A STUDY OF CONTENT BASED IMAGE RETRIEVAL … is most widely used in CBIR. CBIR uses HSV(Hue ... Final Project Report, EE K 381 ... ...

A STUDY OF CONTENT BASED IMAGE RETRIEVAL BASED WITH

LOW LEVEL FEATURE AND SVM CLASSIFIER

1Frenisha Modi

*,

2Milin Patel

1M.E. Student, Dept. of Computer Engineering, Sardar Vallabhbhai Patel Institute of

Technology,Vasad,Gujarat, India.. 2Assistant Professor, Dept. of Computer Engineering,Sardar Vallabhbhai Patel Institute of

Technology,Vasad,Gujarat, India

Corresponding Author Address:

Mrs. FrenishaModi

M.E. Student, Dept. of Computer Engineering,

Sardar Vallabhbhai Patel Institute of Technology,

Vasad, Gujarat, India

ABSTRACT

Image Retrieval Process depending on a new Matching Strategy. Content Based Image

Retrieval uses the computer vision Technology to solve the Image Retrieval Problem. We

have used Color Co-occurrence Matrix , Gabor Wavelet Transform to Extract Image Feature.

The Images are classified using Support Vector Machine Classifier which effectively

distinguishes between relevant and irrelevant images.

KEYWORDS: Color, Texture, Color Co-occurrence Matrix, Gabor Wavelet Transform,

SVM Classifier

INTRODUCTION

“Content Based” means Search analyzes the Contents of the image rather than the metadata

such as Keywords, Tags or Descriptions associated with the Image. “Content” means Context

refer to Colors, Shapes, Texture or any other Information that can be derived from Image

itself. An “Image Retrieval System” is a Computer System for Searching and Retrieving

image from a Database of Digital Image. Many Features of Content Based Image Retrieval

but Four of them are considered such as Color, Texture, Shape and Spatial Properties [2]

].“Low Level Feature” means Image like color, Texture, Shape can be extracted from the

Image. Image Retrieval Process depend on new matching strategy. Image Retrieval Process

can be divided into two categories: annotation-based Image Retrieval (ABIR) and Content

Based Image Retrieval (CBIR). Color feature include Color Histogram, Color auto-

Correlogram, Color Dominant Descriptor, Color Co-occurrence matrix (CCM). Texture

Feature includes Tamura Texture Feature, Steerable Pyramid, Wavelet Transform, Gabor

Wavelet Transform. Shape Feature include normalized inertia, Zernike moments, Histogram

of edge Detection, edge map [1]

. Color Features are better than Texture and Shape. Color

International Journal of Scientific Research in Engineering (IJSRE) Vol. 1 (4), April, 2017

IJSRE Vol. 1 (4), April, 2017 www.ijsre.in Page 105

feature are very stable and robust because it is not sensitive to rotation, translation and scale

changes. Color feature calculation is very simple [1]

.

COMPONENT OF CBIR SYSTEM

Query Image Means teting the image of Image Database. Query Image is Input

Image.Feature Extraction has two categories: Global Feature and Local Feature[1]

. Global

Features Consider whole image.and mostly System used in Global Feature like Color

Histogram. Global features Color abd Shape provide an overall idea but not detail of whole

image.advantage of global feature as extraction and matching is done with high speed. Local

Feature refer to the small pixel blocks obtained by segmenting the image. Local Feature

better than the global feature because various domains local features give a good

classification.Image Database means many images stored in database such as WANG Dataset

as like images are Africa, Beach, Monuments, Buses etc[1]

. Feature Vector Database means

finding the feature and feature isgenerated. Similarity measurment means one or more image

I s similar to the image database.and it is extracted to the feature vector.Retrieving images

means retriev the Image Similar to the Query image like as Color, Texture or Shape.one or

more similar images are retrieved.

LOW LEVEL FEATURE

Color

Color is most widely used in CBIR. CBIR uses HSV(Hue, Saturation and Value) and

HLS(hue, Lightness and Saturation) CBIR method for retrieving images based on color are

comparing color similarity between the query image and image database. Many Method used

in Color such as Color Histogram is most commonly used in global color feature[2].

