools can digitally beautify the face in a click. Even beauty industries embrace virtual makeup to support their customers’ online shopping experience.
The overall purpose of this project is to evaluate the proof of concept behind virtual makeup. Through a simple automated algorithm, curious individuals can experiment with how their appearance will change according to the latest trends. To support the idea of digital makeup, I implemented a state-of-the-art algorithm that was able to do semantic segmentation of portrait images using fully convolutional networks (FCN) and other open-source libraries (GluonCV, Face-Parsing.PyTorch, Caffe), followed by an example-based skin and hair colour transfer using N-dimensional Probability Density Function (PDF) statistical transfer. This algorithm transfers the full RGB colour histogram through a series of 1-dimensional histograms, allowing us to transform the target image part with respect to the reference image.

• Implemented state-of-the-art semantic segmentation techniques with a part-to-part color transfer to create a digital makeup look in 284s (average for professionals 1.5hr ≈95% improvement).

Results when apply the color-transfer algorithm on features of skin, eyes, lips, hair, eyebrows.

Python MatLab

The purposes of this paper is to develop a web search engine interface illustrating the latest news from the top 11 news channels from Instagram. The search engine will display relevant results as per the user’s query such that the user is able to receive a comprehensive overview of the different perspectives of various news channels.
My role was to prepare the dataset for the web search engine.

• Successfully scraped 1000 top posts with critical metadata (30s/post) from 11 news channel in Instagram using Python and Selenium.
• Conducted language processing using Python to clean and prepare the data.

The problem statement given was to: Gamify & socialize the teaching & learning of software engineering courses​; Empower Students to learn and compete with each other​;Help Professors to assess the mastery of a course through relevant data analysis​.
My role was to prepare the designs and user experience of the game, and to build the game platform.

• Developed a mobile learning game platform on Unity to develop an android-based application
• Utilized Photoshop to build an interactive and aesthetic graphical user interface.

We propose a game application called Timely Buddy that is targeted at individuals who are determined to resolve their tardiness. Gamification provides the fun element to motivate participation by leveraging people's natural tendencies for competition, achievement and collaboration using game design tools such as rewards and leaderboards. By making the process fun and enjoyable, users are encouraged to use the game application daily and consistently to resolve the problem of tardiness.

• Proposed a gamified mobile application by utilizing Behavioural Economics principles (Nudge Theory, Gamification, Goal Setting Commitment / Ranking Strategies) and concept of gamification to motivate participation by leveraging people's natural tendencies for competition, achievement and collaboration.

Our objective of this research is to propose a methodology of enforcing a government-initiated compulsory maximum limit on the amount of money gamblers can spend. This limit varies with respect to the individual’s monthly income. This is for the scenario of venue-based gambling amongst an older Asian cohort in Singapore. Pre-commitment limit-setting involves users specifying the maximum they would like to spend before engaging in the act of gambling.

• Identified a specific target population to explore how harm associated with state lotteries could be minimized through involuntary pre-commitment limit-setting.
• Proposed an econometric model of Ordinary Least Squares (OLS) with Difference-In-Differences Estimator to evaluate the impact of the treatment in the Randomized Controlled Trial (RCT) setting.