Targeted Visual Content Recognition Using Multi-Layer Perceptron Neural Network

ISSN (e): 2250 – 3005 || Volume, 06 || Issue, 03||March – 2016 ||
International Journal of Computational Engineering Research (IJCER)
www.ijceronline.com Open Access Journal Page 1
Targeted Visual Content Recognition Using Multi-Layer
Perceptron Neural Network
1
Meer Tayyab Ali Moosavi,2
RafiahTabassum, 3
Perisetti Sravani,4
Panuganti
Devi Mounika,5
Lingam Harish Babu,6
Dr.P.S.Suhasini,7
P.Venkata Ganapathi.
1
D.M.S.S.V.H College of Engineering, Machilipatnam, Andhra Pradesh, India
2
3
4
5
6
Associate Professor, E.C.E dept., D.M.S.S.V.H.C.E, MTM, Andhra Pradesh, India
7
N.I.T (Warangal),Andhra Pradesh, India
I. INTRODUCTION
There are various recognition applications of artificial neural network like pattern recognition, face recognition, traffic sign
recognition, character recognition, etc. Recognition of visual content is a typical application for artificial neural network.
This work aims monument image recognition which will be useful to the tourists. The images obtained may be with the
following variations:
[1] Different lengths.
[2] Different angles.
[3] Any occlusions.
[4]Different illuminations.
ABSTRACT
Visual Content Recognition has become an attractive research oriented field of computer
vision and machine learning for the last few decades. The focus of this work is monument
recognition. Imagesof significant locations captured and maintainedas data bases can be used
by the travelers before visiting the places. They can use images of a famous building to know
the description of the building. In all these applications, the visual content recognition plays a
key role. Humans can learn the contents of the images and quickly identify them by seeing
again. In this paper we present a constructive training algorithm for Multi-Layer Perceptron
Neural Network (MLPNN) applied to a set of targeted object recognition applications. The
target set consists of famous monuments in India for travel guide applications. The training
data set (TDS) consists 3000 images. The Gist features are extracted for the images. These are
given to the neural network during training phase.The mean square error (MSE) on the
training data is computed and used as metric to adjust the weights of the neural network,using
back propagation algorithm. In the constructive learning, if the MSE is less than a predefined
value, the number of hidden neurons is increased. Input patterns are trained incrementally
until all patterns of TDS are presented and learned. The parameters or weights obtained
during the training phase are used in the testing phase, in which new untrained images are
given to the neural network for recognition. If the test image is recognized, the details of the
image will also be displayed. The performance accuracy of this method is found to be 95%
.
Keywords: Visual content recognition, Multi-Layer Perceptron Neural Network, Gist feature.

Targeted Visual Content Recognition Using…
II. WORKING PRINCIPLE
Recognition using artificial neural network include two major parts. They are training phase & testing phase. Humans can
see once an image and can recognize it but artificial neural network requires more and more training for recognition.
Fig1. Block diagram
Training:
Training is done in step by step process as shown in the figure1. In this work images of different locations are taken
5locations (classes) per each class. So 50*5=250.Each image is trained 12 times for accurate training.Thus250*12=3000,
having a training set of 3000 images. Preprocessing is done to each image. Gist feature is extracted for each image and given
to the training unit. Training unit applies the back propagation algorithm on the ANN model. Classification gives the class of
the current image.If the error is less than accuracy (95%), the process is repeated with a new training example.
Testing:
Testing is done in step by step process as shown in the figure1. In this work images of different locations are taken
5locations (classes) per each class. So 20*5=100, having a testing set of 100 images.Test images are preprocessed as that of
training images by gist global feature. After preprocessing image is given to 3 layers ANN Model.3 layer ANN Model
recognizes the test image and classifies the class of it.
III. GIST-GLOBAL-FEATURE EXTRACTION
The whole scene recognition is very hard task for an artificial neural network. So Gist-Global feature is extracted for only
the image of interesteliminating,surrounding unwanted objects like trees, fountains and humans in the scene.
Gist feature is extracted for each image using the following step:
:
 Convert the image into gray scale image
 Resize the image 256x256.
 Normalize the image intensities to be between 0 to 255.
 Compute the FFT of the image
 Multiply with Gabor transformation function (8 orientations and 4 scales – total 32 outputs at this stage),
 Find the inverse FFT of the 32 images.
 Divide each image into 4x4 blocks (total 16 blocks). Find the energy in each block averaging the intensitiesin that
block.
 Total 16 outputs per image and total 32 images. This gives a vector of 32x16 = 512 elements.
 This is the final Global feature vector of 512 elements.
 v1= {e0,e1,e2…..e511}
Fig2. Gist feature descriptor

