In this thesis, we have addressed on the applications of cellular neural networks (CNN’s) in high-order image processing. We start with introducing the fundamentals and design guidelines of CNN and GA to deduce the later architecture that we proposed in this thesis. For using CNN’s in some specific image application, we use genetic algorithm (GA) to optimize a better CNN template. The major reason that we used GA as an analytical training tool for our issuing problems lies in the prevailing applied field and evolution mechanism of GA. As we know, search of CNN templates could be sometimes too complicated to follow the rules in higher-order image processing. Thus, we introduced some basic but significant CNN template examples at the first place so as to inspire more applications based on CNN’s by observing the nature of images. Most significantly, we make use of CNN’s to deal with some difficult issues of image processing, image descreening in particular, which has been regarded as a tough question since no single solution can be applied to all cases.
However in this thesis, we classify the original screening images in two classes beforehand by our defined index, and a different simple filter would be applied to each of these two classes. In this application, CNN plays an important role in classification of images by looking upon screening patterns as different textures. The descreened images hence could be more satisfactory and acceptable to human perception.
What’s more, we focus on texture discrimination and representation for more image applications related to texture analysis. In this aspect, we proposed a brand-new structure to proliferate CNN templates, which makes decision of CNN templates more flexible. We do not have to assume how many templates to be optimized at the beginning of texture classification, and instead we adaptively determine the number of
optimized templates for the characteristics of the texture set. CNN can be no doubt used here on account of its filtering capability and attraction of hardware implementability. To characterized CNN more sensitively in various kinds of texture patterns, we also presented a series of features, Transition Bit String (TBS) defined in this thesis, for more practical uses. Also, our feature series could not be necessarily carried out by CNN. To be more organized, we have the following major contributions in this thesis.
A. We propose a systematical method for image descreening.
B. We inspire an idea of screening pattern classification in advance for image descreening, which makes image documentation simpler and more impressive.
C. We present a proliferation structure to determine CNN templates in a more flexible fashion.
D. We innovate a kind of characterizing features for texture analysis.
E. We give more possibilities of hardware implementations in high-order image processing.
To sum up, this thesis not only provides an approach in the recent state-of-the-art but also gives more chances and perspectives for hardware implemented in high-order image applications. In addition, the indices and features proposed in this thesis would bring about more solutions in the complicated problems of image processing. We shall at this place list some researching fields that might be carried out in the future.
A. Hardware implementation
The main content of this thesis would like to apply the CNN structure to more applications in pursuit of future hardware implementation or chip design.
Therefore, this future work might be the most important and ultimate issue among many kinds of applications inspired by this thesis.
B. Augment of more CNN channels
CNN used to apply to applications of image processing only in gray-scale, which might cause the loss of some important information of original images. It is certainly believed that more channels of CNN implementation could do a great help for some specific applications of images that should depend on the color information of original images. So the implementation of more CNN channels might provide more solutions for more complicated problems of image processing.
C. Feature selection of CNN proliferation structure
In this thesis, we only take into consideration the generative way of extracted features for our CNN proliferation structure due to the augmented properties of TBS feature series. Of course, for the concept of feature selection, how to reduce the number of features might also be useful in pattern recognition. The consideration of how to select useful features could be put forth in the future.
D. More applications or generalization of TBS
We have presented the basic definition and formula of TBS in this thesis, and TBS series have also been proven to be useful in representation of various texture patterns. As what we have concluded earlier, TBS series would not necessarily be applied to CNN outputs, and these feature curves could be quite useful for analyzing image data in software-oriented applications. More specifically, TBS series could be expressed in various cases of requirements of images. For example, the setting of threshold, the scanning directions, the way of counting pixels, and so on would provide more flexible tuning parameters for more versatile image applications.
E. More applications of image processing by CNN
No doubt, more applications of image processing are the major goals that this
thesis pursues. This thesis has offered more way of using CNN output and illustrated the relationship between CNN structure and image data. It is observed that the results convolved from CNN outputs could also be expressed in terms of various sorts of definitions for images. More complicated applications of image processing could be consequently carried out in the future by our proposed feature series as well as the constructed structure.
In this section, we have concluded the major contributions and the possible future works in details. Hence, we would be dedicated to solving more and more issues of digital image processing by our proposed strategies using CNN in the future.
Bibliography
[1] L. O. Chua and T. Roska, “Stability of a class of nonreciprocal cellular neural networks,” IEEE Trans. Circuits Syst., vol. 37, pp. 1520-1527, 1990.
