DOI : 10.17577/IJERTV2IS121161
- Open Access
- Total Downloads : 278
- Authors : Mr. Neeraj Shrivastava, Mr. Naveen Kumar Joshi
- Paper ID : IJERTV2IS121161
- Volume & Issue : Volume 02, Issue 12 (December 2013)
- Published (First Online): 27-12-2013
- ISSN (Online) : 2278-0181
- Publisher Name : IJERT
- License:
This work is licensed under a Creative Commons Attribution 4.0 International License
Content Based Image Retrieval Using GNW Hybrid Method
1 Mr. Neeraj Shrivastava 2 Mr. Naveen Kumar Joshi
1Associate Professor, 2 M.E. Student,
1Associate Professor, Dept. of CSE, Institute of Engineering& Science ( IES),IPS Academy, Indore, MP, India
2M.E Student, Dept. of CSE, Institute of Engineering & Science ( IES),IPS Academy, Indore, MP, India
Abstract
The dramatic rise in the sizes of images databases has affected the development of effective and efficient retrieval systems and is one of the most active research areas. In this paper we proposed a new effective image retrieval method Content based image retrieval using GNW Hybrid method which usage Wavelet transform for Color feature extraction and Gabor transform for Texture feature extraction. Also in this paper we compare all three methods GNW Hybrid method, Wavelet transform and Gabor transform based on the parameter Precision, Recall and F-score for same images. The experimental results show that the proposed approach significantly improves the effectiveness of the image retrieval system.
Index terms- Wavelet Transform, Haar, DWT, Gabor Transform and Content based Image Retrieval.
-
INTRODUCTION
With the rapid growth of computer power, rapidly declining cost of storage and ever-increasing access to the Internet, digital acquisition of information has become increasing popular in recent years [1]. A huge amount of information is available, and daily gigabytes of new visual information is generated, stored, and transmitted. Since 1970s, image retrieval has been studied by mainly two community: Database Management and Computer Vision [2]. They adopted two different techniques:
-
Text-based
-
Visual based.
In Text-based retrieval, images are annotated using one or more keywords, subject heading, or classification codes, which in turn are used as retrieval keys during search and retrieval. Text-based retrieval is non-standardized because different users employ different keywords for annotation .Textual information about images cannot describe by text and involves a vast amount of labor and tends to be colored by personal subjective.
To overcome these difficulties, Content- based Image retrieval (CBIR) emerged as promising means for the describing and retrieving images. According to its objective, instead of being manually annotated by Text-based keywords, images are indexed by their visual content, such as color, texture, shape, and spatial layout. The Importance of Content- based retrieval for many applications, ranging from art galleries and museum archives to the picture collections, criminal investigation, medical, makes the visual information retrieval one of the fastest growing research field in information technology.
This paper proposes new improved Content-based image retrieval using GNW Hybrid method which is usage wavelet transform for extracting the color features and Gabor transform for extracting the texture features of images in horizontal, vertical and diagonal direction and normalized.
This paper is organized as follows: Section II reviews related research on image retrieval. Section III gives detail about our Proposed Method. Section IV tells about Experiments and Results. Section V presents concluding remarks.
-
-
Literature Survey or Related works
In paper 1 it presented the 2-D DWT operates in straight forward manner by inserting array transposition between the two 1-D DWT. The rows of the array are processed first with only one level of decomposition. This essentially divides the array into two vertical halves, with the first half storing the average coefficient, while the second vertical half stores the details coefficient. This process is repeated again with the columns, resulting in four sub-bands within the array defined by filter output [8].
Paper 2 tells that Texture is an important feature of natural images. A variety of techniques have been developed for measuring texture similarity. Gabor wavelet is widely adopted to extract texture from the images for retrieval and has been shown to be very efficient. Basically Gabor filters are a group of the wavelets, with each wavelets capturing energy at specific frequency and specific orientation. The scale and orientation property makes it especially useful for texture analysis [9].
