(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No.

9, December 2010

Nano-particle Characterization Using a Fast Hybrid Clustering Technique for TEM Images M.A. Abdou Researcher, IRI Mubarak City for Scientific Research and Technology Applications Alexandria, Egypt. m.abdou@pua.edu.eg Bayumy B.A. Youssef Researcher, IRI Mubarak City for Scientific Research and Technology Applications Alexandria, Egypt. bbayumy@yahoo.com W.M. Sheta Researcher, IRI Mubarak City for Scientific Research and Technology Applications Alexandria, Egypt wsheta@mucsat.sci.eg

Abstract- This Paper introduces a new fast Transmission Electron Microscopy (TEM) images clustering technique. Since analysis of particle sizes and shapes from two-dimensional TEM images is affected by variations in image contrast between adjacent particles, automatic methods requires more efforts. The proposed hybrid method consists of two main steps: automatic segmentation and nano-particles counting. The segmentation procedure begins with an automatic threshold generator and moves towards a high efficient multiple- regions segmentation technique. Results are observed, compared with existing methods and manual counting. Keywords: TEM, Image segmentation, Threshold generator, Nano-particle counting 1. INTRODUCTION (TEM) images are widely used in field of nanoparticle characterization. These images should be processed via a clustering technique to obtain the distribution of nano-particles on certain surface. Mean diameter can be measured either manually using a ruler or automatically through computer algorithm. Counting pixels belonging to every cluster, calculation of mean diameter are of great importance. Manual methods require extremely hard work and suffer leak of accuracy. Automatic methods, if used properly, will be easier and attain a mass production in this field. Many researches have been achieved concerning the TEM image analysis. Hideyuki et al. [1] have presented a study of construction a 3D geometric model of dermatan sulfate glycosaminoglycans and collagen fibrils, and to use the model to interpret TEM measurements of the spatial orientation and length of dermatan sulfate glycosaminoglycans in the medial collateral ligament of the human knee. This study shown how a 3D geometric model can be used to provide a priori information for interpretation of geometric measurements from 2D micrographs. Schaeublin et al. [2] presented the usage of TEM image simulations for couple the results from molecular dynamics simulations to experimental TEM images. David et al. [3] synthesized discrete single-element semiconductor nano-wires and multicomposition nano-wire hetero-structures, and then characterized their structure and composition using high-resolution electron microscopy and analytical electron

microscopy techniques. Chinthaka et al. [4] have made a study where the nano-sized fluorapatite particles were synthesized using a precipitation method and the material was characterized using X-ray diffraction and transmission electron microscopy (TEM). Yeng-Ming et al. [5] studied the effect of hydrophobic molecules on the morphology of aqueous solutions of amphiphilic block copolymer, which has potential drug delivery applications. Using cryogenic TEM observations, micelles can clearly be visualized and their core size measured. Kurt et al. [6] demonstrated that TEM techniques are focusing on the determination of parameters, such as shape and size of islands. A successful image contrast analysis in terms of shape and strain demands the application of image simulation techniques based on the many-beam dynamical theory and on structure models refined by molecular dynamics or molecular static energy minimization. Kenta et al. [7] developed a spherical aberration corrected TEM technique that allowed them to obtain clearer images in real space than ever before. They applied this technique to titanium oxide, in which light elements such as oxygen are difficult to observe using TEM because of its small cross section and electronic damage. Wang et al. [8] examined the mean diameter of the PtRu nanoparticles using TEM. Jae et al. [9] investigated the catalysts by employing various physicochemical analyses: X-ray diffraction, TEM and extended X-ray absorption fine structure to investigate the structural modification, and X-ray photoelectron spectroscopy and X-ray absorptionnear-edge spectroscopy to characterize the change in electronic features. The data processing was performed with XPSPEAK software program. Because of the lack of Tem image processing, the MRI and Gel image processing were considered in survey. Because both MRI and Gel images are look like TEM images, whereas they have grey level color and the data exists in cluster spots within image background. Many researches have been achieved in the Tem image processing (clustering through image segmentation). Atkins et al. [10] used thresholding and morphology techniques, combined with an anisotropic diffusion process to localize and segment the brain. Ravinda et al. [11] proposed a similar approach. Hahn and Peitgen. [12] proposed a solely intensity-based watershed algorithm, which makes use of a simple merging criterion to avoid the over segmentation problem. Kapur et al. [12] proposed a hybrid approach that

