MSc Thesis Defense: Zeynep Akant,HAIR-ACCESSORY DEFORMATION FOR THREE-DIMENSIONAL AVATARS USING SPHERICAL RADIAL DISPLACEMENT MAPS, Date & Time: 16 July, 2026 – 10:00 AM, Place: FENS L029
HAIR-ACCESSORY DEFORMATION FOR THREE-DIMENSIONAL AVATARS USING SPHERICAL RADIAL DISPLACEMENT MAPS
Zeynep Akant
Computer Science and Engineering, MSc Thesis, 2026
Thesis Jury
Prof. Selim Balcısoy (Thesis Supervisor)
Prof. Bahattin Koç
Prof. Uğur Güdükbay
Date & Time: 16th July, 2026 – 10:00 AM
Place: FENS L029
Zoom: https://sabanciuniv.zoom.us/j/
Keywords: spherical parametrisation, hair volume envelope, accessory deformation, 3D avatar customisation, radial displacement mapping
Abstract
Placing head accessories on a 3D avatar with hair is a recurring problem in avatar customisation: an accessory authored for a bare head will either clip through the strands or sit visibly above them. This thesis presents a closed-form method for adapting head accessories to a wide range of hairstyles. The outer hair surface is summarised as a scalar function on a fixed 40 × 40 spherical grid around the head centroid, built in linear time from a raw strand point cloud without surface reconstruction or simulation. The accessory is then deformed by a single-pass per-vertex radial expansion that interpolates between skull-conforming and hair-conforming limits through one softness scalar α ∈ [0, 1]. A separate path handles rigid hats by uniform scaling and vertical lift, with brim and crown detected from the accessory mesh. Non-penetration of the skull is a property of the formulation rather than a post-hoc fix, and the method is feed-forward: no iterative solver and no neural inference at deformation time. The pipeline was evaluated on all 342 usable USC-HairSalon hairstyles, two accessory classes, three values of α, with four geometric metrics and a uniform-expansion baseline. An online study with N = 40 participants tested α perceptually. Viewers reliably ranked the three α levels by perceived hair compression on both a procedural and an artist-authored beanie (mean Spearman ρ ≈ 0.57, p < 0.0001), and preferred the mid-point over the rigid no-inflation limit on a clear majority of trials; the mid-point and the compliant limit were not distinguished.