Paper 3 described a new and more efficient images descriptor using the weighted combination of color and texture features based on simplified wavelet spatial-color statistics and applying normalized second-order statistics texture via Gabor Wavelet transforms [10].
-
PROPOSED MODEL
Figure 1 described the block diagram of the proposed retrieval approach. A single feature may lack sufficient discriminatory information to permit the retrieval of relevant images so where multiple features utilizing a combination of color and texture features that have been extracted separately. Here we used Wavelet transform for color feature extraction and Gabor transform for the texture feature extraction. For example if I Images is entered the first image is preprocessed so it is converted in to I(X, Y) which is gives separately to the wavelet transform and Gabor transform for extracting color feature Fc and texture feature Ft. Color quantization is carried out using color histogram then it is The normalized histogram is obtained by diving with the total number of pixels. After calculating the color and texture
feature Fc and Ft, they are multiplied by weighting factors Wc and Wt for color and texture features respectively, and combined their results to get the features of Hybrid method Fh, and store the value in database.
Same procedure is apply when a Query image Iq is entered it is per processed so it is converted into Iq (X, Y) Then it passes separately to the wavelet transform to get its color feature Fqc and to Gabor transform to get its texture feature Fqt. Color quantization is carried out using color histogram then its normalized histogram is obtained by diving with the total number of pixels. After calculating query image color feature Fcq, texture feature Ftq they are multiplied by weighting factor Wc and Wt respectively, and combined their result to get the Hybrid Method Query image feature Fhq, store the value in database. To get the similarity between the query image and stored images in the database we are calculate Euclidian distance between them and that result will show the relevant images according to the Query image Iq.
-
Proposed Algorithm Steps
Step1: Input image I.
Step2: Extract the Red, Green, Blue Components from an Image.
Step3: Decomposed each Red, Green, Blue Component using Haar Wavelet transform to get approximate Coefficient, and vertical, horizontal, diagonal detail coefficients.
Step4: Combine approximate coefficient of Red, Green, and Blue Component.
Step5: Color quantization is carried out using color histogram.
Step6: The normalized histogram is obtained by diving with the total number of pixels.
Step7: Repeat the Step3 to Step6 for image to get the Color Features Fc.
Step8: Apply Gabor transform on Red, Green and
Blue components in vertical, horizontal and diagonal direction to Get its approximate and detail coefficient.
Step9: Combined approximate coefficient of Red, Green, and Blue Components.
Step10: Calculate the Standard deviation.
Step11: Repeat the Step8 to Step10 for image to get the Texture Features Ft for Images.
Step12: Combined the Color Features (Fc) and Texture Features (Ft) to gt the Hybrid Method Feature Fh.
Fh=Wt*Ft+Wc*Fc. Where Wt=0.9 and Wc=0.1.
Step13: Repeat Step1 to Step12 on all images of image database.
Step14: Apply same procedure to get Query Image Features Fcq, Ftq and Fhq
Step15: Calculate the Euclidean Distance between
the Query Image and Images of database. Step16: Retrieve the relevant Images
Images
Query Image
I Iq
Pre Process the Image Pre Process the Image
Extract Color
I(X, Y)
Extract Texture
Extract Color
Iq(X, Y)
Extract Texture
Features Using Wavelet Transform
Features Using Gabor Transform
Features Using Wavelet Transform
Features Using Gabor Transform
Fc Ft Fcq Ftq
Combined Features Combined Features
Fh Fhq
Fh
Feature Database
Similarity Measure
Fhq Query Image
Feature
Relevant Images
Figure 4.5 Block Diagram of the Proposed Retrieval GNW Hybrid Method
for GNW Hybrid Method used of combining wavelet and gabor method outputs
Fh=Wt*Ft+Wc*Fc.
Where Wt=0.9 and Wc=0.1.
