Browse Prior Art Database

FEATURE POINT DETECTION OF 2-D SHAPES

IP.com Disclosure Number: IPCOM000009447D
Original Publication Date: 1999-Sep-01
Included in the Prior Art Database: 2002-Aug-26
Document File: 4 page(s) / 135K

Publishing Venue

Motorola

Related People

Mikhail Shnaider: AUTHOR [+2]

Abstract

The technology to provide a key to the develop- ment of new methods of visual information indexing and retrieval is the extraction of the feature sets for the description of objects within images. Shape is one of the most important characteristics which allows us to distinguish and identify different objects. The appropriate usage of some properties of shapes, such as corners and their relative posi- tions or, in other words, feature points, can be high- ly beneficial for object similarity estimation that is essential for indexing and retrieval.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 50% of the total text.

Page 1 of 4

Developments Technical 0 M MO-OLA

FEATURE POINT DETECTION OF 2-D SHAPES

by Mikhail Shnaider and Xing Zhang

BACKGROUND OF THE PROBLEM

  The technology to provide a key to the develop- ment of new methods of visual information indexing and retrieval is the extraction of the feature sets for the description of objects within images. Shape is one of the most important characteristics which allows us to distinguish and identify different objects. The appropriate usage of some properties of shapes, such as corners and their relative posi- tions or, in other words, feature points, can be high- ly beneficial for object similarity estimation that is essential for indexing and retrieval.

  Being extremely useful for description of shapes viewed from a given angle, a majority of the exist- ing algorithms for feature point detection fail in the cases when the view angles of the same object are significantly different. In this paper we describe a feature point detection algorithm which overcomes this drawback.

PROPOSED ALGORITHM

  The algorithm for feature point detection con- sists of a number of steps, which are depicted in Figure 1.

Fourier Descriptors

          f Remove Redundant Descriptors

+

Inverse Fourier Transform

+

Interpoint Distance Calculation and HPF

+

Zero-Crossing Detection

Correction

Fig. 1 Feature Point Detection Algoritihm

81 September 1999

0 Motomla, 1°C. ,999

[This page contains 14 pictures or other non-text objects]

Page 2 of 4

MOTOROLA Technical Developments

  Let us assume that there exists an ordered col- lection of points in a two-dimensional space which represent the contour of a shape.

  Initially, the contour of the 2-D shape is expand- ed using the Fourier transform. The result of this expansion is a set of Fourier descriptors. Then, a number of redundant Fourier descriptors are identi- fied and removed. Next, we calculate the inverse Fourier transform using the remaining descriptors. The result of this operation is a shape which closely resembles the input 2-D shape but on the affine invariant plane. The interpoint distance function is

generated and subsequently high-pass filtered (HPF). As a result, we obtain a function which intersects the horizontal axis at a number of points. The points at which this function intersects the hori- zontal axis with a positive slope corresponds to either the feature points or their neighbors. In order to finalize the positions...