Analysis of border ownership cues and improvement of depth prediction using border ownership

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Bilgisayar Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2014

Öğrenci: MEHMET AKİF AKKUŞ

Danışman: SİNAN KALKAN

Özet:

Border Ownership is the problem of identifying which image regions own the image border. This information is essential and important for large variety of high-level vision problems such as object segmentation, object recognition, depth perception, motion perception etc. Current computational approaches to Border Ownership (BO) estimation either use artificial or limited number of real images. In this thesis, we propose a new comprehensive BO database, including 500 indoor and 500 outdoor images whose BO information is labeled by human participants. Using this dataset, BO estimation capability is investigated for several visual cues such as T-junction, L-junction, curvature, lower region and contrast both individually and combinatorially. Then using these cues, a basic computational model is proposed which estimates BO information based on the majority rule. Moreover, a new method, which merges a feature-based stereo algorithm and BO information, is proposed for more accurate depth prediction on homogeneous areas.