Abstract
PERFORMANCE EVALUATION OF SHAPE-SPECIFIC AND GENERAL DEFINITION OF MODELS USING DYNAMIC DEFORMABLE TEMPLATE FOR IMAGE-BASED OBJECT RECOGNITION
Bilkis Jamal Ferdosi* and Muhammad Masud Tarek
ABSTRACT
Model-based segmentation and recognition can be invaluable if a prior knowledge about the object?s shape is available. It enables a model description that may even override data information in case of inaccurate and missing information. However, achieving proper model description and its in-class variation is not a trivial task. In literature there exist several approaches. Some require enough representative samples, some fails in case of large size and shape variation etc. In our work, we propose a template combining deformable templates and physically based models together. Using the proposed template we study i) whether object?s shape-only-template is sufficient or information about the space surrounding the object should be incorporated, iii) whether there exist a useful quality measure for matching templates. We found that the proposed dynamic deformable template works successfully. Incorporation of the surrounding space information in the template significantly improves the recognition rate which also enables a better template matching criteria.
[Full Text Article] [Download Certificate]