Automatic Robust Designs of Template Parameters for a Type of Uncoupled Cellular Neural Networks
This paper proposes a robust design theorem for a kind of uncoupled cellular neural/nonlinear networks (CNNs) in aim to process images with some specific image properties. A novel algorithm scheme is introduced to automatically determine the required imag
- PDF / 713,819 Bytes
- 14 Pages / 439.37 x 666.142 pts Page_size
- 91 Downloads / 148 Views
Abstract This paper proposes a robust design theorem for a kind of uncoupled cellular neural/nonlinear networks (CNNs) in aim to process images with some specific image properties. A novel algorithm scheme is introduced to automatically determine the required image parameters given in the theorem, and implement image processing. Using the proposed theorem designs a convex corner detection, CNN template, and some edge detection CNN templates. Four image processing examples are given. Comparing the edge detection algorithms ‘‘Sobel,’’ ‘‘Prewitt,’’ ‘‘Roberts,’’ and ‘‘Canny’’ with our algorithm shows that the presented algorithm has promising results.
Keywords Cellular neural network Robust design detection Convex corner detection Edge detection
Automatical image
1 Introduction The CNN was first introduced by Chua and Yang [3, 4] in 1988. The original motivation of CNNs production is to find out a more practical structure of neural network to replace the theoretically existent Hopfield neural networks [1]. CNN has been widely studied for theoretical foundations and practical applications in M. Zhang (&) L. Min X. Zhang Schools of Mathematics and Physics, University of Science and Technology Beijing, Beijing 100083, China e-mail: [email protected] L. Min e-mail: [email protected] X. Zhang e-mail: [email protected]
Z. Wen and T. Li (eds.), Foundations of Intelligent Systems, Advances in Intelligent Systems and Computing 277, DOI: 10.1007/978-3-642-54924-3_55, Springer-Verlag Berlin Heidelberg 2014
577
578
M. Zhang et al.
image and video signal processing, robotic and biological visions, and higher brain functions [1, 2, 5, 20, 21]. During the past 20 years, the stability of standard CNNs and their generalizations has been widely studied and attracted many researchers’ attention [1, 2, 4, 18]. Practically, the parameter levers in an analog CNN usually have 5–10 % of perturbation [7]. The robustness design is an important issue for the study in CNNs. Chua and Dogaru [1, 5] have studied the robustness of a kind of uncoupled Boolean CNN, which has provided optimal design scheme for CNN with prescribed tasks. Lately, some robust designs for uncoupled and coupled CNNs have been studied. Su and Min [19] have designed a robust CNN which can directionally enhance gray levels of pixels in images. Liu and Min [13] have studied the robust design of global connectivity detection CNN for binary images [1], and then extended the result to gray level images [8]. Li et al. [9] and [10] have introduced robust edge detection CNN and corner detection CNN, respectively. Jian et al. [8] have studied the robust designs for dilation CNN and erosion CNN. Zhao et al. [23] have designed robust template for pattern matching CNN. Li et al. [11, 12] have studied the robust design for extracting closed domain CNN and shadow projection CNN. Liu et al. [14, 16] have proposed the robust CNN templates with the functions for selected object extractions and directional object extractions. Robust design of bipolar wa
Data Loading...