[2] F. Zou and J. A. Nossek, “Stability of cellular neural networks with opposite-sign templates,” IEEE Trans. on Circuits Syst., vol. 38,pp. 675–677, June 1991.
[3] I. Fajfar and F. Bratkovi¢c, “Design of monotonic binary-valued cellular neural networks,” in Proc. IEEE Int.Workshop Cellular Neural Networks Applications, Sevilla, Spain, pp. 321–326, June 1996.
[4] B. Mirzai, D. L&&ym, and G. S. Moschytz, “On the robust design of uncoupled CNN’s,” in Proc. European Symp. Artificial Neural Networks, Bruges, Belgium, pp. 297–302, Apr. 1998.
[5] J. A. Nossek, “Design and learning with cellular neural networks,” in Proc. IEEE Int.Workshop Cellular Neural Networks Applications, Rome, Italy, pp. 137–146, Dec. 1994.
[6] C. G&& uzelis and S. Karamahmut, “Recurrent perceptron learning algorithm for completely stable cellular neural networks,” in Proc. IEEE Int. Workshop Cellular Neural Networks Applications, Rome, Italy, pp. 177–182, Dec. 1994.
[7] T. Kozek, T. Roska, and L. O. Chua, “Genetic algorithm for CNN template learning,” IEEE Trans. Circuits Syst. I, vol. 40, pp. 392–402, June 1993.
[8] B. Chandler, C. Rekeczky, Y. Nishio, and A. Ushida, “Using adaptive simulated annealing in CNN template learning—A powerful alternative to genetic algorithms,” in Proc. European Conf. Circuit Theory Design, Budapest, Hungary, pp. 655–660, Sept. 1997.
[9] G. Seiler, A. J. Schuler, and J. A. Nossek, “Design of robust cellular neural networks,” IEEE Trans. Circuits Syst. I, vol. 40, pp. 358–364, Mar. 1993.
[10] I. Fajfar and F. Bratkovi˘c, “Optimizing cellular neural networks for robustness and speed,” in Proc. European Conf. Circuit Theory Design, Istanbul, Turkey, Aug. pp. 781–784, 1995.
[11] M. H¨anggi and G. S. Moschytz, “Genetic optimization of cellular neural networks,” in Proc. IEEE Int. Conf. Evolutionary Computation, Anchorage, AK, pp. 381–386, 1998.
[12] Hanggi, M. and Moschytz, G.S., “Stochastic and hybrid approaches toward robust templates,” in Proc. IEEE Int. Workshop Cellular Neural Networks Applications, London, U.K., Apr. pp. 366–371, 1998.
[13] L. O. Chua and L. Yang, “Cellular neural networks: Applications,” IEEE Trans.
Circuits Syst. I, vol. 35, pp. 1273–1290, Oct. 1988.
[14] J. K. Hawkins, “Textural properties for pattern recognition,” Picture Procesping and Psychopictorics, Bernic Sacks Lipkm and Azriel Rosenfeld (EDS). New York:
Academic Press, 1969.
[15] J. Serra, “Theoretical bases of the Leitz texture analyses system,” Leifz Sci. Tech.
Inform., Supplement 1, 4, pp. 125-136, Apr. 1974 (Wetzlar, Germany).
[16] G. Matheron, “Elements Pour Une Theorie des Milieux Poreu,” Paris, France:
Masson, 1967.
[17] J. Weszka, C. Dyer, and A. Rosenfeld, “A comparative study of texture measures for terrain classification,” IEEE Trans. on Sys., Man , and Cyber., vol. SMC-6, no. 4, pp. 269-285, Apr. 1976.
[18] D.E. Goldberg, Genetic Algorithms in Search, Optimization and Machine Learning, Addison-Wesley, Reading, Mass., 1989.
[19] L. Davis, ed., Handbook of Genetic Algorithms, Van Nostrand Reinhold, New York, 1991.
[20] Z. Michalewicz, Genetic Algorithms + Data Structures = Evolutionary Programs, Springer-Verlag, Berlin, 1992.
[21] J.H. Holland, Adaptation in Natural and Artificial Systems, Univ. of Michigan Press, Ann Arbor, Mich., 1975.
[22] R. R. Brooks, Lynne Grewe and S. S. Iyengar, “Recognition in the wavelet domain: A survey”, Journal of Electronic Imaging, pp. 757-784, July 2001.
[23] L.-K. Shark and C. Yu, “Denoising by optimal fuzzy thresholding in wavelet domain,” Electronics Letters, Vol. 36, pp. 581 –582, March 2000.
