Suputra, Putu Hendra (2023) Framework Rekonstruksi Kraniofasial 3D Menggunakan Deformasi Permukaan Dan Positioning Landmark Otomatis Berdasarkan Surface Curvature Feature. Doctoral thesis, Institut Teknologi Sepuluh Nopember.
Text
07111860012001-Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 October 2025. Download (12MB) | Request a copy |
Abstract
Rekonstruksi kraniofasial atau craniofacial reconstruction (CFR) berbantuan komputer adalah proses yang bertujuan untuk mengestimasi impresi wajah berdasarkan sisa-sisa tengkorak. Proses ini mengadaptasi metode konvensional menggunakan kerangka kerja berbasis model konseptual. Masalah yang ada dalam CFR saat ini adalah (1) anotasi tengkorak masih bergantung pada ahli, (2) pemrosesan tengkorak dalam domain tiga dimensi (3D) memiliki tantangan data volumetrik, dan (3) perlu metode yang didasarkan pada karakteristik morfologi populasi atau statistical model template. Kami mengusulkan sebuah framework rekonstruksi kraniofasial berbasis komputasi yang terdiri dari tiga tahap, yaitu membangun model kraniofasial, deteksi landmark otomatis, dan deformasi permukaan. Machine learning digunakan untuk menarik korelasi antara bentuk permukaan lokal dan bentuk landmark dan secara otomatis mendeteksi posisinya. Fitur permukaan lokal diekstraksi menggunakan Surface Curvature Feature (SCF) sebagai deskriptor 3D. Dengan menggunakan filter berbasis klaster, jarak rata-rata (ke ground truth) dari 20 titik teratas adalah 0,0326 unit, lebih kecil dari radius titik pengambilan sampel 0,05. Filter berbasis klaster lebih baik daripada filter berbasis mass-radius dan secara konsisten memberikan akurasi yang lebih baik, terutama dalam kasus multi-klaster. Data training terdiri dari 140.000 SCF untuk sepuluh kelas landmark. Tahap ketiga, yaitu deformasi permukaan, menyesuaikan bentuk template wajah ke tengkorak berdasarkan korespondensi pasangan landmark wajah-tengkorak. Deformasi permukaan Laplacian memberikan estimasi bentuk alami wajah manusia dengan tetap mempertahankan detail permukaan template wajah. Validasi lima orang ahli dari departemen Antropologi menyatakan bahwa dari hasil rekonstruksi, 91,5% dapat mempertahankan detail template dan diterima sebagai bentuk alami wajah manusia.
=================================================================================================================================
Computer-assisted craniofacial reconstruction (CFR) is a process that aims to estimate facial impressions based on skull remains. It adapts conventional methods using a conceptual model-based framework. The existing problems in current CFR are (1) skull annotation still relies on experts, (2) skull processing in the three-dimensional (3D) domain has volumetric data challenges, and (3) there is a need for methods based on population morphological characteristics or statistical model templates. We propose a computationally-based craniofacial reconstruction framework consisting of three stages: building a craniofacial model, automatic landmark detection, and elastic surface deformation. Machine learning draws correlations between local surface shapes as landmarks and automatically detects their positions. Local surface features are extracted using Surface Curvature Feature (SCF) as a 3D descriptor. Using the cluster-based filter, the average distance (to ground truth) of the top 20 points is 0.0326 units, smaller than the sampling point radius of 0.05. The cluster-based filter is better than the mass-radius-based filter. It consistently provides better accuracy, especially in the multi-cluster case. The training data consists of 140,000 SCFs for ten landmark classes. The third stage, surface deformation, adapts the shape of the face template to the skull based on the correspondence of the face-skull landmark pairs. The Laplacian surface deformation provides an estimation of the natural shape of the human face while maintaining the details of the face template surface. Five experts from the Anthropology department stated that from the reconstruction results, 91.5% could preserve the details of the template and is accepted as the natural shape of the human face.
Item Type: | Thesis (Doctoral) |
---|---|
Uncontrolled Keywords: | craniofacial reconstruction, deformasi, deteksi landmark otomatis, framework, Surface Curvature Feature, surface Laplacian, tigadimensi. |
Subjects: | T Technology > TK Electrical engineering. Electronics Nuclear engineering > TK7882.P3 Pattern recognition systems |
Divisions: | Faculty of Intelligent Electrical and Informatics Technology (ELECTICS) > Electrical Engineering > 20001-(S3) PhD Thesis |
Depositing User: | Putu Hendra Suputra |
Date Deposited: | 17 Jul 2023 06:17 |
Last Modified: | 17 Jul 2023 06:17 |
URI: | http://repository.its.ac.id/id/eprint/98511 |
Actions (login required)
View Item |