   
101 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

used morphological operations and active contour segmentation. Shattuck et al. [14] used adaptive anisotropic diffusion, edge detection and morphological erosions to identify the brain component. Xu et al. [15] introduced the deformation of the active surface under a gradient vector field computed from a binary edge map. Zeng et al. [15] used a coupled surface evolution to extract bounding surfaces of the cortex. Kim et. al. [17] proposed a hierarchical segmentation based on thresholding and the detection of watersheds. They first pre-processed the images to remove noise and enhance contrast, and then thresholding was applied. Takahashi et. al. [18] achieved image enhancement and smoothing based on the definition of threshold values, before defining local maxima in order to label the spots. Umer et al. [19], presented a technique that uses the clustering techniques like K-mean and fuzzy C-mean to distinguish between different types of protein spots and unwanted artifacts. Christopher et al. [20] presented a new technique using the labeling of each image pixel as either a spot or non-spot and used a Markov Random Field model and simulated annealing for inference. Neighboring spot labels were then connected to form spot regions. Feature extraction is usually based on computing commonly used features: mean variance, coefficient of correlation, contrast, homogeneity, skew, and kurtosis [21]. Texture segmentation has been improved by the use of co-occurrence matrices [22, 23]. As the “texture” contained in our electronic microscopic images is not of a regularly repeating variety, it is not clear whether these features would help in segmenting the images manually. Moreover, the seven features used in [21] for segmentation when applied on electronic microscopic images lead to large calculations and high complexity. In this paper, we will present a hybrid feature extraction method based on wavelet transform and pixel texture parameters for microscopic image regions. Those used parameters are: pixel intensity, mean, and variance. The proposed technique is described in the incoming section. 2. HYBRID TECHNIQUE Although it may suffer from real time problems, texture- based algorithms are essential in different sections (segmentation, registration, denoising… etc.). As previously described, the commonly used features in almost all texture based segmentation methods are: mean, variance, coefficient of correlation, contrast, and homogeneity [21]. Taking all those features leads to large computations and thus more complexity. In this paper, we present an automatic hybrid method for feature extraction. Based on wavelet transform, pixel intensity, image mean, and variance this proposed technique is applied to images acquired

from an electronic microscope. In this type of images, features are of great importance. Moreover, errors are not accepted as applications are very restricted to scientific and research purposes. This algorithm deals with TEM grayscale images. These images are acquired by electronic microscope which are used for characterize the material nanostructure. The proposed hybrid segmentation method consists of two main steps: (1). an automatic threshold generator (ATG) (2). High efficient multipleregions segmentation filter. 2.1. AUTOMATIC THRESHOLD GENERATOR (ATG) The proposed method starts with an automatic filter threshold generation. This ATG aims to get exact threshold values used for electronic microscopic input image segmentation. To perform this task, the following step actions are considered: • An input image histogram (contrast and homogeneity) is first generated for each input electronic microscopic image data • Image histogram impulse and false data pixels removal is then applied to reduce errors and increase accuracy of threshold selection • The input image histogram obtained is fitted to a suitable probability distribution function (PDF) to detect its mean and lobe width • A histogram fingerprint for each input image is then defined 2.1.1. HISTOGRAM GENERATOR The input image is preprocessed first to enhance its quality. Histogram shows the intensity distribution within the input image. Image histogram serves in observation of homogeneity or non- homogeneity in different image areas, and thus thresholds decision making within the next step, as shown in Fig 1.

2500

2000

1500

1000

500

0 0 50 100 150 200 250

Fig 1: Input image histogram shape.

   
102 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

2.1.2. HISTOGRAM IMPULSE DATA FILTRATION An impulse data filter is used to omit random values within the histogram data to avoid presence of false critical points, which in turns lead to a false threshold selection. 2.1.3. HISTOGRAM FITTING METHOD Observing Fig 1 for the set of input images acquired from the electronic microscope, it could be easily concluded that their histogram data could be fitted to a suitable PDF curve. Two distributions are selected according to the histogram of each input image: the Normal distribution and the Poisson distribution, as shown in “(1)”,”(2)” respectively. The advantage of this fitting method is that both distributions have known mean and widths.

N1 and N 2 , as in “(3)”. Fig 2 shows the
histogram fingerprint details.

if

n BW = k − 1 2 n If K> BW = n − k 2 And N1 = fingerpr int( K − BW ) K< N 2 = fingerpr int( K + BW )

(3)

I =

σ 2π

e

−( x − μ ) 2 2σ 2

Applying this proposed threshold selection method on a wide data set, it has shown impressive results except when the center of the main lobe (K) lies at either one of the ends of the histogram fingerprint. To overcome this problem, we will introduce a binary constant, the ‘deviation factor’ that will be used in both cases for better values of N1 and N 2 . Equations (4) summarizes the effect of this constant:

(1)

I=

λ e −λ
k!

(2)

If K = 1 or 2 BW = K - 1 N1 = fingerprint(K − BW ± deviation) N 2 = fingerprint(K + BW ) K = n or (n − 1) BW = n - K N1 = fingerprint(K − BW ) If N 2 = fingerprint(K + BW ± deviation)
This binary constant moreover add an integer factor of safety called "Deviation Factor" that could be varied (4)

The mean and bandwidth of this fitted probability distribution function is recorded. Furthermore, side lobes are to be taken into consideration by critical points determination; critical points are those pixels where maximum or inflection points are found. Gradient methods are applied to the fitted curve to detect these critical points. Results of this gradient method are: centers of side lobes, and inflection points: {(I1, ∆1), (I2, ∆2), (I3, ∆3) …}; Where I and ∆ represent the lobe center and lobe width respectively. Since the percentage of infected pixels is usually small enough, infected pixels within the input image are usually away fro nnn the detected lobes (especially the main lobe). Obviously this histogram analysis leads to a good selection of segmentation thresholds or a successful ATG.

BW k

n
2.1.4. FEATURE EXTRACTION An exact method is used to determine threshold values, “Histogram Fingerprint”. What is the meaning of a histogram fingerprint? It is a simple array that contains the exact places and values of histogram critical points (main lobe center- main lobe width- side lobes centers- side lobes widths). Observing these values, we can design the suitable filter accurate parameters: center K and bandwidth Histogram Fingerprint

Consider K as the order of the main lobe center, and n as the length of the fingerprint
Fig 2 : Histogram Fingerprint.

   
103 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

2.2.

THE HYBRID ATG- SEGMENTATION PHASE

A region of interest (ROI) is one area or multiple areas of an image of remarkable importance or distinct features from the rest of the image. ROIs are usually filtered for further investigations. Several methods were proposed in this area either for a single ROI [24] or multiple ROIs filtration [25]. We will define existing ROIs by creating one ‘binary mask’- a binary image that is the same size as the input image with pixels that define the ROI set to 1 and all other pixels set to 0- for all regions. Infected regions or ROIs are selected according to following procedure: • Outputs of the histogram fingerprint analyzer (described in the previous section) are first obtained; • A binary mask is obtained; • Input image is filtered according to this mask Fig 3 shows the block diagram of the whole system. It should be noticed that input data set suffers a major segmentation problem; random occurrence of regions. In the incoming section, we will apply this proposed segmentation procedure on a wide data set and observe the results. Moreover, we will verify our proposed methods to segment electronic microscopic images with unexpected number and shapes of regions of interest.

method mainly depends on getting certain relation between each pixel and its neighbors. The stopping criterion starts when the clustering results remain unchanged during further iterations. This counting yields to: (1). Start with two arrays (A, B) that both have a number of elements corresponding image pixels. (2). Fill these arrays with a large integer number, that must be greater than the expected particles number (M) (3). Put a variable N as counter for particle Number (4). Start N=0 (5). Calculate every array elements according the next relation

Bi , j = Min ( Ai +1, j , Ai −1, j , Ai , j +1 , Ai , j −1 )
(6). If

Ai , j is equal to the larger number M

then

N = N + 1 and Ai , j = N

(7). If Ai , j is less than M , Ai , j remains unchanged (8). Getting

r = A− B

Input Image

(9). If r equals zero, the solution is accomplished. Else repeat steps 5,6,7,8,9 after putting A = B 2.4. PARTICLE DIAMETER

 
Histogram   Generator

Impulse Data Filter

Histogram

The area of particle is denoted by (S), S is calculated by multiplying the number of pixels contained in the particle by the scale factor. If the particle area S is considered as circle the diameter (D) can be calculated according next relation

Output Regions

Threshold – Based Segmentation

D=2
Where

S

π π
is equal 3.14

Fig 3: The proposed hybrid ATG- Segmentation method.

