Feature extraction is an important component in all areas of imageprocessing a fact demonstrated by the wide variety and diversity of themethods available. These range from statistical to human vision basedapproaches. Although progress has been fruitful and uninterrupted, it isalso apparent that as yet to unified theory of feature identification orrepresentation has emerged. It is towards this goal that this work isdirected. The approach adopted has two fundamental principles:meaningful image features are inherently localised in both the spatialand spatial frequency domains; the degree of this locality is notconstant across the range of features, in general, image features existwithin a multiresolution space. Based on these principles, an attempt ismade in this work to provide a unified feature extraction framework.Starting from a general image model, a feature estimation scheme isdeveloped which, by way of example, assumes the image to consist of amultiresolution set of line or edge features. The estimation is achievedby the use of an invertible transform, which by definition incorporatesthe multiresolution structure underlying the model. The work concludeswith a discussion on the appropriateness of the approach to more complexfeatures, such as curvature and shape
展开▼
