CONTENT BASED IMAGE CLASSIFICATION BY IMAGE FEATURE USING TSVM

Transcription

1 CONTENT BASED IMAGE CLASSIFICATION BY IMAGE FEATURE USING TSVM K.Venkatasalam* *(Department of Computer Science, Anna University of Technology, coimbatore ABSTRACT The approach for content based color image classification is done by using Transductive support vector machine. The traditional classifiers [4] are not effectual in classifying the images due to the high dimensionality of the image feature space. This paper mainly deals with histogram of the color components and shape of the image. The major purpose of using color image histogram and shape is to obtain better efficiency and accuracy in image classification. Experimental results on test sets of 1000 images are reported and compared on histogram color features for RGB, CMYK, Lab, YUV, YCBCR, HSV, HVC, YIQ and YPbPr color spaces. Keywords- Transductive support vector machine, color image histogram, image classification. 1 INTRODUCTION We are having different types of multimedia data such as text, image, audio, video and Graphic objects. Nowadays, the development in science and technology leads to have huge amount of images are existing in the database [1]. So that image is classified by using color and shapes to increase the index and retrieval efficiency of the visual information of image. Images are classified by comparing other images from database corresponding to their features [3]. Normally multimedia image occupies large amount of memory so we have to compress the image by using transform coding techniques. A typical image s energy often varies significantly throughout the image, which makes compressing it in the spatial domain difficult; however,images tend to have a compact representation in the frequency domain packed around the low frequencies, which makes compression in the frequency domain more efficient and effective[1]. Transform coding is an image compression technique that first switches to the frequency domain, then does its compressing. The transform coefficients should be decor related, to reduce redundancy and to have a maximum amount of information stored in the smallest space. These coefficients are then coded as accurately as possible to not lose information.in this project, we use transform coding. 2 Image Features: In literature, feature of the images are based on color and texture properties [15].Even though there is inequity among the images, some difficulty arises in classification problem [12]. We are choosing some specific features among the many positive features that are extracted from an image for classification [12]. In traditional classification bit map representations had used to represent the image.it has main drawback such as memory intensive and high resolution [1], larger the file size and when an image is enlarged the individual color squares become visible and illusion of smooth image is lost to the viewer. This pixilation makes the image look coarse.here tiff representation is used to represent the images. 2.1 Research track paper: Features from color histogram: A color histogram is a graphical representation showing a visual representation of the distribution of colors in an image [6]. It provides a good consequence in indexing [3] and retrieval tasks in color histogram techniques [12]. Color image histogram makes sure full transformation invariance in its task [14].A three dimensional vector by which it represents the position of the color in color space. The color space and quantization steps are chosen by using the three dimensional vector. RGB is chosen as first for the color space because it is vastly used in literature [1]. By keeping the color space condition identical, other color spaces are also considered for comparison and completeness. Classifier does not use any color space information after quantization is moved to bin [4].Although RGB and HSV color spaces are often used in three dimensional vectors, for any type of color spaces, the color histogram could be constructed [6]. Intensity histograms are used in monochromatic images that is the histogram of an image normally refers to a histogram of the pixel intensity value [5]. A histogram can be N Page 10

