IMAGE FEATURE EXTRACTION THROUGH SCALE-SPACE FILTERING

Abstract

We present new approaches on feature extraction and representation of planar objects. Description about efficient features of an object is first addressed and the effects of scale-space filtering on these features are discussed in detail. For image feature extraction purpose, two approaches are proposed, both are based on scale-space filtering. In the first approach, circular arc segments are extracted, while in the second one, a set of image features including CORNERs, ENDs, ARCs and LINEs is determined. This forms an integral feature set for describing an object. For image representation purpose, three kinds of invariant feature parameters are proposed. A number of experimental results show that our approaches are robust on translation, rotation, scaling of the objects as well as noise corruption. Further research work includes recognition of non-occluded and partially occluded objects based on the extracted feature set and its representations.