Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more ➡
Download
Standard view
Full view
of .
Add note
Save to My Library
Sync to mobile
Look up keyword
Like this
5Activity
×
0 of .
Results for:
No results containing your search query
P. 1
Image Retrieval using Shape Texture Patterns generated from Walsh-Hadamard Transform and Gradient Image Bitmap

Image Retrieval using Shape Texture Patterns generated from Walsh-Hadamard Transform and Gradient Image Bitmap

Ratings: (0)|Views: 1,303|Likes:
Published by ijcsis
The theme of the work presented here is gradient mask texture based image retrieval techniques using image bitmaps and texture patterns generated using Walsh-Hadamard transform. The shape of the image is extracted by using three different gradient operators (Prewitt, Robert and Sobel) with slope magnitude method followed by generation of bitmap of the shape feature extracted. This bitmap is then compared with the different texture patterns namely ‘4-pattern’, ‘16-pattern’ and ‘64-pattern’ generated using Walsh-Hadamard transform matrix to produce the feature vector as the matching number of ones and minus ones per texture pattern. The proposed content based image retrieval (CBIR) techniques are tested on a generic image database having 1000 images spread across 11 categories. For each proposed CBIR technique 55 queries (randomly selected 5 per image category) are fired on the image database. To compare the performance of image retrieval techniques average precision and recall of all the queries per image retrieval technique are computed. The results have shown improved performance (higher precision and recall values of crossover points) with the proposed methods compared to the mask-shape based image retrieval techniques. Further the performance of proposed image retrieval methods is enhanced using even image part. In the discussed image retrieval methods, the combination of original and even image part for 4-pattern texture with shape masks generated using Robert gradient operator gives the highest crossover point of precision and recall indicating better performance.
The theme of the work presented here is gradient mask texture based image retrieval techniques using image bitmaps and texture patterns generated using Walsh-Hadamard transform. The shape of the image is extracted by using three different gradient operators (Prewitt, Robert and Sobel) with slope magnitude method followed by generation of bitmap of the shape feature extracted. This bitmap is then compared with the different texture patterns namely ‘4-pattern’, ‘16-pattern’ and ‘64-pattern’ generated using Walsh-Hadamard transform matrix to produce the feature vector as the matching number of ones and minus ones per texture pattern. The proposed content based image retrieval (CBIR) techniques are tested on a generic image database having 1000 images spread across 11 categories. For each proposed CBIR technique 55 queries (randomly selected 5 per image category) are fired on the image database. To compare the performance of image retrieval techniques average precision and recall of all the queries per image retrieval technique are computed. The results have shown improved performance (higher precision and recall values of crossover points) with the proposed methods compared to the mask-shape based image retrieval techniques. Further the performance of proposed image retrieval methods is enhanced using even image part. In the discussed image retrieval methods, the combination of original and even image part for 4-pattern texture with shape masks generated using Robert gradient operator gives the highest crossover point of precision and recall indicating better performance.

More info:

Published by: ijcsis on Jan 20, 2011
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See More
See less

06/07/2013

pdf

text

original

 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 08, No.09, 2010
Image Retrieval using Shape Texture Patternsgenerated from Walsh-Hadamard Transform andGradient Image Bitmaps
Dr. H.B.Kekre
1
, Sudeep D. Thepade
2
, Varun K. Banura
3
 
1
Senior Professor,
2
Ph.D.Research Scholar & Associate Professor,
3
B.Tech (CE) Student
 
Computer Engineering Department, MPSTME,
SVKM‟s
NMIMS (Deemed-to-be University), Mumbai, India
1
hbkekre@yahoo.com,
2
sudeepthepade@gmail.com,
3
varunkbanura@gmail.com
 Abstract
 — 
 
The theme of the work presented here is gradientmask texture based image retrieval techniques using imagebitmaps and texture patterns generated using Walsh-Hadamardtransform. The shape of the image is extracted by using threedifferent gradient operators (Prewitt, Robert and Sobel) withslope magnitude method followed by generation of bitmap of theshape feature extracted. This bitmap is then compared with thed
ifferent texture patterns namely ‘4
-
pattern’, ‘16
-
pattern’
and
‘64
-
pattern’ generated using Walsh
-Hadamard transform matrixto produce the feature vector as the matching number of onesand minus ones per texture pattern. The proposed content basedimage retrieval (CBIR) techniques are tested on a generic imagedatabase having 1000 images spread across 11 categories. Foreach proposed CBIR technique 55 queries (randomly selected 5per image category) are fired on the image database. To comparethe performance of image retrieval techniques average precisionand recall of all the queries per image retrieval technique arecomputed. The results have shown improved performance(higher precision and recall values of crossover points) with theproposed methods compared to the mask-shape based imageretrieval techniques. Further the performance of proposed imageretrieval methods is enhanced using even image part. In thediscussed image retrieval methods, the combination of originaland even image part for 4-pattern texture with shape masksgenerated using Robert gradient operator gives the highestcrossover point of precision and recall indicating betterperformance.
 Keywords- CBIR, Gradient operators, Walsh-Hadamard  transform, Texture, Pattern, Bitmap.
I.
 
