A system for extracting local features for the recognition of partially occluded objects

Abstract EN

A number of local features extraction techniques have been proposed in the literature.

Global techniques are unable to recognize objects that are partially visible.

The objectives of this paper is to describe a methodology to identify significant corners as local features that can be used for the recognition of partially occluded two dimensional objects.

The first stage is the detection of edges using a gradient operator.

The edges were thinned using Hilditch's algorithm and then chain-coded.

A corner detection algorithm was implemented based on the ideas of Beus-Tiu.

The most significant corners of a number of two dimensional objects were then stored in a database designed for ease of matching for recognition of partially visible objects.

A number of experiments were made to illustrate the effectiveness of the approach proposed in this paper.