2.3.

PIXEL COUNTING METHOD

3. 3.1.

RESULTS AND DISCUSSION THE SAMPLE RESOURCES

Output of the previous section is a binary image, where nano-particles are represented by black pixels and the background is represented by the white areas. Clustering of those black pixels is preformed taking into consideration the connectivity between them. The connectivity will be accomplished using an iterative method. The

The samples of TEM are obtained from obtain from the literatures surveying such as Ref. [8] and Ref. [26] as shown in Fig 4 (a) (b) (c) (d) (e). They represent a variety of image of TEM image with different conditions.

   
104 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

(a)

(b)

(c)

(d)

(e)

Fig 4 Sample Case studies chosen from [8 and 26].

3.2. RESULTS OF THE PROPOSED THRESHOLD- BASED SEGMENTATION METHOD It is applied to a wide data set. Images acquired from an electronic microscope are all used. Fig 5 shows segmentation method by applying threshold values .

acquired when fitting the histogram curve to normal and Poisson PDF respectively. Observing several results, it can be concluded that the histogram curve fitted to normal distribution will give more realistic results. All incoming tests are performed by fitting to normal distribution

(a)

(b)

(c)

Fig 5 : Segmentation method with threshold values acquired by fitting the histogram curve to normal and Poisson PDF respectively (a) TEM image (b) TEM image processed according to Normal distribution (c) TEM image processed according to Poisson distribution. Filter edges (N1=80 and N2=190) grayscale level.

Fig 6 shows the processed TEM image of Fig 4 (a) [8] using the proposed algorithm and the associated particle size histogram. This case was chosen as an example of TEM images with a partially unclear background and the particles have some variation of intensity; hence, the histogram fingerprint

distribution indicates one main lobe at 148, the filter edges are at (N1=110 and N2=190) and the deviation factor are equal to zero. Fig 7 shows the processed TEM image of Fig 4 (b) [8] processed image using our algorithm and the associated particle size histogram. In this case, the histogram fingerprint

   
105 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

distribution indicates one main lobe at 148, the filter edges are at (N1=128 and N2=244) and the deviation factor are equal to zero, 510 particles could be counted against manually counted of nearly 600 particles. Fig 8 shows the processed TEM image of Fig 4 (c) [8] and the associated particle size histogram. This case was chosen as an example of

TEM images with a partially unclear background and the particles have some variation of intensity; hence, the histogram fingerprint distribution indicates one main lobe at 174, the filter edges are at (N1=174 and N2=220) grayscale level and the deviation factor are equal to zero. 70 particles could be counted against manually counted of 79 particles.

30% 25% Frequency % 20% 15% 10% 5% 0% 2 3 4 5 6 7 8 9 10 11 14

Diameter (nm)

(a)
.
35% 30% Frequency % 25% 20% 15% 10% 5% 0% 5

(b)
Fig 6: (a) Processed TEM image of Fig 4 (a), (b) associated particle size histogram

6

7

8

9

10

11.5

Diameter (nm)

(a)

(b)
Fig 7: (a) Processed TEM image of Fig 4 (a) (b) associated particls size histogram.

30% 25% Frequency % 20% 15% 10% 5% 0% 0.5 1 1.5 2 2.5 3 3.5 Diameter (nm) 4 5 5.5 6.5

(a)

(b)

Fig 8 : (a) Processed TEM image of Fig 4 (c), (b) associated particls size histogram.

 

106

http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

Fig 9 shows the processed TEM image of Fig 4 (d) [8] and the associated particle size histogram. This case was chosen as an example of TEM images with a unclear background and the particles have some variation of intensity; hence, the histogram

fingerprint distribution indicates one main lobe at 174, the filter edges are at (N1=114 and N2=244) grayscale level and the deviation factor are equal to one. 116 particles could be counted against manually counted of nearly 150 particles.