I
NTRODUCTION
 Today the information technology experts are facingtechnical challenges to store/transmit and index/manage imagedata effectively to make easy access to the image collections of tremendous size being generated due to large numbers of images generated from a variety of sources (digital camera,digital video, scanner, the internet etc.). The storage andtransmission is taken care of by image compression [4,7,8].The image indexing is studied in the perspective of imagedatabase [5,9,10,13,14] as one of the promising and importantresearch area for researchers from disciplines like computervision, image processing and database areas. The hunger of superior and quicker image retrieval techniques is increasingday by day. The significant applications for CBIR technologycould be listed as art galleries [15,17], museums, archaeology[6], architecture design [11,16], geographic informationsystems [8], weather forecast [8,25], medical imaging [8,21],trademark databases [24,26], criminal investigations [27,28],image search on the Internet [12,22,23]. The paper attempts toprovide better and faster image retrieval techniques.
 A.
 
Content Based Image Retrieval
For the first time Kato et.al. [7] described the experimentsof automatic retrieval of images from a database by colour andshape feature using the terminology content based imageretrieval (CBIR). The typical CBIR system performs two majortasks [19,20] as feature extraction (FE), where a set of featurescalled feature vector is generated to accurately represent thecontent of each image in the database and similaritymeasurement (SM), where a distance between the query imageand each image in the database using their feature vectors is
used to retrieve the top “closest” images [19,20,29].
 For feature extraction in CBIR there are mainly twoapproaches [8] feature extraction in spatial domain and featureextraction in transform domain. The feature extraction inspatial domain includes the CBIR techniques based onhistograms [8], BTC [4,5,19], VQ [24,28,29]. The transformdomain methods are widely used in image compression, as theygive high energy compaction in transformed image [20,27]. Soit is obvious to use images in transformed domain for featureextraction in CBIR [26]. But taking transform of image is timeconsuming. Spatial feature based CBIR methods are given in[30] as mask-shape CBIR and mask-shape BTC CBIR. Theproposed CBIR methods are further attempting to improve theperformance of these shape based image retrieval with help of shape texture patterns. Here the query execution time is furtherreduced by decreasing the feature vector size further andmaking it independent of image size. Many current CBIRsystems use the Euclidean distance [4-6,11-17] on the extractedfeature set as a similarity measure. The Direct EuclidianDistance between image P and query image Q can be given asequation 1, where Vpi and Vqi are the feature vectors of image
P and Query image Q respectively with size „n‟.
 
ni
VqiVpi ED
12
)(
 (1)
76http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 08, No.09, 2010
II.
 
EDGE DETECTION MASKS
 Edge detection is a very important in image analysis. Asthe edges give idea about the shapes of objects present in theimage so they are useful for segmentation, registration, andidentification of objects in a scene. An edge is a jump inintensity. An ideal edge is a discontinuity (i.e., a ramp with aninfinite slope). The first derivative assumes a local maximumat an edge. The various gradient operators [13] used for edgeextraction are Prewitt, Roberts and Sobel.III.
 
SLOPE MAGNITUDE METHOD
 The problem with edge extraction using gradient operatorsis detection of edges in only either horizontal or verticaldirections. Shape feature extraction in image retrieval requiresthe extracted edges to be connected in order to reflect theboundaries of objects present in the image. Slope magnitudemethod is used along with the gradient operators (Prewitt,Robert and Sobel) to extract the shape features in form of connected boundaries. The process of applying the slopemagnitude method is given as follows. First one needs toconvolve the original image with the Gx mask to get the xgradient and Gy mask to get the y gradient of the image. Thenthe individual squares of both are taken. Finally the twosquared terms are added and square root of this sum is taken asgiven in equation 2.
=
 
2
+
2
 (2)
IV.
 
TEXTURE PATTERNS USING WALSH
-
HADAMARDTRANSFORM MATRIX
 Walsh transform matrix [21,22,26] is defined as a set of Nrows, denoted Wj, for j = 0, 1, .... , N - 1, which have thefollowing properties:
 
Wj takes on the values +1 and -1.
 
Wj[0] = 1 for all j.
 
WjxWkT=0, for j not equal to k and WjxWkT =N,for j=k.
 
Wj has exactly j zero crossings, for j = 0, 1, .... , N-1.
 
