International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 04 Issue: 07 | July -2017
p-ISSN: 2395-0072
www.irjet.net
SVM-KNN Hybrid Method for MR Image Ms. Priti Kale1 , Prof. Priti Subramanium2 1,2Department of Computer Science and Engineering Shri Sant Gadge Baba College of Engineering& Technology, Bhusawal, Maharashtra, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - the most of automated systems in each area are
operating with high definition multimedia contents, especially in the field of medical and healthcare. This allows experts to take tricky decisions very quickly and accurately. The data in the form of image is a most important factor which is used by the physicians to make conclusion. The image data is not sufficient; there should be a proper technique to classify those images according to the training data. In case of brain tumour detection magnetic resonance image of the brain plays an important role. Today getting MR images are not that much difficult, but analyzing and classifying those images afterward, needs an expertise work to extract interesting and potential data. Here comes an image mining concept. Classification of numerical data is really easy compare to image mining. Some classification techniques we study in this paper in the relevance of image classification are SVM (Support Vector Machine), KNN (Kth nearest Neighbour). Both the techniques having their own advantages and disadvantages and try to find best of them to form something interesting. Keywords – SVM, KNN, Classification, Texture
1. INTRODUCTION . Analysis and classification both are easy for humans because of their understanding, learning, and reasoning abilities. But it get worst in case of machines, as machines do not able to understand, learn or being reasonable without human support. Then most important question arises, why human needs machine? An answer to this question is human tendency of getting bored or fed up of doing the same job. There human needs machine to operate on the same job often. But for this purpose, machines should be loaded with intelligence tools like humans have. The detection of brain tumor through the MR images first needs set of images which are the prototypes for the class brain tumor detected, rest of the images automatically termed as brain tumor not detected class. The basic steps in image classification are as follows Collection of images (Digital Data) Designing scheme
Image
Classification
Preprocessing of images Feature Extraction
© 2017, IRJET
|
Impact Factor value: 5.181
|
Selection of Training Data
Decision and Classification Classification Output Post Processing
Brain tissues in MRT images [1][10] can always be divided into two main types: normal tissues, including gray matter, white matter and cerebrospinal fluid (CSF), and abnormal tissues, usually containing tumor, etc. In case of some major disease patient diagnosis data should be observed and analyzed over the specific time period. Huge amount images are generated and needs to be analyzed in case of medical field for diagnosis of major disease. All these images should be in correct form for the purpose of classification; hence the raw images are pre-processed to remove noise or irrelevant information. In real world medical diagnosis physician get consult to the experts in case of major disease to be diagnosed. Here the fact is expert is also a human being. We can replace the human expert by a machine expert. Here machine expert is a system which having a capable hardware as well as the software installed. The software is responsible to classify the images according to given class labels. For this image classification purpose some of the well-known classifiers are Probabilistic Neural Network (PNN), Support Vector Machine (SVM), Bidirectional Associative Memory (BAM), Artificial Neural Network (ANN), Hidden Markov Model (HMM), Learning Vector Quantization (LVQ) and KNearest Neighbor (KNN) etc, and every classification technique having its own advantage and disadvantage.
II. Comparison among Classification Techniques
Different
Image
PNN is slow at classifying new cases as well as it consumes more memory space to store the model [3]. ANN performs better than other methods in terms of high dimensional features. The high computation requirement of ANN needs the high consumption of CPU as well as memory. KNN’s performance is degraded when the noisy or unrelated data encountered and number of attributes increases [7]. In SVM, there is lower classification accuracy, if the sample data of the two classes in a binary classification are all close to the separating hyper plane[5][6][10]. In this paper, we study a hybrid algorithm designed by merging the concepts of the SVM and KNN classification algorithms to classify MR Images to conclude that human brain having tumor or not. The KISO 9001:2008 Certified Journal
|
Page 1136