Use this URL to cite or link to this record in EThOS:
Title: Study of object recognition and identification based on shape and texture analysis
Author: Wang, Guanqi
ISNI:       0000 0004 2743 620X
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2012
Availability of Full Text:
Access from EThOS:
Access from Institution:
The objective of object recognition is to enable computers to recognize image patterns without human intervention. According to its applications, it is mainly divided into two parts: recognition of object categories and detection/identification of objects. My thesis studied the techniques of object feature analysis and identification strategies, which solve the object recognition problem by employing effective and perceptually important object features. The shape information is of particular interest and a review of the shape representation and description is presented, as well as the latest research work on object recognition. In the second chapter of the thesis, a novel content-based approach is proposed for efficient shape classification and retrieval of 2D objects. Two object detection approaches, which are designed according to the characteristics of the shape context and SIFT descriptors, respectively, are analyzed and compared. It is found that the identification strategy constructed on a single type of object feature is only able to recognize the target object under specific conditions which the identifier is adapted to. These identifiers are usually designed to detect the target objects which are rich in the feature type captured by the identifier. In addition, this type of feature often distinguishes the target object from the complex scene. To overcome this constraint, a novel prototyped-based object identification method is presented to detect the target object in the complex scene by employing different types of descriptors to capture the heterogeneous features. All types of descriptors are modified to meet the requirement of the detection strategy’s framework. Thus this new method is able to describe and identify various kinds of objects whose dominant features are quite different. The identification system employs the cosine similarity to evaluate the resemblance between the prototype image and image windows on the complex scene. Then a ‘resemblance map’ is established with values on each patch representing the likelihood of the target object’s presence. The simulation approved that this novel object detection strategy is efficient, robust and of scale and rotation invariance.
Supervisor: Stathaki, Tania Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral