(1)Inte rnatio nal Jo urnal o f Sc ie ntific & Eng inee ring Re se arc h, Vo lume 3, Issue 2, February -2012 1
ISS N 2229-5518
Morphological Texture Synthesis Algorithm Using
Pixel and Patch Based Approach
G. Venkata Rami Reddy, Dr. V.VijayaKumar, Dr. M. Anji Reddy
Keyw ords: pixel and patch-based linear searching, Hit-Miss Transf orm, region f illing.
—————————— ——————————
ne of the very important research areas in computer graphics and computer vision over years is the texture synthesis. Texture refers to the class of imagery that can
be categorized as a portion of an infinite pattern consisting of
stochastically repeating elements. The key behind texture syn-
thesis techniques is the inherent repeatability present in tex- tures. These techniques generate output textures that are larg- er in size than the input sample but perceptually similar to it. Texture synthesis techniques can be broadly divided into local region-growing methods and global optimization -based me- thods. In local methods the texture grow one pixel or patch at a time with the goal of maintaining coherence with nearby pixels in the growth region [1,2,3]. This leads to inconsisten- cies, in these approaches, because the small errors in the syn- thesized texture can accumulate over large distances. The dis- advantage of the local methods is they consume lot of time and hence they do not sufficiently meet real-time applications. The global methods evolve the entire texture as a whole, based on some criteria for evaluating similarity with the input.
Most existing global approaches either model complex
formulations that are difficult to optimize or model an only pixel-to-pixel interaction which leads to difficulties in captu r- ing large scale structures of the texture. [4,5,6,7]. Many re- searchers used the pyramid structure [4], tree-based accelerate mechanism [8] and multi-resolution approach [9,10, 11] to im- prove the efficiency of texture synthesis.
The Pixel-based synthesis algorithms are based on spatia l neighborhood methods which match across different frequen- cy bands [1, 2, 12, 13] and they grow an output texture pixel by pixel. These approaches are fit for stochastic textures, but usually fail on textures with more coherent structures. One of the methods which are generally more successful on synth e- sizing structural textures is path based methods [3, 14, 15, 16] which copy selected source regions into the output instead of single pixels. Some intermediate techniques [11, 17] between pixel and patch-based methods have also been presented, which somewhat combine the advantages of both.
The entire algorithm is explaned below.
Let Iin be an example texture sample as the input. Initialize the texture synthesis unit Iout to be a square block of user- specified size. The complete process to synthesis a new texture image Iout is given in thefollowing:
1. Take the original example image Iin, if the original image is color, converted into gray image.
2. First identify all the regions (R1, R2, R3, ….) of the image.
3. Identify seed pixel of each region using Hit-Miss Trans-
form.
4. Extract contours of R1.
5. Apply morphological thickening operation of extracted contour (R1) with suitable structuring element and the re- sult is stored in T1.
6. Extract contours of T1.
7. If the number of pixels in the contour of T1 is more than the number of pixels in contour of R1 then superimpose R1 with T1 else repeat step 5 using T1 with same structur- ing element and repeat step-7.
8. Take the selected seed pixel of the region R1; apply R e- gion-fill by using patch matching algorithm of the sup e- rimposed image i.e RT1.
9. Resulting RT1 image is placed into the synthesized image
Iout.
10. Repeat steps (4)-(9) till the whole image Iout are synthe- sized.
To detect the regions a reference image called a segment map is introduced, in the present approach, to make a correct selec- tion from the candidate pool, which corresponds to a coarse segmentation of the original image. As an initializ ation step, the whole image is segmented into several separate regions according to the texture similarity. The present paper adopted
IJSER © 2012
Inte rnatio nal Jo urnal o f Sc ie ntific & Eng inee ring Re se arc h, Vo lume 3, Issue 2, February -2012 2
ISS N 2229-5518
contours to achieve this segmentation. The contour based segmentation preserves the well connectivity between regions. This is very use full for patch filling. The entire process is represented by the equation 5. In the equation 5 the image is denoted by R, N represents the number of regions with each region as Ri, by the following Eqn. (1)
(1)
The Contour segmentation segments the original image with thin contours. To make thin contours as more visible thicken- ing operation is performed using morphology. The thickened image preserves the original shape. The thickening operation is calculated by translating the origin of the structuring el e- ment to each possible pixel position in the image, and at each such position comparing it with the underlying image pixels. If the foreground and background pixels in the structuring element exactly match, with the foreground and background pixels in the image, then the image pixel underneath the origin of the structuring element is set to foreground (one). Other- wise it is left unchanged. Thickening is the morphological dual of thinning and defined by the following Eqn. (2).