Fh = features of GNW Hybrid method
Ft= Texture feature
Fc= Color feature
Wt= multiplying coefficient of texture features Wc= multiplying coefficient for color features
-
Wavelet transform
-
Gabor transform –
Gabor transform is a good multi resolution approach that represents the texture of an image in an effective way using multiple orientations and scales. This approach has a spatial property that is similar to mammalian perceptual vision, thereby providing researchers a good opportunity to use it in image processing. Gabor filters are found to perform better than wavelet transform and other multi resolution approaches in representing textures and retrieving images due to its multiple orientation approach
The Gabor function is a complex exponential modulated by a Gaussian function. For a given image I(X,Y) with size P X Q, its discrete Gabor wavelet transform is given by convolution:
Wavelet transform is an automatic multilevel decomposition of a signal. It represents the input signal as a superposition of a family of the basic function called wavelets. Translating and dilating the
Gmn(X, Y) =
, *mn (s, t)
mother wavelet corresponding to the particular basic can generate a set of basic functions. The signals are passed through a low pass (LPH) and fallowed by high pass (HPF). The outputs of the filter are down- sampled for next level decomposition, thus allowing information from signal to be represented at different scales.
Where, s and t are the filter mask size variables, *mn and is a complex conjugate of mn which is a class of self similar functions generated from dilation and rotation of the following mother wavelet:
mn(x,y)
= + ( + ) . ()
Haar wavelets are widely being used since its invention after Haar. Haar wavelet is the simplest type of wavelet. In discrete form Haar wavelets are related
to a mathematical operation called Haar transform. Like all wavelet transform the Haar transform decomposed a discrete signal into two sub signals of half of its length one sub signal is running average the other sub signal is a running difference or fluctuation. We have used Haar wavelets to compute color
Where W is called modulation frequency. The space constants x and y define the Gaussian envelope along the x and y axes. The self-similar Gabor wavelets are obtained through the generating function:
_ _
features, because they are the fastest to compute and also have been found to perform well in practices. Haar wavelet function enables us to speed up.
mn
(x, y) =a-m (x, y)
The Haar mother wavelet function (t) can be described as:
1, 0t1/2 (t) = -1, 1/2t1
0, otherwise
Where m and n specify the scale and orientation of wavelet respectively, with m=0,1,M-1, n=0,1,N-1.
_
X= a-m (x cos + y sin)
_
Y= a-m (-x sin + y cos)
Where a>1 and = n / N. The scale factor a-m is meant to ensure the energy is independent of m.
-
Similarity measure-
For calculating the similarity we calculate Euclidean Distance between the query image (Iq) and images (I) in database formula of Euclidean Distance is as fallow:
Where and , features of query image and images in database. Where xi represents the ith feature of the , and yi features of .
-
-
EXPERIMENTAL RESULTS
The Proposed model is simulated on an Intel Pentium IV, 2.66GHz and 1GB RAM. The GNW Hybrid Content based technique was tested on an images database of 425 images of 17 categories each category has 25 images and 100×100 pixel resolution [3]. The system is capable of accepting images as input and finding the best match images from within the database. The performance parameters are Precision, Recall, and F-Score. Y. H. Lee et al. [10] the most common evaluation measures used in IR (Information Retrieval) are precision and recall, usually presented.
Figure 4.2 Show relations between irrelevant items retrieved, relevant items retrieved and relevant items not retrieved.
Precision
Precision is the probability of retrieving images that is relevant to query or it is of ratio of relevant records retrieved to the total number of relevant and irrelevant record retrieved it is usually express as percentage.
Figure 4.3 Precision
Precision= =
Recall
+
Figure 4.1 Show the relation between set of items retrieved and set of relevant images
Recall is the probability of relevant being retrieved or it is ratio of the number of relevant record retrieved to the total number of records in the database. It is usually represent as percentage.
Figure 4.4 Recall
Recall=
=
+
Where A= is the number of relevant images retrieved.
B= is the number of relevant that were not retrieved. C= is the number of irrelevant images retrieved.
F-Score
F-Score is the parameter often used to combined Precision and Recall into single performance measure, Higher the value of F-Score when value of both Precision and Recall is higher.
Fscore=
+
Table 4.1 show comparing different method (Local color histogram, Scalable Color Descriptor, Color Correlogram, Wavelet Correlogram, GLCM, Descriptor, Wavelet Method, Gabor Method, GNW Hybrid Method) based on parameter average precision.
Table 4.2 show comparing different method (Local color histogram, Scalable Color Descriptor, Color Correlogram, Wavelet Correlogram, GLCM, Descriptor, Wavelet Method, Gabor Method, GNW Hybrid Method) based on parameter average recall.
Table 4.3 show comparing different method (Local color histogram, Scalable Color Descriptor, Color Correlogram, Wavelet Correlogram, GLCM, Descriptor, Wavelet Method, Gabor Method, GNW Hybrid Method) based on parameter average Fscore.
Table (4.1) Table Comparison of Different Method Based on Average Precision
Method
Average Precision
Local color histogram[13]
0.333
Scalable Color Descriptor[7]
0.354
Color Correlogram[5]
0.324
Wavelet Correlogram[12]
0.325
GLCM[6]
0.156
Descriptor[10]
0.391
Wavelet Method
0.814
Gabor Method
0.917
GNW Hybrid Method
0.994
Wavele
t,81.40
%
GNW
Hybrid, 99.40%
Wavele
t, 81.40
%
GNW
Hybrid, 99.40%
Gabor,
91.70%
Gabor,
91.70%
Figure 4.5 Comparison of Proposed system Wavelet method, proposed system Gabor method, proposed GNW Hybrid Method based on Average Precision
Table (4.2) Table Comparison of Different Method Based on Average Recall
Table (4.3) Table Comparison of Different Method Based on Average F-score
Method
Average Recall
Local color histogram[13]
0.709
Scalable Color Descriptor[7]
0.716
Color Correlogram[5]
0.658
Wavelet Correlogram[12]
0.705
GLCM[6]
0.185
Descriptor[10]
0.771
Wavelet Method
0.8624
Gabor Method
0.894
GNW Hybrid Method
0.9508
Method
Average F-score
Local color histogram[13]
0.453
Scalable Color Descriptor[7]
0.474
Color Correlogram[5]
0.424
Wavelet Correlogram[12]
0.445
GLCM[6]
0.169
Descriptor[10]
0.519
Wavelet Method
0.8375
Gabor Method
0.9052
GNW Hybrid Method
0.97192
Method
Average Recall
Local color histogram[13]
0.709
Scalable Color Descriptor[7]
0.716
Color Correlogram[5]
0.658
Wavelet Correlogram[12]
0.705
GLCM[6]
0.185
Descriptor[10]
0.771
Wavelet Method
0.8624
Gabor Method
0.894
GNW Hybrid Method
0.9508
Method
Average F-score
Local color histogram[13]
0.453
Scalable Color Descriptor[7]
0.474
Color Correlogram[5]
0.424
Wavelet Correlogram[12]
0.445
GLCM[6]
0.169
Descriptor[10]
0.519
Wavelet Method
0.8375
Gabor Method
0.9052
GNW Hybrid Method
0.97192
Wavele t, 86.24
%
Gabor, 89.40%
GNW
Wavelet
, 83.75%
Wavelet
, 83.75%
GNW
Hybrid
, 97.19%
GNW
Hybrid
, 97.19%
Hybrid, 95.08%
Gabor,
90.53%
Gabor,
90.53%
Figure 4.6 Comparison of Proposed system Wavelet method, proposed system Gabor method, proposed GNW Hybrid Method based on Average Recall
Figure 4.7 Comparison of Proposed system Wavelet method, proposed system Gabor method, proposed GNW Hybrid Method based on Average F-Score
Figure 4.8 Query Image brick2.jpg
Figure 4.9 Result of Query Image (Brick2.jpg) Using Wavelet Transform
Figure 6.6 Result of Query Image (Brick2.jpg) Using Gabor Transform
Figure 4.10 Result of Query Image (Brick2.jpg) Using GNW Hybrid Method
-
CONCLUSION
Many research have been done to develop some algorithms that solve some problems and achieve the accuracy when retrieving images and distinguishing between them. Many proposed algorithms use images to extract features and use their features for similarity matching. When only wavelet transform is applied (extract color feature) it show good result for some images and for some images we noticed it showing more irrelevant images. In the same way when Gabor transform is applied (extract texture feature) it show good result for particular type of images and for some images its result is not appropriate and show more number of irrelevant images.
