Anda di halaman 1dari 5



Image Retrival of Domain Name system Space Adjustment Technique

A. N. Karthikeyan and R. Dhanapal

AbstractToday the time large amount of picture elements are captured and stored invisual information systemsshould be effective and efficient.the new image indexing and manipulation techniques are required. The picture elements in visual information systems are stored in some compressed forms. The needed simple method desirable to explore image technologies for feature extraction and image manipulation in the compressed domain systems.The image of feature extraction and manipulation are performed on compressed imagesupon video without decoding or with minimal decoding only. The compressed new domain approach imposes many constraints. It provides great poten- tial for reducing computational complexity, because of reduction of the amount of data after compression. This thought try to provide an idea of work area in domain systems . Describes the results and analysis of the future directions of our work on compressed-domain texture fea- ture extraction, image matching and image manipulation.

Index Terms—DCT, Image Retrival., TranformZero, Domain inverse




T Oday operative techniques are need today for image indexing and searching are re-

quired for photo elements in information systems of image databases and video servers. In genral methods that allow users to search images based on keywords like charactristic of texture, shape, color.The image query carried on feature- based image search provide powerful tools to complement existing keyword-based search tech- niques. In genral image charactristic of texture, shape, color, and object motion are extracted and stored as side information as like in blue tooth tranmission data techniques methodology [1]. The possibility to use networkservices of connec- tion is almost transparent to the user of data or image tranmission is concerned theory with stored and genral techniques [3]. The similarity retrieval is performed based on the comparison of the charactristics associated with each image in the database. On a desktop video editing system, users would like to have general tools for image geometrical transformation image ltering, multi-image composition, and video seg- ment cut-and-paste. In a networked video applica- tion, users may want to subscribe multiple im- age,video sources from different locations and combine them to a single displayable format. In a multi-point video conferencing application a net- work device such as a video bridge may receive multiple video sources and generate multiple vid-




B. Dr.R.DHANAPAL. Research Supervisor, HOD, Department of Com- puter Application, Easwari Engineering College,Chennai,India.

.KARTHIKEYAN, Research Scholar, Bharathiar Universi-


eo streams of various forms to different end users [2] A timely and important research issue would be the following: given today’s existing compres- sion charactristic of transform coding and interframe predictive coding functionalities that one can possibly achieve in the existing com- pressed domaintechnique pursuing the maximal functionalities in the compressed domain has advantageof less data in the compressed domain than the original uncompressed domain. And also most stored visual materials are compressed. Ap- plying image searchingmanipulation techniques in the compressed domain can avoid the overhead of coding process .The compression algorithms actu- ally perform some forms of information ltering. And content decomposition which can provide good foundations for subsequent image content analysis. This paper gives an rview of our research on new compressed domain techniques for image video indexing and manipulation. The image charactristics of visual feature extraction, image matching, image manipulation, video editing in the compressed domain. Our idea concerd with technique that operate on the compressed data directly without any decodingtry to eliminate of empty space in order to extract useful data from the compressed images. The work compressed domain in this paper refers to both the ideal case without decoding and the sub optimal case with minimal decoding.


2.1 Retrieval process

Image retrieval process genrally followed by acess the image, search by example, serach by the sektech, search by thetext, navigation with image



catgoryies.In the part of the process,Image matching has been used in many applications including image registration, pattern recognition, and stereoscopic image correspondence matching. Two critical factors in image matching are determination of the match- ing criterion and the search space. One Example is the minimal distortion matching used in the popular motion estimation algorithm for video coding[5] In we have derived algorithms for doing motion estima- tion and inverse motion compensation in the DCT domain. For any ortho normal transforms like DCT, the Euclidean distance is preserved in the transform domain. However, because motion compensation is pixel-based while DCT is block-based, computation of the DCT of each reference block may involve sig-

of the DCT of each reference block may involve sig- nificant overhead in realigning the DCT

nificant overhead in realigning the DCT block struc- ture in genral retrival shown in figure(1)

To compensate for this overhead, the search space may need to be reduced, using some heuristics such as the 3-

Fig .(1) Image retrival system Architecture

point motion estimation techniqueAnother promising technique for image matching in the wavelet subband domain is to incorporate the zero-crossing representa tion. In [3] a stabilized zero-crossing representation was used in stereo image correspondence matching. It was shown that under certain conditions, the sta- bilized zero-crossing representation is complete and stable. A unique signal can be reconstructed from its stabilized zero-crossing representation. One interest- ing application is to use the distance of the stabilized zero-crossing representation to approximate the dis- tance between two signals. Zero-crossing represen- tation can be computed from the wavelet transform of a signal if the wavelet function is the second- order or first-order derivative of a smooth function.

2.2 Extraction process

The figure(2) shows imge extraction process Feature extractionof image matching, image manipulation, and video editing inthe compressed domain have techniques that operate on the compressed data di-

rectly without any decoding. Compression system method in DCT is a feature- based image query; we have derived automatic al- gorithms for extracting low-level signal features from the transform compressed images. Textures have been used to describe content of many real- world images; for example, clouds, trees, bricks, hair, fabric all have textural characteristics. Psycho- physical studies have shown that humans perceive textures by decomposing signals. We use the feature sets defined in transform decomposition to approx- imate the texture feature. Transform decomposition of images can be obtained by taking from DCT, subband wavelet transform of the images. From the decomposed signal bands, texture feature sets are defined by measuring each subband energy.

2.3 Algorithem theory

DCT a derivative of values from simulationwork with real sinusoidal basis functions We take the 5-level wavelet values worked with measurments from derived units.

DCT transform

N * N =N 2

- - - - - ( 1)

N2 signal bands can be obtained by regrouping

transform coefficient. The statistical measures such as


subband in forming the

features ,reduce the empty spaces to produce the

transform-domain texture feature to extract promi- nent Regions from each image in the database. One

image may have Zero or multiple

geneous texture regions. In fig(3)Given the input im- age key, the texture feature vector is derived from the transform domain [11]and compared against every region contained in every image in the database.

moments can also be derived from each

transformdomain texture

prominent homo-

be derived from each transformdomain texture prominent homo- Fig (3) DCT based compression system MC –Motion

Fig (3) DCT based compression system

MC –Motion estimation. FM-Frame memory VLC-Variable length code

3 Compressed-Domain Image Manipulation

3.1 Exprimental calculation

We calclaute the above compressed-domain ap- proach to image manipulation in this section. Im- age manipulation involves many useful opera- tions for general multimedia applications; In gen-



eral, it includes linear and non-linear operations. We have been focusing on the compressed- domain solutions for linear operations, such as ltering, geometrical transformation, multi-object composition, pixel multiplication, and convolu- tion. The two dimensional separable linear filter- ing of the images can be expressed as

Y=i W i P i H i

--- 3

Where Pi is the input image blocks, Hi and Wi are filter coefficient matrices in the horizontal and verti- cal directions respectively, and Y is the output fil- tered image block.

3.2 Image matching

If the images are encoded by wavelet or subband

transforms,image matching can be implemented in an intelligent, hierarchicalway as well. Suppose we adopt correlation as the matching criterion and use the exhaustive search space. Searching for the posi- tion with the highest correlation is equivalent to finding the peak value in the convolution. One can

prove that the correlation criterion is closely related

to the MSE or the correlation coefficient criterion. It

has been shown that convolution of two 1-D se-

quences can be decomposed to the summation of

convolutions of their subband components. we took

a similar approach to implement a hierarchical im- age matching method using for image matching.

3.3 Indexing and Editing

A method for indexingediting and similarity search-

ing in Image DataBases IDBs. ImageMap answers queries by example involving any number of objects or regions and taking into account their inter- relationships. The management of large volumes of digital images has generated additionalinterest in methods and tools for real time archiving and re- trieval of images by content

4 Domain retrival implementation 4.1Image Processing:

The image processing function with searching image processed with charactristic of color, local shape,texture.the process of apporachcolor axes(R- G,2B-R-G,R+G+B) by RGB color theory.The full of information about the image could be the image process Search for clusters in a color histo- gram to identify which pixels in the image originate from one uniformly colored object.


Classesof indexing methods with image pro- cessing1)Space partitioning2)Data partition- ing3)Distance-based technique4)Varies tree struc- ture5)O log N.Compared to still images, a video sequence can be further characterized by two addi- tional features are the video is captured ,image fea- tures change over time .There are existing tech- niques for extracting these dynamic visual features

in the uncompressed domain. Work has been re- ported in to detect scene changes in the transform domain and the MPEG.the process first derive in- terpretation from feature set.second generate a probability distributionthird: four semantic layers already discussed.where Pi is the input image blocks, Hi and Wi are lter coef cient matrices in

the horizontal and vertical directions respectively,

and Y is the output ltered image block. Using the distributive property of separable orthogonal transform with respect to matrix multiplication

[6] To extend the image manipulation techniques to the motion compensation domain is not directly feasible, due to the complication of the motion compensation algorithm. we have provided a par- tial solution which applied the transform-domain inverse motion compensation to convert the input video to the transform domain and kept the ma- nipulation. manipulation operations in the trans- form domain(fig.2). This will incur some overhead associated with the transform-domain inverse mo- tion compensation, whose net impact on the overall computation cost actually depends on the motion vector distribution for each speci c input of image and video stream.

5. Experiment results discussion

5.1 Methodology of Retriaval

we took a similar approach toimplement a hierar- chical image matching method. If {h1, h2}and {g1, g2} are subsampled low-band and high-band signal decomposition of the original sequences h and g. Their convolution can be calculated as expressed in the z transform form method

h(z)g(z) = h1(z2)g1(z2) S11(z) + h2(z2)g2(z2) S22(z)


h1(z2)g2(z2) S12(z) + h2(z2)g1(z2) S21(z)


Where Sij are the product synthesis filters for each subband convolution.The above equation includes two intra-subband convolutions and two cross- subband convolutions. If the analysis filters are ideal half-band low-pass and high-pass filters, thecross-subband terms will be zero. For practical filters, such asthe Harr filter and QMF filters, these terms are non-zero althoughthey are relatively small compared to the intra-subband convolutions. In [8], we described an adaptive convolution scheme which adaptively approximates the com- plete convolution with the dominant subband con- volutions. The criteria for choosing the dominant



subband convolutions are based on two possible features energy and feature. The energy-based ap- proach chooses the subbands with the highest en- ergy and approximate the complete convolution with the intra- and cross-subband convolutions as- sociated with those dominant subbands. Note that the subband decomposition can be iterated more times in a uniform, logarithmic, or adaptive way to create signal decompositions at more levels. The above adaptive, hierarchical convolution can be easily repeated in each iteration. The hierarchical image searching method has been studied earlier in [7],but only low-low band convolution was used to approximate the complete result. One alternative criterion for choosing the significant subbands is to use the signal features, such as edge and texture, in each subband. For example, if one subband has strong indication of edge or texture content, it is better to include that subband in the approxima- tion. with incorrect files or poorly formatted graphics. To extend the image manipulation techniques to the motioncompensationdomain is not directly feasible, due to the complicationof the motion compensation algorithm to eliminatethe extra empty spaces to avoid time delayby our algorithm method. In [5], we have provided a partial solution which applied the transform-domain inverse motion compensation to convert the input video to the transform domain and kept the manipulation operations in the trans- form domain. This will incur some overhead associ- ated with the transform-domain (inverse) motion compensation,whose net impact on the overall com- putation cost actually depends on the motion vector distribution for each specific input video stream. Some operationssuch as shearing and rotation, can- not be directly modeled by a linear operation like that in EQ 2. In general, they require different opera- tions on different rows and columns This problem can be solved by using the divideandconquer ap- proach.



We have presented a method of compressed-domain image technologies for image manipulation in this paper. We believed by taking advantage of some nice properties of existing compression algorithms we will be able to provide some extent of content accessibility for today’s compression algorithms. This will be a good evaluation criterion for compar- ing various existing image compression techniques. In the context of feature extraction for image query, one future direction is to nd effective ways for in- tegrating multiple features, such as color, texture, shape, and motion in the same domain and to test them on concrete, specic applications. We believe by low-level signal features alone it will not be a suf- ficient solution. One critical component will be the

integration of domain user knowledge and other complementary indexing techniques, such as text keywords. On the image analysis research front, techniques for de ning visual features that are in- variant to geometry and noise will be crucial as well.


[1] M.Sughasiny,Dr.R.Dhanapal.“A Study on Local AreaNetwork Access Point Using Bluetooth De- vices”. International Journal of Computing, Vol. 2, Is- sue 12, pp. 25 – 32, 2010.


