Önem Tahminleme Tabanlı Tek Sınıf Sınıflayıcı ile Doku Tanıma

Tek Sınıf Sınıflayıcı (TSS) ile doku tanıma, farklı sınıflara ait dokular içinde sadece ilgilenilen hedef dokuya ilişkin eğitim verileri ile hedef dokunun tanınması problemidir. Bu çalışmada, önem değerinin tahminlenmesine dayanan bir doku sınıflayıcı önerilmiştir. Önem değeri, test ve eğitim verilerinden elde edilen yoğunluk fonksiyonlarının oranından hesap edilmiştir. Girilen test verisi ile hedef sınıf verisi birbirine benzer ise önem değeri bire yakın olduğundan elde edilen önem değerlerine eşikleme işlemi uygulanarak sınıflama işlemi gerçeklenmiştir. Önem değeri, Kısıtsız En Küçük Kareler Önem Uydurma (KEKK-ÖU) yöntemi ile elde edilmiştir. Önerilen yöntemin etkinliği, farklı doku setleri üzerinde farklı sınıflama ölçütleri ile değerlendirilmiştir. Elde edilen sonuçlar önerilen yöntemin TSS problemlerinde başarılı ve güvenilir olduğunu göstermektedir. Sonuçlar, ayrıca literatürde referans yöntem olarak alınan tek-sınıf destek vektör makinaları yöntemine göre de karşılaştırılmış ve önerilen yöntem ile daha yüksek TSS başarım performansı elde edilmiştir

Texture Recognition Using Importance Based One-Class Classifier

Texture recognition by Single-Class Classifier (SCC) refers to the problem of recognizing the target texture by using only the training data pertaining to the texture under concern. In this study, a texture classifier, based on the estimation of importance, is proposed. Importance is calculated as the ratio of two density functions obtained from the training and test data. Since the importance value is close to the one if the inputted test and target class data are similar to each other, classification is performed by appliying a thresholding process to the obtained importance values. Importance is estimated by Unconstrained Least Square Importance Fitting (uLSIF) algorithm. The effectiveness of the proposed method is examined on different texture sets with different classification metrics. Our results show that the proposed algorithm is powerful and reliable in SCC problems. Results are also compared with the one-class support vector machines which is the reference algorithm in the literature and higher SCC performance is obtained with the proposed method for applied textures

PDF

___

[1] Tax, D. M.J. 2001. One-Class Classification. TU Delft, Delft University of Technology.
[2] Maglaras, L. A., Jiang, J., Cruz, T. 2014. Integrated OCSVM mechanism for intrusion detection in SCADA systems. Electronics Letters, Cilt: 50(25), s. 1935-1936.
[3] Guerbai, Y., Chibani, Y., Hadjadji, B., 2015. The effective use of the oneclass SVM classifier for handwritten signature verification based on writer-independent parameters, Pattern Recognition, Cilt 48, s. 103- 113.
[4] Guillermo, L.G. , Lucas, C.U., Pablo M.G. 2013. Abrupt Change Detection with One-Class Time-Adaptive Support Vector Machines, Expert Systems with Applications, Cilt 40, s. 7242- 7249.
[5] Hyun, J.S., Dong-Hwan, E., Sung-Shick, K., 2005. One-class Support Vector Machines—An Application In Machine Fault Detection and Classification, Computers and Industrial Engineering, Cilt 48, s. 395-408.
[6] Ciesielski, V., Phuong Ha V. 2009, Texture Detection Using Neural Networks Trained on Examples of One Class, s.140-149, Nicholson, A., Li, X. ed. 2009, LNAI 5866, SpringerVerlag Berlin Heidelberg.
[7] Sanchez-Yanez, R.E., Kurmyshev, E.V., Fernandez, A. 2003. One-class texture classifier in the CCR feature space, Pattern Recognition Letters Cilt. 24, s. 1503–1511
[8] Chen, C.H. (Ed.), 2015. Handbook of pattern recognition and computer vision. World Scientific.
[9] Demir, G.K. 2017, A Spectral Graph Theoretical Approach to Oriented Energy Features, International Journal of Pattern Recognition and Artificial Intelligence, Cilt. 31, s. 1755001_1 – 1755001_27.
[10] Depeursinge, A., Püspöki, Z., Ward, J. P., & Unser, M. 2017. Steerable wavelet machines (SWM): learning moving frames for texture classification. IEEE Transactions on Image Processing, Cilt 26, s. 1626- 1636.
[11] Scholkopf, J., Platt, J., Shawe-Taylor, J., Smola, A., Williamson, R. 2001, Estimating the Support of a HighDimensional Distribution. Neural Computation, Cilt 13, s.1443–1471.
[12]Erfani, S. M., Rajasegarar, S., Karunasekera, S., & Leckie, C. 2016. High-dimensional and large-scale anomaly detection using a linear oneclass SVM with deep learning. Pattern Recognition, Cilt 58, s. 121-134.
[13] Gondra, I., Heisterkamp, D., Peng, J. 2004, Improving Image Retrieval Performance by Inter-query Learning with One-class Support Vector Machines, Neural Computing and Applications, Cilt 13, s. 130–139
[14] Breunig, M.M., Kriegel,H.P., Ng,R.T., Sander J. 2000, Lof: Identifying Density-Based Local Outliers, ACM Sigmod Record, Cilt 29, s. 93–104.
[15].Hempstalk, K., Frank, E., Witten, I.H., 2008, One-Class Classification by Combining Density and Class Probability Estimation, Daelemans, W., Goethals, B., Morik, K. ed. Machine Learning and Knowledge Discovery in Databases. ECML PKDD 2008. Lecture Notes in Computer Science, Cilt 5211, Springer, Berlin, Heidelberg.
[16] Kanamori, T., Hido, S., Sugiyama, M. 2009, A least-squares approach to direct importance estimation, Machine Learning Research, Cilt 10, s.1391-1445.
[17] Haralick,R.M., Shanmugam, K., Dinstein, I. 1973, Textural Features for Image Classification, IEEE Transactions on Systems, Man and Cybernetics, Cilt 3, s. 610–621.
[18] Hido, S., Tsuboi, Y., Kashima, H., Sugiyama, M., Kanamori, T. 2011, Statistical Outlier Detection Using Direct Density Ratio Estimation, Knowledge and Information Systems, Cilt 26, s. 309–336.
[19] Liu, S., Yamada, M., Collier, N., & Sugiyama, M. 2013. Change-point detection in time-series data by relative density-ratio estimation. Neural Networks, Cilt 43, s. 72-83.
[20] Balzi, A., Yger, F., & Sugiyama, M. 2015. Importance-weighted covariance estimation for robust common spatial pattern. Pattern Recognition Letters, Cilt 68, s. 139- 145.
[21] Brodatz, P. 1966, Textures: A Photographic Album for Artists and Designers, Dover, Newyork.
[22] Otsu, N. 1979, A Threshold Selection Method from Gray-Level Histograms, IEEE Transaction on Systems Man and Cybernetics, Cilt. 9, s. 62–66.