landscape image of regional tourism classification using neural network

Landscape Image of Regional Tourism Classification using Neural Network Thai Hoang Le, Computer Science Department, University of Science HCM City - Vietnam, lhthai@fit.hcmus.edu.vn Nguyen Thai Do Nguyen, Math and Computer Science Department, University of Pedagogy HCM City - Vietnam, nguyenndt@math.hcmup.edu.vn Hai Son Tran, Math and Computer Science Department, University of Pedagogy – HCM City - Vietnam, haits@hcmup.edu.vn Abstract— In recent years, pattern classification and image classification have received much attention. Many approaches are suggested to solve these problems such as Neural Network, Support Vector Machine, or K-NN. In this paper, we improve Multi Artificial Neural Network model to apply for Landscape Image of Regional Tourism classification. This model evaluates the reliability of each image space and gives the final classification conclusion. In order to compare this method to others, we setup these methods into a set of 904 landscape image of regional tourism Ha Long, Ha Noi, Nha Trang. The classified results show the feasibility of our improvement model. Keyword: Neural Network, Multi Artificial Neural Network (MANN), Image Classification. I. INTRODUCTION Landscape image of regional tourism classification is a kind of pattern classification, which has large pattern representation space. The k-nearest neighbor (k-NN) decision rule is a common tool in image classification but its sequential implementation is slowly and requires the high calculating costs. Thus using them to apply for landscape image of regional tourism classification is not feasible. While SVM may be errors in the case of the landscape is not in all region tourism, because SVM will classify it into the nearest regional tourism based on the calculation parameters. Therefore, we use Neural Network to apply for landscape image of regional tourism classification. In this paper, we improve the Multi Artificial Neural Network (MANN) model to apply for landscape image classification. Firstly, landscape images projected to difference representation spaces. Secondly, based on one-by-one spaces, landscape images are classified into responsive regional tourism class using a Neural Network called Sub Neural Network (SNN) of MANN. Lastly, we use MANN’s to compose the classified result of all SNN. II. MULTI ARTIFICIAL NEURAL NETWORK IMPROVEMENT APPLY FOR IMAGE CLASSIFICATION Multi Artificial Neural Network (MANN), applying for pattern or image classification with parameters (m,L), has m Sub-Neural Network (SNN) and a global frame (GF) consisting L Component Neural Network (CNN). In particular, m is the number of feature vectors of image, n is the number of feature vector dimensions and L is the number of classes. MANN model has suggested 4 definitions (SNN, GF, Collective vector R, and CNN. [1] Figure 1. MANN (m,L)[1] In this paper we have some improvements below: Improvement 1: All SNN have n input nodes. N is the dimensions of feature vector. It means that all sub presentation space of landscape image has n dimensions and equivalent each other. Besides, all SNN will be the same structure. So the implementation costs will be reduced and MANN (m,L) becomes MANN (m,,n,L). Improvement 2: The local training phase can be done parallel. It means that SNN 1 , SNN 2 … SNN m are trained in the same time, see Fig. 2. Figure 2. Parallel local training III. LANDSCAPE IMAGE OF REGIONAL TOURISM CLASSIFICATION PROBLEM A. Feature Extraction from Image In the above section, we explain the MANN in the general case with parameters (m,n,L) apply for pattern classification. Now we apply improvement MANN model for landscape image of regional tourism classification. In fact this is an experimental setup with (m=4,n=5,L=3). The training image set has 822 images including 201 Ha Long bay images (getting from Internet), 367 Ha Noi images and 254 Nha Trang images (capture by digital camera). The test set has 82 images of Ha Long, Ha Noi, Nha Trang. Because the input of Neural Network is vector data, an image is extract to m feature vectors [2],[8]. In details, the image separate into 4 sub-image based on gray level, see Fig. 3. Firstly, image is extract to background and foreground. Foreground includes the pixels which has higher gray level. Background includes the pixels which has lower gray level. Foreground will be extracted to Fore of Foreground and Back of Foreground based on gray level. Background will be extracted to Fore of Background and Back of Background based on gray level. Figure 3. Image feature extraction Each of sub-images is extract the position of center (upper left (UL), upper right (UR), lower left (LL), lower right (LR) quarter), the ratio of sub-image’s area and use LHC color [3] instead of RGB color. 1 3 0 * 1 * * 2 * 2 1 1 3 3 * 0 0 1 1 3 3 0 0 1 1 6 1 6 tan ( ) ( ) 5 0 0 * 2 0 0 2 .7 6 9 0 1.7518 1 .1 3 0 0 1 .00 0 0 4 .59 0 7 0 .0 6 0 1 0 .00 0 0 0 .0 5 6 5 5 .59 4 3 Y L Y b H a C a b X Y a X Y Y Z b Y Z X Y Z −   = −       =     = +         = −                       = −                     =          R G B                ( 1 ) Thus, an image is featured by 4 vectors has 5 dimensions. For example, Fig. 4 will be extracted to 4 feature vectors in the Table 1. Figure 4. A landscape image Table 1. An image features Sub Image Pos Ratio area L H C 1 LL 0.26 159 0.59 4.53 2 UR 0.33 226.8 0.42 0.92 3 UR 0.33 56.0 0.66 9.15 4 UR 0.08 116.4 0.75 6.61 B. MANN Architecture apply for Landscape Image of Regional Tourism An image is a pattern featured by 4 vectors which have 5 dimensions. Images need to classify into 3 classes (Ha Long, Ha Noi, Nha Trang). So we apply MANN with parameters (m=4,n=5,L=3) for landscape image of regional tourism classification. Thus, MANN model in this case has four SNN(s) and one GF consisting of three CNN(s). The i th (i=1 4) feature vector of an image will be processed by SNN i in order to create the L=3 dimensional output vector of responsive