(2)
denotes thickening, thickening can be defined as a s e- quential operation as follows
A {B} = ((… ((AB1) B2)…) Bn)
In the fourth step to grow the size of the object contours are
formed for the thickened image. .
The proposed method searches for a seed point of a region and in the second part finds all pixels in the image that are connected with the seed point and provide them with a mark which indicates that they have been taken into account. The present paper initiates a new approach for detecting seed point by using Hit-Miss Transform (HMT). The HMT is a fun- damental Morphological Operation on binary images. Ma- thematical morphology is a well-founded non-linear theory of image processing [18, 19, 20, 21]. Its geometry-oriented nature provides an efficient framework for analyzing object shape characteristics such as size and connectivity, which are not easily accessed by linear approaches. Mathema tical morphol- ogy is theoretically founded on set theory. It contributes a wide range of operators to image processing, based on a few simple mathematical concepts. The operators are particularly useful for the analysis of binary images, classification, syn the- sis, boundary detection, noise removal, image enhancement, and image segmentation.HMT is a basic tool for shape detec- tion. Its main objective is to find the location(s) of the object of
———— ——— ——— ——— ———
Dr. V. VijayaKumar, Dean Dept.of Computers & HEAD SRRF-GIET,
GIET, Rajahmundry.vakulabharanam@yahoo.com
Dr. M. Anji Reddy, Prof. & Director University Foreign Relations,
JNTUH, Hyderabad.AP, India.mareddyanjireddy@hotmail.com
interest in a given image. The HMT uses a pair of contradic t- ing SEs (A, B), and looks for all positions where A can be fitted within a figure X, and B within the background Xc, in other words The morphological HMT transform is defined by the following Eqn. (3)
(3) 
- Denotes hit-or-miss transform AΘ X- Denotes erosion transforms,
W- Denotes a small window
One can generalize the notation somewhat by letting B = (B1,
B2), where B1 is the set formed from elements of B associated with an object and B2 is the set of elements of B associated with the corresponding background. From the preceding dis- cussion, by considering B1 = X and B2 = (W-X), then the equa- tion 3 becomes
) (4)
Thus, the set 
contains all the points of A at which B1 is
‗hit‘ and B2 is found a match in Ac i.e missing. By using the
definition of set differences and the dual relationship between erosion and dilation the equation 4 can be written as
) (5)
However, equation 4 is considerably more intuitive. We refer
to any of the preceding three equations as the morphological
HMT.
The present paper utilized a structuring element B1 associated
with objects and an element B2 associated with the bac k- ground based on an assumption that two or more objects are distinct only if they form disjoint (disconnected) sets. This is guaranteed by requiring that each object have at least a one- pixel-thick background around it.. For this a one-pixel-thick background operation is required in the HMT, that‘s why eq- uation 2 is adopted in the present paper for detection of seed points.
The target region is filled with patches from the source region possessing similar texture by using seed point reference. The patches are selected from the linear searching of the seed point reference. Once the reference seed of the corresponding region is identified then it is filled with the corresponding region of the original texture image by using patch based algorithm. This process is repeated to fill all the regions. This process is similar to patch matching in texture synthesis. In the present approach matching patch is achieved by searching the sample of texture along seed point from the position of synthesized patch in the sample.
The proposed method is tested on several color and gray i m- ages, some of which are taken from other studies in the litera- ture in order to make a comparison. The Fig.2 shows some texture synthesis results using the proposed algorithm. Pro- posed method works well for a wide range of textures. Table I show that the PSNR results of the proposed technique for s e- lected textures. Table II compares the PSNR performance of the proposed method with the existing methods i.e., seeding
IJSER © 2012
Inte rnatio nal Jo urnal o f Sc ie ntific & Eng inee ring Re se arc h, Vo lume 3, Issue 2, February -2012 3
ISS N 2229-5518
texture synthesis algorithm [22], image quilting for texture synthesis [3] and Graph cut Textures [14].Table II clearly shows that the PSNR value of the proposed method is high when compared to the all other methods.
(a) (b) (c) (d)
(e) (f ) (g) (h)
Fig.1 Input Textures: (a)-(e) are stone Textures, (f) straw berry (g) Litchie
(h) Dates.