Histogram of Image I is defined as Color Ci , Hci(i) represents the number of pixels of Color

Ci in[2]

. Color co-relograms means express how the Spatial Correlations of pairs of Color

Changes with Distance. It is Combination with Global and Local Feature. This Method is

more accurate and effective to the Histogram Method [1]

. Dominant Color Descriptor used in

maximum of 8 Colors. Color co-occurrence matrix is better than another method [1]

.

Color Co-occurrence Matrix

Color Co-occurrence Matrix(CCM) represent the traversal of adjacent Pixel Color difference

in an image. as each Pixel corresponds to four adjacent colors, each image can be represented

by four images of motif of scan pattern ,which can be constructed into four Two-Dimensional

matrices of Image Size.[5]

Figure-1 3x3 Convolution mask generated

1 2 3

4 G(x ,y) 5

6 7 8



Figure-2 four 2x2 Grids are generated

Figure-1 & 2: 3x3 Convolution mask are divided into Four 2x2 grids[5]

For each pixel G(x, y), a 3x3 convolution mask can be generated in Figure-1. 3x3

convolution mask can be generated into four 2x2 grid with including G(x , y) inFigure-2.

Figure-3 The Seven Scan Pattern [5]

There are 25 different Scan pattern in Grid if the traversal goes from four angles. 25 Different

Scan Pattern means Four 2*2 Grids and Seven Motif of Scan Pattern. Motif means scanning

pattern. Motif no 0 signifies the situation where a motif cannot be formed due to the

equivalence.2*2 grids are replaced by motif of scan pattern. Let

G(x,y): Nx*Ny=Z gray level of an Nx * Ny Image I which Z= {0,1,2,3,4,…,255}[1,5]

( ) ( | ) ( [ ] [ ]) (1)

The Co-occurring probabilities of the number I motifs of scan pattern matrix are determined

by dividing Mi(u,v) by the total number of counts across all u and v[1,5]

as shown below

[ ] ( )

(2)

There will be a 7x7=49 two-Dimensional CCM grids in total.

Texture

Texture is natural property of Surface and it is characterized by repetition of patterns or

patterns over a region in an image. Texture is extracted by comparing the color contrast of

each pixel in a group of pixel or region. Texture is extracted by spectral or Statistical

Methods. The Measured of image is texture is given by Contrast, Coarseness, Directionality

and Regularity [4]

. Many Method used in Texture Feature such as Tamura features are

Coarseness, Contrast, Directionality, line likeness, Regularity, Roughness. Coarseness means

Texture granularity. Size and Number of Texture Primitives. Image size is n*n. Contrast

means the difference in intensity among neighboring pixels. Directionality means Shape of

Texture primitives and their placement rule. Line-likeness means Shape of Texture

primitives. It is often simultaneously directional. Regularity means Variations of the texture-

1 2

4 G(x ,y)

2 3

G(x, y) 5

4 G(x ,y)

6 7

G(x ,y) 5

7 8



primitive placement. Regular texture is composed of identical or similar primitives. And

Irregular texture is composed of various primitives. Roughness means Variation of Physical

Surface. A Rough texture contains angular primitives. A Smooth texture contains rounded

blurred primitives. Steerable pyramid means basic filter are translation and rotation of a

single function. Filter is linear combination of basis function and only for rotation-invariant

texture retrieval [4,12]

. Gabor wavelet are better than another texture Method [2]

.

Gabor Wavelet Transform

The process of feature extraction using Gabor wavelet transform is shown below steps,

Step 1: Convert Color image into gray image.

Step 2: Make Gabor filter using different phase angles and frequencies.

Step 3: Calculate mean and standard deviation for each and every filtered image and store in

feature vector.

SVM CLASSIFIER

SVM means Support Vector Machine. SVM is one of technique that can solve Classification

Problem. Classification is Method for complied data Systematically, Collect or Grouping into

Certain groups according to the rules or Predetermined Criteria.[13]

Image Classification

methods are categorized into two types Supervised and unsupervised Method.[5]

SVM

generate Hyper plane which separates all points in Same Class on one side of Hyper plane .

Different hyper plane which can classify all the correctly feature Set. The best Choice will be

the best hyper plane that leaves the maximum margin from both Classes. The region between

the hyperplanes on both sides of the separating hyper plane is called the margin band.