Our Proposed method GNW Hybrid method that is uses wavelet transform for color features extraction and Gabor transform for texture feature extraction. And for calculate the similarity between query image and database images using Euclidian distance. Several experiments were performed to analyze of the performance of the proposed system. Our proposed GNW Hybrid method show good result in both cases where wavelet or Gabor transforms method showing more irrelevant images.
Our experiments with image also prove that only color or only texture feature is not sufficient to describe an image. The Results also reflect that GNW Hybrid method is more efficient in terms of average Precision, average Recall, and average F-score in compare to the previous methods, value of average precision 99.4%, value of average Recall 95.08%, and value of F-score 97.19%.
-
REFERENCE
-
N Gnaneswara Rao, Dr. V Vijaya Kumar, V Venkata Krishna, Texture Based Image Indexing and Retrieval, IJCSNS International Journal of Computer Science and Network Security, Vol 9 No 5, May 2009, page number 206-210.
-
Xin Liu, Xinge You, Yin-ming Cheung, Texture Image Retrieval Using Non- Separable Wavelets and Local Binary Patterns, IEEE International Conference on Computaion Intelligence and Security, July 2009, page number 287-291.
-
Web site http://www.ee.cityu.edu.hk/~mirror/
-
Passawan Skulsujirapa, Supavadee Aramvith, Supakorn Siddhichai, Development of Digital Image Retrieval Technique using Auto correlogram and Wavelet based Texture, 47th IEEE International Midwest Symposium on Circuits and Systems, vol.1, page number , 273-276,2004.
-
T. Ojala, M. Rautiainen, E. Matinmikko, M. Aittola, Sementic Image Retrieval with HSV Correlograms, Proceeding 12th Scandinavian Conference in Image Analysis, pp.621~627, June, 2001.
-
Peter Howarth, Stefan Ruger, Evaluation of Texture Features for Content-based Image Retrieval, LNCS, vol.3115, Springer, 2004, Page number 326-334.
-
Akio Yamada, Robert Ocallaghan, Sang- Kyun Kim, MPEG-7 Visual Part of eXperimentation Model (XM) Version 27.0, ISO/IEC JTC1/SC29/WG11/N7808, Jan., 2006.
-
Ms. Yamini S. Bute, Prof. R. W. Jasutkar, Implementation of Discrete Wavelet
Transform Processor For Images Compression,International Journal of Computer Science and Network(IJCSN),volume 1,Issue 3, june 2012.
-
S. Mangijao Singh, K.Hemachandran, Contnet-Based Images Retrieval using Color Moment and Gabor Texture Feature,IJCSI International Journal of Computer Science Issue 5,Vol 9,page number 299-309, Sep 2012.
-
Yong-Hwan Lee, Sang Burm Rhee, Content-based Image Retrieval Using Wavelet Spatial-Color and Gabor Nomalized Texture in Multi-resolution Database , IEEE Sixth International Confernce on Innovative Mobile and Internet Services In Ubiquitous Computing, page number 371-377,Jan 2012.
-
Ricardo Baeza-Yates, Berthier Ribeiro- Neto, Modern Information Retrieval: The Concepts and Technology behind Search, 2011.
-
H. Abrishami Moghaddam, T. Taghizadeh Khajoie, A.H. Rouhi, A New Algorithm for Image Indexing and Retrieval using Wavelet Correlogram, Proceedings of International Conference on Image Processing, vol.3, page number.III.497-500, 2003.
-
Bongani Malinga, Daniela Raicu, Jacob Frust, Local vs Global Histogram-based Color Image Clustering, Technical Report TR06- 010, School of Computer Science, DePaul University, USA, 2006.