2 dimensional. Sharing of data in an image is briefly given by histogram [14]. Using the invariant of translation and rotation of the axis we can find the color histogram of the images and it varies slowly with an angle of view [6]. While comparing the histograms signatures of two images and the color content of one image with the other is separating. The color histograms are well suited for the problem of recognizing an object from unknown position and rotation within a scene [14]. Translation of an RGB image into the illumination invariant rg -chromaticity space allows the histogram to operate well in varying light levels [6]. 2.2 Features extracted from shape: In visual information, the shape of object plays a vital role. When using the similarity search and retrieval, the shape is a very dominant property [12]. The method called contour based shape description technique is implemented. For the shape representation and [3] Fixed-Resolution format is used. The resolution is higher when there are more blocks and data quantity. The isometric blocks are classified from the image. Edge detection is used first when shape is used as a feature [15]. 3 Transductive Support Vector Machine Transductive support vector machines extend SVM s (2) in that they could also treat partially labeled data in semi [4] learning by following the principles of transduction. Here, in addition to the training set, the learner is also given a set[11].in the primal problem, from each sub optimal point that satisfies all the constraints, there is a direction or a sub space of directions to move that increase the objective function[6].moving in any such direction is said to remove slack between the candidate solution and one or more constraints[1]. An infeasible value of the candidate solution is one that exceeds one or more of the constraints [4] A slack variable is a variable that is added to an inequality constraint to transform it to equality [10]. Introducing a slack variable replaces an inequality constraint with an equality constraint and a non-negativity constraint [1]. Transduction or transductive inference is reasoning from observed specific (training) cases to specific (test) cases [4]. In contrast, induction is reasoning from observed training cases to general rules, which are then applied to the test cases [10]. The distinction is most interesting in cases where the predictions of the transductive model are not achievable by any inductive model. Note that this is caused by transductive inference on different test sets producing mutually inconsistent predictions [4]. It is one of the most simple machine learning methods. In feature space, the closest training examples are based on dividing the objects by K- nearest algorithm in pattern recognition. KNN is part of supervised learning [4] that has been used in many applications in the field of data mining, statistical pattern recognition and many others.majority vote of neighbors are classified by objects. K is always referred a positive integer value. a set of objects are taken by neighbors whose known correct classification are taken[12]. The usage of baseline methods with many extensions. The usage of Inverse Square of weighted distance as the distance which reduce the noisy training data. It is effective when the training data is large [11]. 5 Supervised learning: For creating a function from training data, supervised learning technique is used [4]. It consists of pair of input objects and desired output. The output of the function can be a continuous value, of can predict the class label of the input object. For any valid input object, the predicted value function is collected after seeing a number of training examples. Data are labeled with predefined classes in supervised learning [4]. Supervised learning can be done by two steps. Learning: learn a model using the training data. Testing: test the model using unseen test data to assess the model accuracy. The training set must known what kind of data to be added gathering the training set. the real world use a function is represented in the training set [1]. 6 Histogram calculations: In histogram, tabulated frequencies are given by graphical display [6]. A set of predefined bins are organized by histogram which are collected in the form of counts in data [11].the intensity range of matrix ranges from These ranges are subdivided into 256 segments; those segments are called as Bins. the range of each bin can be calculated by using the number of pixels in the count[5]. Histograms are not only for color intensities but also used for different image features [6].Fig1 is shows the histogram for an image by passing the parameter as the input images. 4 K-nearest neighbor algorithm: Page 11

3 Fig1 shows the histogram for an image 7 Matrix calculations: The most commonly used method in transform geometry for 3D rendering is four by four matrices. To scale RGB colors, to control hue, saturation and contrast, RGB color matrices are used. By using standard matrix multiplication, any number of color transformation can be done in matrices is one of the major advantage [8]. Before using the matrix operations for input image, the non-linear brightness space RGB colors must be converted into linear space. Input RGB color is maintained in the saturation matrix is one of the properties. 8 Color image to gray scale: Grayscale images have many shades of gray in between. Monochromatic is the name for gray scale image. A shade of gray exclusively composed of varying from black at the weakest to white at the strongest intensity [8].Vision is probably the most important human sense. It allows us to observe and assess the environment around us with the texture and depth information [15].A lot of computer image processing can be done by changing the RGB image to gray level pictures for the simplified implementation of various algorithms [13]. Gray level images provide a wealth of information but at the same time eliminate the need of color channels. intensity of light at each pixel [5] in the single band of electromagnetic spectrum. Conversion of color image to gray scale is not unique; different colored photographic filters on the camera with different weighting of the color channels effectively present the effect of shooting black and white film. The matching of luminance of the color image to the luminance of the gray scale image is the common strategy [1]. In fig 2, by using gamma expansion, the primary colors red, green, blue linear intensity must be obtain for the conversion of gray scale representation of its luminance [8]. 9 Optimal separating hyperplane: The margin between hyperplane and any point is maximal such that the hyperplane cleanly separates the points in the dataset x into sets that contain only members of the own type [11]. The minimum distance between any data point in x and the hyperplane is 1 with respect to canonical form in the hyperplane. The finite set of learning patterns with the maximum margin of the linear classifier [4] is the optimum separation hyperplane. The input set from which the classification function is learned is called the training set[12]. Fig3 margin of the linear classifier 10 RESULTS Fig 2 gray scale image The number of applications that utilize color in images. The measuring of gray scale image is the Here the images are classified according to the color of the image. The color should be taken from the background or color of that image [1]. The images which are present in the bin are classified by using the particular color of the input image [2]. In figure 4 the images are classified by the color of the image. The image colors are the brown so all the image which contain that color are classified as single group. Page 12