.S.-F. Chang, W.-L. Chen, and D.G. Messerschmitt,


“Video Compositing in the DCT Domain,” IEEE Intern. Workshopon Visual Signal Processing and Commu- nications, Raleigh,North Carolina, September, 1992 .S. Mallat, “Zero-Crossing of a Wavelet Transform,”

IEEE Transactions on Information Theory, Vol. 37, No. 4, July 1991, pp.1019-33. [4] P. Brodatz, Textures: a Photographic Album for Art- ists andDesigners, Dover, New York, 1965. [5] S.-F. Chang and D.G. Messerschmitt, “Manipulation and Compositing of MC-DCT Compressed Video,” IEEE Journalof Selected Areas in Communications, Special Issue on Intel-ligent Signal Processing, pp. 1-11, Vol. 13, No.1, Jan. 1995

[6] J.R. Smith and S.-F. Chang, “Quad-Tree Segmenta- tion for Texture-Based Image Query” Proceedings, ACM 2nd Multi-media Conference, San Francisco, Oct. 1994.

[7] E.L. Hall, R.Y. Wong, and J. Rouge, “Hierarchical Search for Image Matching,” IEEE Decision and Con- trol Conference,1976, pp. 791-796.


. H. Wang and S.-F. Chang, “Adaptive Hierarchical


Image Matching in the Subband Domain,” Submitted to SPIE Sym-posium on Electronic Imaging 1996 S.-F. Chang, Compositing and Manipulation of Video Signals for Multimedia Network Video Services, Ph.D. Dissertation,U.C. Berkeley, Aug., 1993.

[10] R. W. Picard, “Content Access for Image/Video Coding: TheFourth Criterion,” MIT Media Lab. Per- ceptual Computing Section Technical Report, No.


[11] . M. K. Mandal, F. Idris and S. Panchanathan ,”Image and Video Indexing in the Compressed Domain:A Critical review”, Image and Vision Computing Jour-



First A. N.karthikeyan obtained his Mphil in Computer Science from Bharathidasan University, Tamil Nadu, and India. He is currently,Research Scholar in Bharathiyar Universi- ty,Coiambatore,&Head (B.C.A Dept), Department of Computer Applications, Srimad Andavan Arts&Science College, Affiliated to Bharathidasan University Tiruchitrapalli, Tamil Nadu, India. He has teaching, researches experience which includes 11 years of Private Sectors and colleges



Professor & Head Research Department of Computer Applications Easwari Engineering College Chennai – 600 089

Easwari Engineering College Chennai – 600 089 Phone:9940867665


Prof.Dr.R.Dhanapal obtained his Ph.D in Computer Science from Bharathidasan University, Tamil Nadu, India. He is currently Profes- sor & Head, Research Department of Computer Applications, SRM Easwari Engineering College, Affiliated to Anna University Chennai, Tamil Nadu, India. He has 25 years of teaching, research and administrative experi- ence which includes 21 years of Government Service. Besides being Professor, he is also a prolific writer, having authored twenty one books on various topics in Computer Sci- ence. His books have been prescribed as text books in Bharathidasan University and Autonomous colleges affiliated to Bharathidasan University. He has served as Chairman of Board of Studies in Computer Science of Bharathidasan Uni- versity, member of Board of Studies in Computer Science of several universities and autonomous colleges. Member of standing committee of Artificial Intelligence and Expert Sys- tems of IASTED, Canada and Senior Member of International Association of Computer Science and Information Technology (IACSIT), Singapore and member of International Association of Engineers, Hongkong. He has Visited USA, Japan, Malay- sia, and Singapore for presenting papers in the International conferences and to demonstrate the software developed by him. He is the recipient of the prestigious ‘Life-time Achieve- ment’ and ‘Excellence’ Awards instituted by Government of India. He served as Principal Investigator for UGC and AICTE, New Delhi funded innovative, major and minor research projects worth of 1.7 crore especially in the area of Intelligent systems, Data Mining and Soft Computing. He is the recognized su- pervisor for research programmes in Computer Science leading to Ph.D and MS by research in several universities including Anna University Chennai, Bharathiar University Coimbatore, Manonmaniam Sundaranar University Tirunelveli, Periyar University Salem, Mother Teresa Uni- versity Kodaikanal and many Deemed Universities. He has got 59 papers on his credit in international and national jour- nals. He has been serving as Editor In Chief for the Internation- al Journal of Research and Reviews in Artificial Intelligence (IJRRAI) United Kingdom and serving as reviewer and member of editorial in accredited peer reviewed national and interna- tional journals including Elsevier Journals