Int. Journal of Electrical & Electronics Engg.
Vol. 2, Spl. Issue 1 (2015)
e-ISSN: 1694-2310 | p-ISSN: 1694-2426
A Review of Feature Extraction Techniques for CBIR based on SVM Navneet Kaur1, Sonika Jindal2 1
M.Tech, Department of Computer Science and Engineering Assistant Professor, Department of Computer Science and Engineering Shaheed Bhagat Singh College of Engineering and Technology, Ferozepur 2
dhahnavneet@gmail.com & sonikamanoj@gmail.com
Abstract: As with the advancement of multimedia technologies, users are not gratified with the conventional retrieval system techniques. So a application “Content Based Image Retrieval System” is introduced. CBIR is the application to retrieve the images or to search the digital images from the large database .The term “content” deals with the colour, shape, texture and all the information which is extracted from the image itself. This paper reviews the CBIR system which uses SVM classifier based algorithms for feature extraction phase.
Relevance Feedback Query Formation
Descrip tion Image Database
NITTTR, Chandigarh
EDIT -2015
Visual Content
Feature Vectors Feature Database
Similarity Comparison Indexing & Retrieval
Descript ion
Index Terms— Content Based Image Retrieval , Support Vector Machine , Feature Extraction, Relevance Feedback.
I. INTRODUCTION1 Content Based Image retrieval is the method to retrieve images based on various derived features such as colour, texture and shape[1]. According to the users requirement, this technique basically use visual contents to search images from the large databases. Early techniques were based on textual annotation of images rather than the visual contents [2]. In other terms, firstly images are annotated with text, and then search can be done using text based method from traditional database management system. The first and foremost retrieval approach based on combination of textual data into each image and retrieve those images by keywords which is the traditional database query technique which is time consuming and too much gruelling task. In CBIR system the images are extracted form the database on the basis of visual contents and represented as feature vectors. Furthermore, SVM is the classifier which is basically for the regression and classification on the basis of various tools and techniques. Various algorithms are used to extract the features of images by using the SVM classifier. Moreover learning techniques are used it may be supervised or unsupervised learning techniques which are based on the training and testing phases. Firstly retrieve the feature vectors from the images (features can be colour, texture and shape) and then keep feature vectors into different databases for the future purpose. The two images in the database is similar to the query image only when the distance between two images feature vectors is small. Figure 1 shows the various phases of CBIR.
Visual Content
Output
Retrieval results
Figure 1: Content Based Image Retrieval II. FEATURE EXTRACTION Feature extraction for CBIR is the method of computing the attributes of various digital images which can be used to define information regarding the contents of the image. A feature can be associated with the single attribute or composite description of distinguished attributes. The classification of features is general purpose or domain dependent. The general purpose features can be designed anywhere in the context whereas domain dependent features are used for a specific application[6]. The advantage for feature extraction is to detect the different types of features which are used in images[9].There has been huge work done on various approaches to detect the different features among images. Feature extraction will be solicited very customarily; therefore, it would be exact, accurate and time efficient. The dimension of the vector can be reduced by using the feature extraction techniques on the basis of: Colour Texture Shape COLOR Color is more considerable visual content for the retrieval of images. It reveals the most broadly used feature in the CBIR system. The preference for the selection of features of the colour, based on results of the segmentation. For illustration, if homogeneous colour is not provided by the segmentation method then obviously this is not a better choice[4]. Firstly to represent the colour images, colour space is used. Typically, the summation of red, green and blue gray level intensities is represented by the gray level intensities which represent the RGB space.RGB space mainly used for image display in colour space. This 116