Table1: PSNR results f or the proposed method
Texture Name | PSNR (dB) of Proposed Method |
Stone-1 | 32.29 |
Stone-2 | 27.54 |
Stone-3 | 29.6.4 |
Stone-4 | 26.50 |
Stone-5 | 31.25 |
Strawberry | 21.45 |
Litchie | 32.44 |
Dates | 35.19 |
Table2: PSNR results for the proposed and existing methods
Proposed Method | Existing Methods | |||
Texture Name | PSNR (dB) of Proposed Method | Seeding Algorithm | Image Quilting | Graphcut |
Stone-1 | 32.29 | 27.34 | 21.2 | 18.96 |
Stone-2 | 29.54 | 27.52 | 25.96 | 22.5 |
Stone-3 | 29.6.4 | 29.45 | 21.67 | 27.87 |
Stone-4 | 29.5 | 20.96 | 21.45 | 26.58 |
Stone-5 | 31.25 | 23.56 | 19.78 | 21.56 |
Strawberry | 29.45 | 19.67 | 15.54 | 23.87 |
Litchie | 32.44 | 24.43 | 21.46 | 22.8 |
Dates | 35.19 | 28.5 | 25.6 | 26.7 |
As demonstrated in this document, the numbering for sections upper case Arabic numerals , then upper case Arabic numerals, separated by periods. Initial paragraphs after the section title are not indented. Only the initial, introductory paragraph has a drop cap.
Fig.2 Synthesized Textures of above input images.
The proposed approach integrates the pixel and patch based local searching methods using morphological approach. The present paper initiates a new approach for detecting seed point by using Hit-Miss Transform (HMT). The morphological seed identification is useful for selection of the regions and also easy searching of filling the corresponding regions. The present paper presents a new texture synthesis algorithm fea- tured with a new search rule that searches matching patch using the idea of seed point. The main adva ntage of proposed algorithm is finding seeds are very fa st, very simple and also filling by seeds of the region is fast. The method can effectively and efficiently handle large complex and stochastic regions. The proposed algorithm is particularly effective for structured textures. The experimental results indicate the efficacy of the proposed method over the other method.
IJSER © 2012
Inte rnatio nal Jo urnal o f Sc ie ntific & Eng inee ring Re se arc h, Vo lume 3, Issue 2, February -2012 4
ISS N 2229-5518
I would like to thank to Prof. Rameswara Rao, Vice Cha n- cellor for encouraging research Programmes. The authors would like to express their gratitude to Sri K.V.V. Satyanara- yana Raju, Chairman, and Sri K. Sasi Kiran Varma, Mana ging Director, Chaitanya group of Institutions for providing neces- sary Infrastructure. Authors would like to thank the anon y- mous reviewers for their valuable comments.
[1] EFROS, A., AND LEUNG, T. 1999. Texture synthesis by nonparametric sampling. InIntl. Conf. ComputerVision, 1033–1038.
[2] WEI, L.-Y. 2001. Texture Synthesis by Fixed Neighborhood Searching. PhD
thesis, Stanford University.
[3] EFROS, A., AND FREEMAN, W. 2001. Image quilting for texture synthesis
and transfer. InSIGGRAPH‘01, 341–346.
[4] HEEGER, D,J., AND BERGEN, J. 1995. Pyramid-based texture analysis/
synthesis. InProc. of SIGGRAPH, 229–238.
[5] PAGET, R., AND LONGSTAFF, I. D. 1998. Texture synthesis via a noncausal nonparametric multiscale markov random field. IEEE Transactions on Image Processing 7, 6 (June), 925–931.
[6] PORTILLA, J., AND SIMONCELLI, E. P. 2000. A parametric texture model
based on joint statistics of complex wavelet coefficients. International Journal of ComputerVision 40, 1 (October), 49–70.
[7] FREEMAN, W. T., JONES, T. R., AND PASZTOR, E. C. 2002. Example based
super-resolution. IEEE Comput. Graph. Appl. 22, 2, 56–65.
[8] L. Y. Wei and M. Levoy, ―Fast texture synthesis using tree structured vector quantization,‖ inSIGGraph-00, 2000, pp. 479–488.
[9] S. Lefebvre, and H. Hoppe, ―Parallel Controllable Texture Synthesis,‖ ACM
Trans. on Graphics, pp. 777-786, 24(3), 2005.
[10] S. Lefebvre, and H. Hoppe, ―Appearance-Space Texture Synthesis,‖ ACM
Trans. on Graphics, pp.541-548,25(3), 2006.
