Bilkent University
Department of Computer Engineering

---

S E M I N A R

 

Finding Compound Structures in Images using Image Segmentation and Region Relations

 

Daniya Zamalieva
MSc.Student
Computer Engineering Department
Bilkent University

The increase in the amount of available high-resolution remotely sensed data is subsequently causing augmentation of applications that aim automatic information extraction and knowledge discovery. One interesting way of enabling high-level understanding about the image content is to identify significant image regions that are internally heterogeneous. These image regions are comprised of more than one object of diverse types and hence can be also referred as compound objects. In comparison to single object detection, studies that aim to detect the compound objects are not encountered frequently in literature. Several segmentation algorithms have been proposed to partition the image into homogeneous regions; however,in order to obtain meaningful regions that are internally heterogeneous further exploration must be performed. In this work, we propose a method for discovering interesting and significant compound objects regardless of their type. We translate the image segmentation into the relational graph and apply the graph knowledge discovery algorithm to find interesting and repeating substructures that correspond to desired compound objects.

 

DATE: 4 May, 2009, Monday@ 16:20
PLACE: EA 409