30% 25% Frequency % 20% 15% 10% 5% 0% 1.5 2 2.5 3 4 4.5 5

Diameter (nm)

(a) Fig 10 shows the processed TEM image of Fig 4 (e) [26] and the associated particle size histogram. This case was chosen as an example of TEM images with a clear background but the particles have big variation of intensity; hence, the histogram fingerprint distribution indicates one main lobe at

(b) 174, the filter edges are at (N1=128 and N2=244) grayscale level and the deviation factor are equal to zero. 315 particles could be counted against 369 particles counted in [26]. When the counted manually were found 333 particles.

Fig 9: (a) Processed TEM image of Fig 4 (d) using the proposed algorithm (b) Associated particls size histogram.

80 70 60 Frequency 50 40 30 20 10 0 1.1 1.3 1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3.1 3.3 3.5 3.7 Diameter in (nm)

(a)

(b)

Fig 10: (a) Processed TEM image of Fig 4 (e) using the proposed algorithm (b) associated particls size histogram.

3.3. RESULTS COMPARISON The comparison has been preformed based on the results of Refs. [8, 26]. The case of Fig 4 (c) from Ref. [8] and the case of Fig 4 (e) from [26] are chosen to perform the comparison. Ref [8] has proposed a manual method to count and size the nano

particles where were obtained over 100 particles in randomly chosen areas not over the all image. The Fig 11 shows the particle size histogram of [8] against the particle size histogram using our algorithm. The results which included in Fig 11 (a) are not reasonable because as we see in the figure the

   
107 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

summation of the frequency percentage over the all

registered particle size is nearly equal to 135%.

(a) Ref. [26] has proposed that the contrast and darkness of the separate sections of the TEM image were adjusted in Adobe Photoshop prior to the analyzing of particles by NIH-Image. The Fig 12 shows the particle size histogram of Ref. [26] against the particle size histogram using our algorithm. 315 particles could be counted against 369 particles

(b)
Fig 11: Particle size histogram (a) of Ref. [8] (b) The proposed algorithm.

counted in [26] and 333 particles manually counted. The results which included in Fig 12 gives an indication that the particles size has almost the same distribution trend. Fig 12 (a) shows that the standard deviation and the mean diameter are 0.437 nm and 2.4 nm respectively while Fig 12 (a) shows 0.54 nm of standard deviation and 2.3 nm of mean diameter.

80 70 60 Frequency 50 40 30 20 10 0 1.1 1.3 1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3.1 3.3 3.5 3.7 Diameter in (nm)

(a) 4. CONCLUSION

(b) and proposed techniques shows the impact of our work. REFERENCES

Fig 12: Particle size histogram (a) Ref. [26] (b) The proposed algorithm.

The presented method for nano-particles size characterization in TEM using the fast hybrid automatic threshold segmentation and counting, shows promising results. Since the algorithm is performed automatically without human interaction, fidelity and better accuracy are both achieved. Moreover, this fast algorithm gives more advantage in execution time. Many cases from [8 and 26] have been tested; comparison between manual, existing,

[1]. Heath B. Henninger, Steve A. Maas , Clayton
J. Underwood, Ross T. Whitaker, JeVrey A. Weiss “Spatial distribution and orientation of dermatan sulfate in human medial collateral ligament” Journal of Structural Biology, PP13, 6 November 2006.

