Occluded objects recognition using multiscale features and Hopfield neural network

Jiann Shu Lee, Chin-Hsing Chen, Yung-Nien Sun, Guan Shu Tseng

Research output: Contribution to journalArticle

10 Citations (Scopus)

Abstract

A new method to recognize partially visible two-dimensional objects by means of multiscale features and Hopfield neural network was proposed. The Hopfield network was employed to perform global feature matching. Since the network only guarantees to converge to a local optimal state, the matching results heavily depend on the initial network state determined by the extracted features. To acquire more satisfactory initial matching results, a new feature vector was developed which consists of the multiscale evolution of the extremal position and magnitude of the wavelet transformed contour orientation. These features can even be used to discriminate dominant points, hence good initial states can be obtained. The good initiation enables our proposed method to recognize objects even heavily occluded, that cannot be achieved by using the Nasrabadi-Li's method. In addition, to make the matching results more insensitive to the threshold value selection of the network, we replace the step-like thresholding function by a ramp-like one. Experimental results have shown that our method is effective even for noisy occluded objects.

Original languageEnglish
Pages (from-to)113-122
Number of pages10
JournalPattern Recognition
Volume30
Issue number1
Publication statusPublished - 1997 Jan 1

Fingerprint

Hopfield neural networks
Object recognition

All Science Journal Classification (ASJC) codes

  • Software
  • Signal Processing
  • Computer Vision and Pattern Recognition
  • Artificial Intelligence

Cite this

@article{dbcc3e9e864a4cf28b31232aae8716e7,
title = "Occluded objects recognition using multiscale features and Hopfield neural network",
abstract = "A new method to recognize partially visible two-dimensional objects by means of multiscale features and Hopfield neural network was proposed. The Hopfield network was employed to perform global feature matching. Since the network only guarantees to converge to a local optimal state, the matching results heavily depend on the initial network state determined by the extracted features. To acquire more satisfactory initial matching results, a new feature vector was developed which consists of the multiscale evolution of the extremal position and magnitude of the wavelet transformed contour orientation. These features can even be used to discriminate dominant points, hence good initial states can be obtained. The good initiation enables our proposed method to recognize objects even heavily occluded, that cannot be achieved by using the Nasrabadi-Li's method. In addition, to make the matching results more insensitive to the threshold value selection of the network, we replace the step-like thresholding function by a ramp-like one. Experimental results have shown that our method is effective even for noisy occluded objects.",
author = "Lee, {Jiann Shu} and Chin-Hsing Chen and Yung-Nien Sun and Tseng, {Guan Shu}",
year = "1997",
month = "1",
day = "1",
language = "English",
volume = "30",
pages = "113--122",
journal = "Pattern Recognition",
issn = "0031-3203",
publisher = "Elsevier Limited",
number = "1",

}

Occluded objects recognition using multiscale features and Hopfield neural network. / Lee, Jiann Shu; Chen, Chin-Hsing; Sun, Yung-Nien; Tseng, Guan Shu.

In: Pattern Recognition, Vol. 30, No. 1, 01.01.1997, p. 113-122.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Occluded objects recognition using multiscale features and Hopfield neural network

AU - Lee, Jiann Shu

AU - Chen, Chin-Hsing

AU - Sun, Yung-Nien

AU - Tseng, Guan Shu

PY - 1997/1/1

Y1 - 1997/1/1

N2 - A new method to recognize partially visible two-dimensional objects by means of multiscale features and Hopfield neural network was proposed. The Hopfield network was employed to perform global feature matching. Since the network only guarantees to converge to a local optimal state, the matching results heavily depend on the initial network state determined by the extracted features. To acquire more satisfactory initial matching results, a new feature vector was developed which consists of the multiscale evolution of the extremal position and magnitude of the wavelet transformed contour orientation. These features can even be used to discriminate dominant points, hence good initial states can be obtained. The good initiation enables our proposed method to recognize objects even heavily occluded, that cannot be achieved by using the Nasrabadi-Li's method. In addition, to make the matching results more insensitive to the threshold value selection of the network, we replace the step-like thresholding function by a ramp-like one. Experimental results have shown that our method is effective even for noisy occluded objects.

AB - A new method to recognize partially visible two-dimensional objects by means of multiscale features and Hopfield neural network was proposed. The Hopfield network was employed to perform global feature matching. Since the network only guarantees to converge to a local optimal state, the matching results heavily depend on the initial network state determined by the extracted features. To acquire more satisfactory initial matching results, a new feature vector was developed which consists of the multiscale evolution of the extremal position and magnitude of the wavelet transformed contour orientation. These features can even be used to discriminate dominant points, hence good initial states can be obtained. The good initiation enables our proposed method to recognize objects even heavily occluded, that cannot be achieved by using the Nasrabadi-Li's method. In addition, to make the matching results more insensitive to the threshold value selection of the network, we replace the step-like thresholding function by a ramp-like one. Experimental results have shown that our method is effective even for noisy occluded objects.

UR - http://www.scopus.com/inward/record.url?scp=0030834665&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0030834665&partnerID=8YFLogxK

M3 - Article

VL - 30

SP - 113

EP - 122

JO - Pattern Recognition

JF - Pattern Recognition

SN - 0031-3203

IS - 1

ER -