Support vector machines can be used for text categorization, image classification, particle

identification, database marketing and bioinformatics.

PERFORMANCE EVALUATION

Recall means measures the ability of the System to retrieve all models that are relevant [1]

.

(3)

TP means number of retrieved images which are similar to the query. FN means number of

images in the database which are similar to the query but not Retrieved. Precision means

measures the ability of the system to retrieve only models that are relevant. [1]

(4)

FP means number of retrieved images dissimilar to the query.



CONCLUSIONS

The Color co-occurrence matrix for Color Feature Extraction can capture the Color Variation

efficiently. The Gabor Wavelet Transform for Texture Feature Extraction can give the good

representation of the texture. These two Techniques can improve the retrieval results. As we

have classification before retrieval so if my Classification is True then no negative retrievals.

Here we retrieve images on classified class only so Computational time is reduced.

REFERENCES

1. ElAlami, M. E. "A new matching strategy for content based image retrieval

system." Applied Soft Computing 14 (2014): 407-418.

2. Murala, Subrahmanyam, Anil Balaji Gonde, and Rudra Prakash Maheshwari. "Color and

texture features for image indexing and retrieval." In Advance Computing Conference,

2009. IACC 2009. IEEE International, pp. 1411-1416. IEEE, 2009.

3. Zhang, Dengsheng, Aylwin Wong, Maria Indrawan, and Guojun Lu."Content-based

image retrieval using Gabor texture features." In IEEE Pacific-Rim Conference on

Multimedia, University of Sydney, Australia. 2000.

4. Howarth, Peter, and Stefan Rüger. "Evaluation of texture features for content-based

image retrieval." In Image and Video Retrieval, pp. 326-334. Springer Berlin Heidelberg,

2004.

5. Lin, Chuen-Horng, Rong-Tai Chen, and Yung-Kuan Chan. "A smart content-based

image retrieval system based on color and texture feature." Image and Vision

Computing 27, no. 6 (2009): 658-665.

6. Saad, Michele. "Low-level color and texture feature extraction for content-based image

retrieval." Final Project Report, EE K 381 (2008): 20-28

7. Jhanwar, N., Subhasis Chaudhuri, Guna Seetharaman, and Bertrand Zavidovique.

"Content based image retrieval using motif cooccurrence matrix."Image and Vision

Computing 22, no. 14 (2004): 1211-1220KR. Nicole, “Title of paper with only first

word capitalized,” J. Name Stand. Abbrev., in press

8. Yue, Jun, Zhenbo Li, Lu Liu, and Zetian Fu. "Content-based image retrieval using color

and texture fused features." Mathematical and Computer Modelling54, no.3 (2011):

1121-1127.

9. Zhang, Jianguo, and Tieniu Tan. "Brief review of invariant texture analysis

methods." Pattern recognition 35, no. 3 (2002): 735-747.

10. http://docs.opencv.org/2.4.10/doc/tutorials/imgproc/histograms/histogram_calculation/hi

stogram_calculation.html.



11. Rao, M. Babu, B. Prabhakara Rao, and A. Govardhan. "CTDCIRS: content based image

retrieval system based on dominant color and texture features."International Journal of

Computer Applications 18, no. 6 (2011): 40-46.

12. Deselaers, Thomas, Daniel Keysers, and Hermann Ney. "Features for image retrieval: an

experimental comparison." Information Retrieval 11, no. 2 (2008): 77-107

13. Muhammad Athoillah, M.Isa Irawan, Elly Matul Imah.”Supprt Vector Machine with

Multiple Kernal Learning for Image Retrieval” 2015 International Conference on

Information, Communication Technology and System(ICTS).

14. Sushant Shrikant Hiwale, Dhanraj Dhotre, Dr. G.R.Bamnote. “ Quick Interactive Image

Search in Huge Databases using Content-Based Image Retrieval.” 2015 IEEE Sponsored

2nd

International Conference on Innovations in Information Embedded and

Communication System.



A STUDY OF CONTENT BASED IMAGE RETRIEVAL … is most widely used in CBIR. CBIR uses HSV(Hue ... Final Project Report, EE K 381 ... ...

Documents