4 4. Guangxia Li,Steven C.H. Hoi,Kuiyu Chang in Two View Transductive Support Vector Machines Journal of Machine Learning Research - Proceedings Track 15: (2011) 5. FrancescoBovolo,Lorenzo Bruzzone, Lorenzo Carlin in A Novel Technique For Sub pixel Image Classification Based On Support Vector Machine IEEE Transactions on Image Processing, 19, 2010: Fig 4 images are classified by the brown color In figure 5 the background of the image is green. So the images which contain background in green color are classified as single group [4]. 6. Olivier Chapelle, Patrick Haffner,Vladimir Vapnik in SVMs for Histogram-Based Image Classification.Journal of Computer Vision, vol. 7, pp Rajesh Garg,Bhawna Mittal,Sheetal Garg in Histogram Equalization Techniques For Image Enhancemant. IJECT Vol. 2, IssuE 1, MarCh Tarun Kumar,Karun Verma in A Theory Based On Conversion Of RGB Image To Gray Image International Journal of Computer Applications on (2):7 10, Fig 5 images are classified by the brown color Reference: Journal Papers: 1. Saurabh Agrawal,Nishchal K Verma, Prateek Tamrakar, Pradip Sircar in Content Based Color Image Classification Using SVM. Electronic Edition pubzone.org... International Journal of Computational Intelligence and Applications on (1): Gidudu Anthony,Hulley Gregg, Marwala Tshilidzi in Image Classification Using Svms : One - Against-One Vs One-Against-All. Journal-ref: Statistics and Computing on november (4) 3. Yanni Wang,Bao-Gang Hu in Hierarchical Image Classification Using Support Vector Machines The 5th Asian Conference on Computer Vision, January 2002, Melbourne, Australia 9. Dr.H.B.Kekre,Sudeep Thepade,Nikita Bhandari in Colorization Of Grayscale Images Using Kekre s Biorthogonal Color Spaces And Kekre s Fast Codebook Generation. IJACSA_SpecialIssueNo3 5 Sep 2011 PAGE Lorenzo Bruzzone in A Novel Transductive Svm For Semisupervised Classification Of Remote Sensing Images. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 44, NO. 11, NOVEMBER Yunpeng Li,David J.Crandall,Daniel P.Huttenlocher in Landmark Classification In Large-Scale Image Collections. ICCV on September 29,2009 page Forum RSS Twitter... Cristian Sminchisescu 12. Nikita Orlov,Lior Shamir,Tomasz Macura, Josiah Johnston,D.Mark Eckley,Ilya G.Goldberg in WND-CHARM:Multi-Purpose Image Classsification Using Compound Image Transforms. NIHPA Author Manuscripts on 27 Oct Sergvan.S in Color Histogram Features Based Image Classification In Content-Based Image Retrieval Systems. IEEE TRANSACTIONS ON Page 13

5 Ieee Applied Machine Intelligence and Informatics, on january Zhenhua Zhang in An Improving Technique Of Color Histogram In Segmentation-Based Image Retrieval. This paper appears in: Information Assurance and Security, IAS '09. Fifth International Conference on Aug Ashour M.W, Hussin, M.F,Mahar,K.M on April 2008 in Supervised Texture Classification Using Several Features Extraction Techniques Based On ANN and SVM. Computer Systems and Applications, AICCSA IEEE/ACS International Conference on April Page 14