[24] R.D. Nowak and R.G. Baraniuk, “Wavelet-domain filtering for photon imaging systems,” IEEE Trans. on Image Processing, Vol. 8, pp. 666 –678, May 1999.
[25] D. Margulis, Mathematics, Morie, and the Artist, CATFWorld, Sep/Oct, 1997.
[26] R. Ulichney, Digitial halftoning , MIT press, Cambridge, Massachusetts, 1987.
[27] P. W. Wong, “Inverse halftoning and kernel estimation for error diffusion,” IEEE Trans. Image Processing, vol. 4, pp. 486–498, Apr. 1995.
[28] S. Hein and A. Zakhor, “Halftone to continuous-tone conversion of error-diffusion coded images,” IEEE Trans. Image Processing, vol. 4, pp.
208–216, Feb. 1995.
[29] D. C. Youla and H. Webb, “Image restoration by the method of convex projections—Part I: Theory,” IEEE Trans. Med. Imag. vol. MI-1, pp. 81–94, Oct.
1982.
[30] D. F. Dunn, T. P. Weldon and W. E. Higgins, “Extracting halftones from printed documents using texture analysis,” Optical Engineering, vol 36, no 4, pp.
1044-1052, 1997.
[31] A. Jaimes, F. Mintzer, A.R. Rao, and G. Thompson, “ Segmentaion and automatic descreening of scanned documents,” Spie, vol. 3648, pp. 517-528, 1999.
[32] J. J. Hopfield, “Neural networks and physical systems with emergent computational abilities,” Proc. Natl. Acad. Sci. USA., vol. 79, pp. 2554-2558, 1982.
[33] J. J. Hopfield and D. W. Tank, “Neural computation of decisions in optimization problems,” Bio. Cybern., vol. 52, pp. 141-152, 1985.
[34] J. J. Hopfield and D. W. Tank, “Computing with neural circuits: a model,”
Science (USA), vol. 233, no. 4764, pp. 625-633, 1986.
[35] D. W. Tank and J. J. Hopfield, “Simple ‘neuron’ optimization networks: An A/D converter, signal decision circuit, and a linear programming circuit,” IEEE Tran.
Circuits Syst., CAS-33, pp. 533-541, 1986.
[36] F. Zou, S. Schwarz, and J. A. Nossek, “Cellular neural network design using a learning algorithm,” in Proc. IEEE Int. Workshop on Cellular Neural Networks and Their Applications, pp. 73-81, 1990.
[37] S. Schwarz and W. Mathis, “A design algorithm for cellular neural networks,” in Proc. 2nd Int. Conf. Microelectronics for neural network, pp. 53-59, 1991.
[38] H. Harrer. “Multiple layer discrete time cellular neural networks using time variant templates,” IEEE TCAS-II, vol 40, n.3, pp. 191-199, 1993.
[39] P. Lopez, D.L. Vilarino and D. Cabello, “Design of multilayer discrete time cellular neural networks for image processing tasks based on genetic algorithms,” ISCAS 2000, n.4, pp. 133-136, 2000.
[40] P. Lopez, M. Balsi, D.L. Vilarino and D. Cabello, “Design and training of multilayer discrete time cellular neural networks for antipersonnel mine detection using genetic algorithms,” in Proc. 6th IEEE Int. Workshop on Cellular Neural Networks and Their Applications, pp. 363-368, 2000.
[41] D.L. Vilarino, D. Cabello, M. Balsi and V. M. Brea, “Image segmentation based on active contours using discrete time cellular neural networks,” in Proc. fifth IEEE Int. Workshop on Cellular Neural Networks and Their Applications, pp.
331-336, 1998.
[42] P. Lopez, D.L. Vilarino and D. Cabello, “Genetic algorithm based training for multilayer discrete time cellular neural networks,” IWANN’99. Lecture Notes in Computer Science, vol. 1607, pp. 467-476, Springer-Verlag, 1999.
[43] Tamas Sziranyi and Marton Csapodi, “Texture classification and segmentation by cellular neural networks using genetic algorithms,” Computer Vision and Image Understanding, vol. 71, No. 3, Sep, pp. 255-270, 1998.
[44] M. Levine, Vision in Man and Machine, McGraw-Hill, 1985.
[45] R. Haralick, “Statistical and structural approaches to texture,” Proc. IEEE, vol.
67, no. 5, pp.786- 804, 1979.