IV. MULTI LAYER PERCEPTRON NEURAL NETWORK
Multi-Layer Perceptron is used for Non-linear way of solving problems. To calculate the Shape, Intensities, Orientation of
Image, and Multi-Layer Perceptron is used. The MLP Learns in two ways. They are: Online Learning and Batch Learning.
In Online Learning instances are seen one by one. In Batch Learning whole scene is shown all at once. The input signal
propagates through the network in forward direction, on a layer by layer basis. These neural networks are common known as
Multi-layer perceptron.
Fig3. Sample MLP
The above figure 3 shows a sample MLP with 8 input nodes , 4 middle layer and 1 output layer. Middle layer is the hidden
layer. In this work 512 input layer nodes, 16 hidden layer nodes and 5 outputs are used.
V. RESULTS
At 5 different locations Viz., TajMahal, Statue Of Liberty, Mount Everest, Waikiki Beach, Koneru Center. 50 images are
taken in this work per each class. Training images=5*50=250. The ANN is trained 12 times for accuracy, 250*12=3000
images.The test images are 20 per each class, 5*20=100 images.For the TajMahal image is recognized and the corresponding
details like history, importance, location, etc., will be displayed for the user example tourists.
Fig4. Recognition and Description using ANN
VI. CONCLUSION & FUTURE SCOPE:
The application of neural network recognition in targeted Visual Content Recognition (Monument Recognition)
is totally a new application. The recognitionof image with ANN of using gist features is done in this work and
the accuracy of recognition is about 95%. In this work the images of 5 locations are taken. This method can be
validated on large data base of different monument images. This work can be extended, using Deep Neural
Networks.

REFERENCES
[1] Simon Haykin Neural Networks, A comprehensive foundation prentice-Hall Of India, New Delhi., 2004
[2] LaureneFausett,Fundamentals Of Neural Networks Architectures, Algorithms and Applications.,1994.
[3] A. Torralba, R. Fergus, and W. T. Freeman. 80 million tiny images: a large database for non-parametric object and
scene recognition. IEEE Transactions on Pattern Analysis and MachineIntelligence, 2008.
[4] B. Yegnanarayana, “Artificial neural networks - Hall of India”, New Delhi, 2006.
[5] Intel Corporation, “Open Source Computer Vision Library,” Reference Manual, 2001
[6] S. Kumar, “Neural Networks: A Classroom Approach”, McGraw Hill 2005.
[7] R. Hecht-Nielsen “Theory of the back propagation neural network,” Proc. IEEE Int ’I Conf. on Neural Networks,
vol. 1, pp. 593-605, 1989.
[8] Golda, Principles of Training multi- layer neural network using back propagation, 2005.
[9] David Kriesel, ”A Brief Introduction to Neural Networks”.
[10] J.Hertz,A.Krogh and R.G.Palmer,”Introduction to the theory of Neural Computation”,AddisonWesley,Redwood
City,1991

Targeted Visual Content Recognition Using Multi-Layer Perceptron Neural Network

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Targeted Visual Content Recognition Using Multi-Layer Perceptron Neural Network (20)

Recently uploaded (20)

Targeted Visual Content Recognition Using Multi-Layer Perceptron Neural Network