Recognition of 2D Object Contours Using Starting-Point-Independent Wavelet Coefficient Matching

✍ Scribed by Hee Soo Yang; Sang Uk Lee; Kyoung Mu Lee

• Publisher: Elsevier Science
• Year: 1998
• Tongue: English
• Weight: 369 KB
• Volume: 9
• Category: Article
• ISSN: 1047-3203
• DOI: 10.1006/jvci.1998.0384

✦ Synopsis

profiles, the object contours, extracted from images, are known to be quite useful for object recognition [2]. In this In this paper, a new recognition algorithm for 2D object contours, based on the decimated wavelet transform, is prepaper, we will focus our attention to the case where an sented, emphasizing the starting point dependency problem. object can be only described by contour.

The proposed matching algorithm consists of two parts: Firstly,

There are many techniques available to describe 2D we present new data structures for the decimated wavelet repreobjects with contours, and the Fourier descriptor is a widely sentation and a searching algorithm to estimate the misalignused shape representation technique [3]. However, the ment between the starting points for the reference model and Fourier descriptor has several shortcomings in shape repreunknown object. We also adopt a polynomial approximation sentation. One of the disadvantages is that it can not protechnique and propose a fast searching algorithm. And then, vide multiresolutional representations of a shape. Since all matching is performed in an aligned condition on the multiresodescriptions and matching procedures are carried out at lutional wavelet representation. By employing a variable-rate one fixed resolution, the Fourier descriptor based apdecimation scheme, we can achieve fast and accurate recognition results, even in the presence of heavy noise. We provide proaches lead to relatively poor accuracy and high compuan analysis on the computational complexity, showing that our tational complexity.

approach requires only less than 25% of the computational load In many computer vision tasks, in order to improve the required for the conventional method [1]. Various experimental accuracy and robustness to the noise, multiresolutional results on both synthetic and real imagery are presented to analysis is preferred. The wavelet representation offers the demonstrate the performance of the proposed algorithm. The global shape information at coarse resolutions and the simulation results show that the proposed algorithm successlocal detail features at fine resolutions. Since the wavelet fully estimates the misalignment and classifies 2D object contransform provides natural multiresolution representatours, even for the input SNR ‫؍‬ 5 dB.