[46] G. Cross and A. Jain, “Markov random field texture models,” IEEE Trans.
Pattern Analysis and Machine Intelligence, vol. 5, no. 1, pp. 25-39, 1983.
[47] A. Pentland, “Fractal-based description of natural scenes,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 6, no. 6, pp. 661-674, 1984.
[48] R. Chellappa and S. Chatterjee, “Classification of textures using gaussian markov random fields,” IEEE Trans. Acoustic, Speech and Signal Processing, vol. 33, no.
4, pp. 959-963, 1985.
[49] H. Derin and H. Elliot, “Modeling and segmentation of noisy and textured images using Gibbs random fields,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 9, no. 1, pp. 39-55, 1987.
[50] B. Manjunath and R. Chellappa, “Unsupervised texture segmentation using markov random fields,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 13, no. 5, pp. 478-482, 1991.
[51] A. Rosenfeld and J. Weszka, “Picture recognition” in Digital Pattern Recognition, K. Fu (Ed.), Springer-Verlag, pp. 135-166, 1980.
[52] J. Daugman, “Uncertainty relation for resolution in space, spatial frequency and orientation optimised by two-dimensional visual cortical filters,” Journal of the Optical Society of America, vol. 2, pp. 1160-1169, 1985.
[53] A. Bovik, M. Clark, and W. Giesler, “Multichannel texture analysis using localised spatial filters,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 12, pp. 55-73, 1990.
[54] S. Mallat, “Multifrequency channel decomposition of images and wavelet models,” IEEE Trans. Acoustic, Speech and Signal Processing, vol. 37, no. 12, pp. 2091-2110, 1989.
[55] A. Laine and J. Fan, “Texture classification by wavelet packet signatures,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol. 15, no. 11, pp. 1186-1191, 1993.
[56] C. Lu, P. Chung, and C. Chen, “Unsupervised texture segmentation via wavelet transform,” Pattern Recognition, vol. 30, no. 5, pp. 729-742, 1997.
[57] B. Mirzai, Z. Cheng, and G. S. Moschytz, “Learning algorithms for cellular neural networks,” in Proc. IEEE Int. Symp. Circuits Systems, Monterey, CA, vol.
3, pp. 159–162, June 1998.
[58] L. O. Chua and P. Thiran, “An analytic method for designing simple cellular neural networks,” IEEE Trans. Circuits Syst. I, vol. 38, pp. 1332–1341, Nov.
1991.
[59] M. Hanggi and G. S. Moschytz, “An exact and direct analytical method for the design of optimally robust CNN templates,” IEEE Trans Circuits Syst. I, vol. 46,
no. 2, pp. 304-311, Feb. 1999.
[60] Tamas Sziranyi and Attila Hanis, “Sub-pattern texture recognition using intelligent focal-plane imaging sensor of small window-size,” Pattern Recognition Letters, pp. 1133-1139, 1999.
[61] Tamas Sziranyi and Josiane Zerubia, “Markov random field image segmentation using cellular neural network,” IEEE CAS-I: Fundamental Theory and Applications, vol. 44, no.1, 1997.
[62] Tamas Sziranyi, “Texture segmentation by the 64X64 CNN chip,” in Proc. IEEE Int.Workshop Cellular Neural Networks Applications, pp. 547-554, 2000.
[63] Tamas Sziranyi and Jozsef Csicsvari, “High-speed character recognition using a dual cellular neural network architecture (CNND),” IEEE CAS-II: Analog and Digital Signal Processing, vol. 40, no. 3, March 1993.
[64] John P. Miller, Tamas Roska, Tamas Sziranyi, Kenneth R. Crounse, Leon O.
Chua, and Laszlo Nemes, “Deblurring of images by cellular neural networks with applications to microscopy,” in Proc. Third IEEE Int.Workshop Cellular Neural Networks Applications, pp. 237-242, 1994.
[65] Tamas Sziranyi, “Texture recognition using a superfast cellular neural network VLSI chip in a real experimental environment,” Pattern Recognition Letters, pp.
1329-1334, 1997.
[66] L. O. Chua and L. Yang, “Cellular neural networks: Theory,” IEEE Trans.
Circuits Syst., vol. 35, pp. 1257-1272, 1988.
[67] Y.-W. Shou and C. T. Lin, “Image descreening by GA-CNN-Based Texture Classification,” IEEE Trans. on circuits and system I, vol. 51, no.11, pp.
2287-2299, 2004.
[68] Leon O. Chua, CNN: A paradigm for complexity, World scientific, 1998.