Each row Wj is even or odd with respect to itsmidpointWalsh transform matrix is defined using a Hadamard matrix of order N. The Walsh transform matrix row is the row of theHadamard matrix specified by the Walsh code index, whichmust be an integer in the range [0, ..., N - 1]. For the Walshcode index equal to an integer j, the respective Hadamardoutput code has exactly j zero crossings, for j = 0, 1, ... , N - 1.Using the Walsh-Hadamard transform assorted texturepatterns namely 4-pattern, 16-pattern and 64-pattern aregenerated. To generate N
2
texture patterns, each column of theWalsh-Hadamard matrix of size NxN is multiplied with everyelement of all possible columns of the same matrix (onecolumn at a time to get one pattern). The texture patternsobtained are orthogonal in nature.Figure 1(a) shows a 2X2 Walsh-Hadamard matrix.The four texture patterns generated using this matrix areshown in figure 1(b). Similarly figure 2(b) shows first fourtexture patterns (out of total 16) generated using 4X4 Walsh-Hadamard matrix shown in figure 2(a).1(a). 2x2 Walsh-Hadamard transform matrix
1(b). „4
-
 pattern‟ texture patterns
 Figure 1. Generating 4 texture patterns2(a). 4x4 Walsh-Hadamard transform matrix
77http://sites.google.com/site/ijcsis/ISSN 1947-5500
 
(IJCSIS) International Journal of Computer Science and Information Security,Vol. 08, No.09, 2010
2(b). First four ‟16
-
 pattern‟ texture patterns
 Figure 2. Generating first four of 16 texture patternsV.
 
GRADIENT IMAGE BITMAPS
 Image bitmaps of colour image are generated using threeindependent red (R), green (G) and blue (B) components of Prewitt/Robert/Sobel image obtained using slope magnitudemethod to calculate three different thresholds. Let
X={R(i,j),G(i,j),B(i,j)} where i=1,2,….m and j=1,2,….,n; be
an m×n slope magnitude gradient of color image in RGBspace. Let the thresholds be TR, TG and TB, which could becomputed as per the equations given below as 3, 4 & 5.
min j
 ji RnmTR
1 1
),(*1
 (3)
min j
 jiGnmTG
1 1
),(*1
 (4)
min j
 ji BnmTB
1 1
),(*1
 (5)Here three binary bitmaps will be computed as BMr, BMg andBMb. If a pixel in each component (R, G, and B) is greaterthan or equal to the respective threshold, the correspondingpixel position of the bitmap will have a value of 1 otherwise itwill have a value of -1.
TR ji Rif  TR ji Rif   ji BMr 

),(...,....1 ),(....,1 {),(
 (6)
TG jiGif  TG jiGif   ji BMg

),(...,....1 ),(....,1 {),(
 (7)
TB ji Bif  TB ji Bif   ji BMb

),(...,....1 ),(....,1 {),(
 (8)To generate tiled bitmaps, the image is divided into four non-overlapping equal quadrants and the average of each quadrantis considered to generate the respective tile of the imagebitmap.VI.
 
DISCUSSED CBIR METHODS
 
 A.
 
 Mask-shape based CBIR
In this method the feature vector is obtained by extractingthe shape of the image by using gradient operators (Prewitt,Robert or Sobel) with slope magnitude method. Then thefeature vectors are compared pixel by pixel using theEuclidian distance. The limitation of this method is that it isdependent on the size of the image. For this method, size of allthe images in the database should be same as query image.
 B.
 
CBIR with Mask-shape and BTC 
First of all the shape of the image is extracted by usingthree gradient operators with slope magnitude method. Theaverage of the obtained shape feature is calculated. The featurevector is obtained by calculating the average of all thosevalues which are greater than the average of the shape featureand average of all those values which are less than or equal tothe average of the shape feature. Here the size of the featurevector is constant and is independent of size of the image.
C.
 
Proposed Shape Texture Pattern based CBIR
In the proposed gradient shape texture method, the shapefeature of the image is extracted using the three gradientoperators Prewitt, Robert and Sobel. Then the bitmap of theshape feature is generated using the modified BTC technique.The bitmap thus obtained is compared with the different
texture patterns like „4
-
 pattern‟, „16
-
 pattern‟ and „64
-
 pattern‟
generated using Walsh-Hadamard transform matrix to producethe feature vector as the matching number of ones and minusones per texture pattern. The size of the feature vector of theimage is given by equation 8.
Feature vector size=2*3*(no. of considered texture-pattern) (9)
Using three different gradient operators in association withthree assorted texture pattern sets along with original andoriginal-even image, total 18 novel feature vector generationmethods can be used resulting into 18 new image retrievaltechniques. Mask-shape based CBIR methods [30] areconsidered to compare the performance of proposed CBIRtechniques. In the proposed CBIR techniques thecombination of original and even part of images gives betterresults than original image alone [1,2]. The main advantage of proposed CBIR methods is improved performance resultinginto better image retrieval. Here also the feature vector size isindependent of image size in proposed CBIR methods.
78http://sites.google.com/site/ijcsis/ISSN 1947-5500

Activity (5)

You've already reviewed this. Edit your review.
1 hundred reads
1 thousand reads
Arun Kumar liked this
yogapriya_j liked this
padmashri1234 liked this

You're Reading a Free Preview

Download
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->