SNN. To join all the k th (k=1 3) element of these output vectors gets the collective vector R k . These collective vectors are the input of CNN(s). The only one output node of CNN is an output node of MANN, see Fig. 5. Figure 5. MANN architecture with (m=4,n=5,L=3) In our implementation uses back-propagation Neural Network which has 3 layers with the transfer function is sigmoid function [4] for SNN and CNN. The number of hidden nodes of SNN i (i=1 4) and CNN j (j=1 3) are experimentally determined from 1 to 10 hidden nodes. Every SNN i has n=5 (the dimensions of feature vector) input nodes and L=3 (the number of classes) output nodes. The k th (k=1 3) output of the SNN i gives the probability of image in the k th class based on the i th feature vector. Every CNN j has m=4 (the number of feature vectors) input nodes and only one output nodes. Input of CNN j is the j th output of all SNN(s). It means that CNN j compose the probability of image in the j th class appraised by all SNN(s). Output of CNN j is the j th output of MANN model. It gives the probability of image in the j th class. It is easy to see that to build MANN model only use Neural Network technology to develop our system. IV. R ESULTS We use the same 904 (include 822 for training and 82 for testing by reference the 10-Fold statically method) images set to classify. We compare our improvement MANN model to selection method (choose only one Sub-Neural Network result), and original MANN. The experimental classified result uses a SNN, original MANN and improvement MANN in the same image database could be seen in the Table 2. Table 2. The experimental classified result Region SNN 1 SNN 2 SNN 3 SNN 4 MANN MANN Improvement Ha Long 13 13 14 12 16 15 Ha Noi 17 21 22 20 24 22 Nha Trang 16 15 10 12 15 22 Total 46 49 46 44 55 59 The above table show the details of classified result based on one by one Sub Neural Network (SNN 1 , SNN 2 , SNN 3 , SNN 4 ), the original model (MANN) and the improvement model. The trend of experiment result show in the Fig. 6. Figure 6. Classified result with different methods It is easy to see that our improvement model has increased the classified result. Besides, the improvement model will have lower implementation costs than original model and can be trained parallel. V. C ONCLUSION In this paper, we have improved Multi Artificial Neural Network (MANN) with parameters (m,n,L) from MANN (m,L), where m is the number of feature vectors of pattern or image, n is the dimensions of the feature vector, and L is the number of classes. This model applies for landscape image classification. MANN model has m Sub-Neural Network SNN i (i=1 m) and a Global Frame (GF) consisting L Components Neural Network CNN j (j=1 L). Each of SNN uses to process the responsive feature vector. Each of CNN uses to combine the responsive element of SNN’s output vector. In fact, the weight coefficients in CNN j are as the reliability coefficients the SNN(s)’ the j th output. So the importance of the feature vectors is identified after the training process. On the other hand, it depends on the image database and the desired classification. To experience the feasibility of improvement MANN model, in this research, we conducted to develop a improvement MANN model with parameters (m=4,n=5,L=3) apply for landscape image of regional tourism classification. The experimental result in the same image database shows that the improvement model increases the classified result more than the selection and original MANN method. VI. R EFERENCES [1] Thai, L., Hai, S. T., Facial Expression Classification Based on Multi Artificial Neural Network, Volume of Extended Abstract, International conference on Advance Computing and Applications, Mar 2010, p. 125-133. [2] Thai, L. Building, Development and Application Some Combination Models of Neural Network (NN), Fuzzy Logic (FL) and Genetics Algorithm (GA), PhD Mathematics Thesis, Natural Science University, HCM City, Vietnam, 2004. [3] Siu-Yeng Cho and Zheru Chi, Genetic Evolution Processing of Data Structure for Image Classification, IEEE Transaction on Knowledge and Data Engineering, Vol 17, No 2, 2005. [4] Bishop, C.: Pattern Recognition and Machine Learning. Springer Press, 2006 [5] Tong, S. and E. Chang, Support vector machine active learning for image retrieval, Proceedings of the ninth ACM international conference on Multimedia, 2001, p. 107-118. Compare of image classification 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 SNN 1 SNN 2 SNN 3 SNN 4 MANN MANN Improve ment Methods [6] Brown, R. and B. Pham, Image Mining and Retrieval Using Hierarchical Support Vector Machines, Proceedings of the 11th International Multimedia Modeling Conference (MMM'05)-Volume 00, 2005, p. 446-451. [7] Ghoshal, A., P. Ircing, and S. Khudanpur, Hidden Markov models for automatic annotation and content-based retrieval of images and video, Proceedings of the 28th annual international ACM SIGIR conference on Research and development in information retrieval, 2005, p. 544-551. [8] Chen, Y. and J.Z. Wang, A region-based fuzzy feature matching approach to content-based image retrieval, Pattern Analysis and Machine Intelligence, IEEE Transactions on, 2002, p. 1252-1267. [9] Hoiem, D., et al., Object-based image retrieval using the statistical structure of images, Computer Vision and Pattern Recognition, CVPR 2004, Proceedings of the IEEE Computer Society Conference on, 2, 2004. . for landscape image of regional tourism classification. In this paper, we improve the Multi Artificial Neural Network (MANN) model to apply for landscape image classification. Firstly, landscape. Network, Multi Artificial Neural Network (MANN), Image Classification. I. INTRODUCTION Landscape image of regional tourism classification is a kind of pattern classification, which has large. Multi Artificial Neural Network model to apply for Landscape Image of Regional Tourism classification. This model evaluates the reliability of each image space and gives the final classification