Region and Shape Extraction Huiqiong Chen Faculty of Computer Science Dalhousie University
Aims Goal of this research Motivation Applications Extract meaningful regions from image and estimate their shapes without intensive computation. Motivation Taking advantage of inherent structure information carried by each GET feature, the perceptual structure of region shape can be obtained easily as well as region interior attributes. Applications Image/Video representation Region-based Image/Video retrieval Video surveillance Medical Imaging License recognition and tracking
Key Idea All meaningful regions can be represented by two types of perceptual GET-based closures in Perceptual region hierarchy. An image can be transformed into GET space on the fly represented by a GET graph, which presents perceptual organization of GET associations. Region detection can be achieved by perceptually grouping closures in GET graph.
Perceptual Region Hierarchy Regions can be classified into two types: Object outline: describe an whole object Basic region: describe basic inner component of object
System Architecture
Perceptual Closure Detection
Example Original image Extracted GET features
Example (Cont’d) Contours in image Basic regions
Advantages Provides a real-time region detection system in which region shape structure can be extracted at the same time. It achieves high accuracy of detected regions without intensive computation. Suitable for segmenting regions with arbitrary shapes. Both object contours and their components can be detected, based upon which all meaningful regions can be conducted hierarchically.
Detected Samples Original image Extracted GET features GET-based region contour Filled regions
Detected Samples (Cont’d)