Control Systems and Computers

DOI: https://doi.org/10.15407/usim.2017.01.059

Upr. sist. maš., 2017, Issue 1 (267), pp. 59-67.

UDC 004.89 + 004.932

R.O. Tkachenko ¹, P.R. Tkachenko ², I.V. Іzonіn ³, D.A. Batyuk ⁴

Methods of Image Pre-Processing Based on Neuro-Paradigm of Geometric Transformation Model

1 – Doctor of Engineering Science, L’viv polytechnic National University, Address: 12 Bandera str., Main building, Room 202, L’viv, Ukraine, E-mail:

2 –  PhD of Eng. Sc., Lviv Educational-Scientific Institute of the Higher Educational Institution “University of Banking Affairs” (Lviv),

3 –  PhD of Eng. Sc., L’viv polytechnic National University, Address: 12 Bandera str., Main building, Room 202, L’viv, Ukraine, E-mail:  ivanizonin@gmail.com

4 – Post graduate student, L’viv polytechnic National University, Address: 12 Bandera str., Main building, Room,  202, L’viv, Ukraine.

Introduction. The task of image preprocessing for the problems of the intellectual analysis become a significant spread in our time.  It is explained by the increasing necessity to apply similar procedures in the areas such as medicine, criminology, video, and more. The realization of problem solution for improving the digital images quality sometimes of the large dimension in online mode and while minimizing the computing resources continues to be very relevant. Similar restrictions required the use of the effective methods and tools for its solution. One possible approach to solve this problem may be the use of the fast and effective machine learning procedures.

Purpose. There are many tools for the machine learning implementation. In this article the  authors use the tools of computational intelligence – artificial neural networks. This apparatus allows the rapid and efficient learning. The use of such tools for solving the problem of improving the quality of digital images is not new. However, the existing methods are based on the classical neural networks have the significant drawbacks. It imposes a number of restrictions.

In the article the  authors use a new paradigm of building artificial neural networks. It is based on the geometric transformation machine. Exactly this advantage is providing the possibility of solution the problem of improving the quality of digital images in online mode.

The authors describe the topology of the neural network of solution to the problem of improving the quality of digital images, the basic steps of the training algorithm. The proposed learning algorithm is different from the existing ones by speed and accuracy, It provides an effective solution of the problem of increasing the quality of the digital images. Also, the authors in detail describe the process of applying trained neural network to solve the problem.

Conclusions. Therefore, in this article a new method of image preprocessing to improve its quality for further intellectual analysis is described. The method is simulated in different images. The estimation of the images quality, using four indicators, is carried out. It is established that the efficiency of the method is the best on one class of images. A comparison of the proposed method with existing ones is conducted. The basic advantages of the developed method for its application in real-time vision systems are described.

Keywords: image resolution, machine learning, neuro-paradigm, Geometric Transformation Model.

Download full text!