[11] V. Kwatra, I. Essa, A. Bobick, and N. Kwatra, ―Texture Optimization for Ex- ample-based synthesis,‖ ACMTrans. on Graphics, pp. 795-802, 24(3), 2005.
[12] Bonet JSD (1997) Multiresolution sampling procedure for analysis and syn- thesis of texture images. In: SIGGRAPH ‘97: Proceedings of the 24th annual conference on Computer graphics and interactive techniques. ACM Press, New York, pp 361–368
[13] Ashikhmin M (2001) Synthesizing natural textures. In: SI3D ‘01: Proceedings of the 2001 symposium on interactive 3D graphics. ACM Press, New York, pp 217–226.
[14] Kwatra V, Schodl, A, Essa I, Turk G, Bobick A (2003) Graph cut textures:
image and video synthesis using graphcuts. ACMTransGraph22:277–286 [15] Wu Q, Yu Y (2004) Feature matching and deformation for texture synthesis.
ACMTrans Graph 23:364–367.
[16] Liu Y, LinWC, Hays J (2004) Near-regulartexture analysis and manipulation.
ACMTrans Graph 23:368–376.
[17] Nealen A, Alexa M (2003) Hybrid texture synthesis. In: EGRW ‘03: Proceed- ings of the 14th Euro-graphics workshop on rendering, Aire-la-Ville, Euro- graphics Association, Switzerland, pp 97–105.
[18] Dougherty E. and Astola J. ―An Introduction to Nonlinear Image
Processing,‖ Vol.16, SPIE Optical Engineering Press, Washington, 1994.
[19] Fisher R. ―Gasteratos Mathematical morphology operations and structuring
elements,‖ http://www.dai.ed.ac.uk/CVonline/transf.html.
[20] Giardina,C.R and Dougherty E.R ―Morphological Methods in Image and
Signal Processing‖ Englewood Cliffs, NJ: Prentice-Hall, 1988.
[21] Haralick R.M and Shapiro L.G ―An Introduction to Nonlinear Image
Processing,‖ Vol.16, SPIE Optical Engineering Press, Washington, 1994.
[22] Lijun Sun, Xueqing Li, ―Seeding Texture Synthesis Algorithm‖, IEEE confe- rence, 2010.
G.Venkata rami reddy received the
M.Tech. (CSE) degree from JNT Univer-
sity Hyderabad in 1998. He is working in JNT University since 2000. Presently he is working as an Ass ociate Professor in Dept of CSE in School of Information Technology, JNT University Hyderabad.
He is more than 11 years of experience in teaching and Soft-
ware Development. . He is pursuing his Ph.D. in the area of Image processing from JNT University Hyderabd in Compu t- er Science and Engineering under the guidance of Dr. M. Anji Reddy. He is presented more than 6 National and Internation- al journal and conference. His areas of interests are image processing, computer networks, analysis of algorithms.
Vakulabharanam Vijaya Kumar received integrated M.S. Engg, degree from Tash- kent Polytechnic Institute (USSR) in 1989. He received his Ph.D. degree in Comput- er Science from Jawaharlal Nehru Tech- nological University (JNTU) in 1998. He has served the JNT University for 13 years as Assistant Professor and Associate Pro- fessor and taught courses for M.Tech stu-
dents. He has been Dean for Dept of CSE and IT at Godavari
Institute of Engineering and Technology since April, 2007. His research interests include Image Processing, Pattern Recogn i- tion, Network Security, Steganography, Digital Watermarking, and Image retrieval. He is a life member for CSI, ISTE, IE, IRS, ACS and CS. He has published more than 150 research publi- cations in various National, Inter National conferences, pro- ceedings and Journals.
Dr. M. Anji Reddy received the M.Tech Civil Engineering from Indian Institute of Technol- ogy, Kanpur, India in 1989 and joined Jawa- harlal Nehru Technological University (JNTU), Hyderabad in 1989 as Assistant pro- fessor. He was awarded the doctoral degree
by JNTU in Remote Sensing, in 1995. He is having more than
22 years of teaching and research experience in Remote Sens- ing, Geoinformatics and Environmental Management. Present- ly he is working as a Professor and Director of University For- eign Relations, Jawaharlal Nehru Technological University Hyderabad. He has been the principal guide for more than 20
Ph.D projects and at present guiding more than 13 PhD stu-
dents in different areas.
IJSER © 2012