The establishment of a shape correspondence is mandatory for localized shape analysis. This step can address fundamental questions about how shapes change over time and which parts or locations differ between them. Our team has developed advanced surface registration techniques that establish shape correspondence across complex shapes, including partially missing or highly variable structures, enabling more accurate population shape analysis. [read more]

Dividing 3D objects into meaningful parts is a crucial step in understanding shape patterns for shape matching and statistical analysis. For instance, segmenting the human body into smaller parts, such as arms and legs, is essential for gaining a deeper understanding of its structure. Our team has developed automatic surface annotation techniques that utilize deep neural nets, incorporating novel architecture designs and data synthesis methods. [read more]

Surface feature extraction provides a way to represent complex shapes using compact landmarks. Typically obtained in the form of curves, such as ridges or valleys, these features allow us to reduce the high dimensionality of 3D objects while preserving their essential characteristics. Our team has developed automatic landmark extraction techniques to identify unique and meaningful features from surfaces for structural variability analysis. [read more]

Quantifying 3D shapes provides a quantitative means of explaining shape patterns and addressing questions related to shape folding and differences between corresponding regions. Our team has developed novel shape quantification techniques that provide quantitative shape markers, supporting the understanding of human cognition, behavior, and brain disorders such as autism spectrum disorder and Alzheimer's disease. [read more]