Improving a portrait photo using face retouching and face morphing
Date of Issue2017-05-19
School of Electrical and Electronic Engineering
This project is aimed at improving a portrait photograph through face retouching and face warping approaches. To localize our implementations on the facial skin, the first step was to develop an efficient and accurate skin map generation algorithm. Out of the two methods to detect skin, region-based methods are more suitable for determining the physical location of the face as a whole and not definitive boundaries and contours of skin. Pixel based methods are location independent and classify a pixel as skin or non-skin based on simplistic heuristics. Hence pixel classification in RGB space was implemented for skin detection. Next step was to smooth the skin with a suitable low pass filtering approach. Normal low pass filtering, median filtering, Gaussian filtering and bilateral filtering were tested. Bilateral filtering produced the best results comprising appropriate smoothing with edge preservation. Face warping comprises mapping of pixels from source to destination via apt interpolation algorithms. Previous researches on face warping techniques suggested that although mesh warping was computationally the fastest, thin plate splines interpolation was best suited for user interaction which is imperative to this project. Number of control points required to achieve qualitatively satisfactory results was high in mesh and feature lines based warping, while thin plate splines was proficient in generating desirable transformations with minimal annotations. Finally, all functions were integrated in a MATLAB based utility involving user interaction to generate a skin map of required strength, smoothing of desired degree and customized warping by interactively selecting source and destination points.
Final Year Project (FYP)
Nanyang Technological University