1. Peleshko, D., Peleshko, M., Kustra, N., Izonin, I., 2011. “Analysis of invariant moments in tasks image processing”. 11th International Conference the Experience of Designing and Application  of  CAD Systems in Microelectronics (CADSM-2011), Polyana-Svalyava, pp. 263–264.
2. Tkachenko, O., Tkachenko, R., 2013. “Neural System Based on the Geometric Transformation Model”. Proceeding of the Second International Conference on Automatic Control and Information Technology 2013 (ICACIT’13), Cracow, Poland, pp.28–34.
3. Peleshko, D., Kovalchuk, A., Kustra, N., Izonin, I., 2011. “Invariantni momenty v prykladnykh zadachakh obrobky ta analizu zobrazhen”. Visnyk Natsionalnoho universytetu “Lvivska politekhnika”: Komp’iuterni nauky ta informatsiini tekhnolohii, 694, pp. 265–270 (In Ukrainian).
4. Tkachenko, R.O., Izonin, I.V., Batiuk, D.A., Sydorenko, R.V., Prostrelchuk, V.I., 2016. “Peredyskretyzatsiia zobrazhen zasobamy mashynnoho navchannia”. Matematychne ta kompiuterne modeliuvannia. Ser. : Tekhnichni nauky, 13, pp. 176–183 (In Ukrainian).
5. Peleshko, D.D., Izonin, I.V., Pelekh, Iu.M., 2013. “Analiz osnovnykh metodiv zbilshennia rozdilnoi zdatnosti zobrazhen na osnovi tekhnolohii super-resolution”. Zbirnyk naukovykh prats Instytutu problem modeliuvannia v enerhetytsi im. H. Ye. Pukhova. Nats. akad. nauk Ukrainy, In-t probl. modeliuvannia v enerhetytsi im. H. Ye. Pukhova, Kyiv, 67, pp. 162–169 (In Ukrainian).
6. Tkachenko, R., 2013. “Information Models of the Geometric Transformation”. Proceeding of the Second International Conference on Automatic Control and Information Technology 2013 (ICACIT’13), Cracow, Poland, pp.48–53 (In Ukrainian).
7. Tsmots, I.H., Shkodyn, A.V., Izonin, I.V., 2014. “Peredyskretyzatsiia zobrazhen v systemakh stereobachennia na osnovi symetrychnoi matrytsi vidstanei”. Intelektualni systemy pryiniattia rishen i problemy obchysliuvalnoho intelektu: zb. nauk. pr. mizhnar. nauk. konf., 28–31 travnia 2014, Zaliznyi port, Kherson, Vyd-vo KhNTU, pp.343–344 (In Ukrainian).
8. Izonin, I.V., 2015. “Doslidzhennia metodiv obchyslennia eksponent Heldera v zadachakh rekonstruktsii zobrazhen”. Suchasni problemy matematychnoho modeliuvannia ta obchysliuvalnykh metodiv: materialy mizhn. nauk. konf.: tezy dopovidi, Rivne, 19–22 liutoho 2015, Rivne: RVV RDHU, pp.61–62 (In Ukrainian).
9. Tsmots, I., Peleshko, D., Izonin, I., 2014. “Parallel algorithms and VLSI structures for median filtering of images in real time”. International Journal of Advanced Research in Computer Engineering & Technology, 3 (8), pp. 2643–2649.
10. Izonin, I. V., Tkachenko, R. O., Peleshko, D. D., Batiuk, D. A., 2015. “Metod nadvysokoi rozdilnoi zdatnosti zobrazhen na osnovi modeli heomet­rychnykh peretvoren”. Intelektualni systemy pryiniattia rishen i problemy obchysliuvalnoho intelektu: zb. mizhnar. nauk. konf., 25–28 travnia 2015, Zaliznyi port, Kherson: KhNTU, pp. 284–286 (In Ukrainian).
11. Yang, X., Hu, X., Liu, Y. 2012. “Modified Gram-Schmidt orthogonalization of covariance matrix adaptive beamforming based on data preprocessing”. IEEE 11th International Conference on Signal Processing, Beijing, pp.373–377.
    https://doi.org/10.1109/ICoSP.2012.6491678
12. Izonin, I. V., Tkachenko, R. O., Hrytsyk, K. Iu., Tytyk, R. O., 2015. “Do metodu zbilshennia rozdilnoi zdatnosti zobrazhen na osnovi ShNM modeli heometrychnykh peretvoren”. Kompiuterni tekhnolohii ta informatsiina bezpeka: mizhn. nauk.-prakt. konf., 2–3 lypnia 2015, Kirovohradskyi natsionalnyi tekhnichnyi universytet, Kirovohrad, pp.54–55 (In Ukrainian).
13. Dong, Chao, Loy, Chen Change, He, Kaiming, Tang, Xiaoou., 2015. “Image Super–Resolution Using Deep Convolutional Networks”. IEEE Transactions on Pattern Analysis and Machine Intelligence, Preprint, pp.14.
14. Dong, Chao, Loy, Chen Change, He, Kaiming, Tang, Xiaoou., 2014. “Learning a Deep Convolutional Network for Image Super–Resolution”. Computer Vision. Proc. of European conf., Zurich, Switzerland, 6–12 September 2014, Springer, pp.184–199.
    https://doi.org/10.1007/978-3-319-10593-2_13
15. The USC-SIPI Image Database of University of Southern California. http:/sipi.usc.edu/database.

Received 31.01.2017