This project presents a Digital Body Reshaping Mirror, a real-time system designed to support body-image research and mirror-exposure therapy. Mirror exposure therapy is a clinically used method for treating body-image-related disorders associated with a fear of weight gain and distress when viewing one's own reflection. While some visualization approaches exist, they often rely on immersive virtual reality or 3D body scanning systems that require specialized hardware and are therefore less practical for clinical use.
This study examines whether a lightweight 2D system based on body tracking and pixel warping can generate believable real-time visualizations of simulated weight gain without requiring specialized hardware. The system captures a live camera feed and dynamically modifies body proportions using image segmentation and detected body landmarks, producing a mirror-like reflection of the user at a specified simulated weight that responds to user movement. Two implementations were developed: a Python pipeline using MediaPipe and OpenCV, and an equivalent TouchDesigner network.
To evaluate usability, a pilot user study (N = 5) was conducted using an adapted System Usability Scale (SUS) and additional survey questions assessing believability, immersion, and emotional response. The prototype achieved a mean SUS score of 63, indicating marginal-to-good usability for an early-stage interactive system. Results suggest that the chosen approach can produce believable body-size modifications in real time, though improvements in tracking stability, image quality, and anatomical realism are required for clinical deployment.
Acknowledgements: I would like to thank my adviser, Professor Shruti Mahajan, for her guidance and support throughout this research. I am also grateful to Professor Walker from the Psychology Department, who originally proposed the idea of body-weight modification for this study. I would like to thank Mai Tran and Lauren Levitt for their prior psychology research that informed this project. I appreciate their feedback and willingness to participate in and support the study.
I would also like to thank Union College Undergraduate Research for providing SRG funding, which supported the equipment and participant compensation necessary for this work.