   
108 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

[2]. R. Schaeublin, M.-J. Caturla, M. Wall, T.
Felter, M. Fluss,B.D. Wirth, T. Diaz de la Rubia, M. Victoria “Correlating TEM images of damage in irradiated materials to molecular dynamics simulations”, Journal of Nuclear Materials 307–311, 988–992, 2002. [3]. David C. Bell, Yue Wu, Carl J. Barrelet, Silvija Gradecak, Jie Xiang, Brian P. Timko, Charles M. Lieber “Imaging and Analysis of Nanowires”, Microscopy Research And Technique 64:373–389 (2004). [4]. G.W. Chinthaka Silva, Longzhou Ma, Oliver Hemmers, Dennis Lindle “Micro-structural characterization of precipitation-synthesized fluorapatite nano-material by transmission electron microscopy using different sample preparation techniques”, Science Direct ,Micron (the international research and review journal for microscopy), vol.39 No. 3, 269– 274,(2008). [5]. Yeng-Ming Lam, Gerhard Goldbeck-Wood, Chris Boothroyd “Mesoscale Simulation and cryo-TEM of Nanoscale DrugDelivery Systems” Molecular Simulation, Vol. 30 (4), pp. 239–247 , 15April 2004. [6]. Kurt Scheerschmidt and Peter Werner “Analysis of Nanostructures by EM Techniques: Quantum Dots”, EUREM 12, Brno, Czech Republic, July 9-14, 2000. [7]. Kenta Yoshida, Tomoyuki Kawai, Takahiro Nambara, Sakae Tanemura, Koh Saitoh, Nobuo Tanaka “Direct observation of oxygen atoms in rutile titanium dioxide by spherical aberration corrected high-resolution transmission electron microscopy”, Nanotechnology 17 3944–3950, 2006. [8]. Wang Y.C., Chou T.M., Libera M. “Preparation of carbon-supported PtRu nanoparticles for direct methanol fuel cell applications – a comparative study”, Journal of Power Sources ,Vol. 142, Pp. 43–49 , 2005. [9]. Jae H. Kim, Sung M. Choi, Sang H. Nam, Min H. Seo, Sun H. Choi, Won B. Kim, “Influence of Sn content on PtSn/C catalysts for electrooxidation of C1–C3 alcohols: Synthesis, characterization, and electrocatalytic activity”, Applied Catalysis B: Environmental,Vol. 82, Pp. 89–102, 2008. [10]. Atkins, M.S., Mackiewich, B.T., Fully automatic segmentation of the brain in MRI. IEEE Trans. Med. Imaging, Vol. 17, Pp. 98– 107. ,1998.

[11]. Ravinda G. N. Meegama, Jagath C. Rajapakse
“Fully Automated Peeling Technique For T1Weighted, High-Quality MR Head Scans”, International Journal of Image and Graphics (IJIG), Volume: 4, Issue: 2, Pp. 141-156,2004. Horst K. Hahn, Heinz-Otto Peitgen . “The Skull Stripping Problem in MRI Solved by a Single 3D Watershed Transform.”, Proc. MICCAI, LNCS 1935: Pp. 134-143, Springer, Berlin, 2000. Tina Kapur, W. Eric L. Grimson, William M. Wells, Ron Kikinis2 “Segmentation of Brain Tissue from Magnetic Resonance Images.” Medical Image Analysis volume 1, number 2, Pp. 109–127, 1996. David W. Shattuck, Stephanie R. SandorLeahy, Kirt A. Schaper, David A. Rottenberg, Richard M. Leahy “Magnetic resonance image tissue classification using a partial volume model.”, NeuroImage 13 (5), Pp. 856– 876, 2001. Xu C., Pham D. L., Rettmann M. E., Yu D. N., Prince J. L., “Reconstruction of the Human Cerebral Cortex from Magnetic Resonance Images”, IEEE Transactions on Medical Imaging, 18(6), pp. 467-480, June, 1999. Zeng, X., Lawrence H. Staib, Robert T. Schultz, James S. Duncan. “Segmentation and Measurement of the Cortex from 3D MR Images Using Coupled Surfaces Propagation”. IEEE Transactions On Medical Imaging, Vol. 18, No. 10, October 1999. Kim Y., Kim J., Won Y., Yongho In, "Segmentation Of Protein Spots in 2d Gel Electrophoresis Images With Watersheds Using Hierarchical Threshold" , LNCS Computer and Information Sciences - ISCIS 2003, 2869:389-396, 2003. Takahashi, K., Nakazawa M., Watanabe Y., "Dnainsight: An Image Processing System For 2-D Gel Electrophoresis Of Genomic DNA", Genome Informatics, Vol. 8, Pp.135-146, 1997. Umer Z. Ijaz , Safee U. Chaudhary, Moon S. Don, Kyung Y. Kim, "Computational Strategies for Protein Quantitation in 2D Electrophoresis Gel Image Processor for Matlab", Proceedings of the 2007 Frontiers in the Convergence of Bioscience and Information Technologies, 129-134, 2007. Christopher S. Hoeflich , Jason J. Corso, "Segmentation of 2D Gel Electrophoresis Spots Using a Markov Random Field", In

[12].

[13].

[14].

[15].

[16].

[17].

[18].

[19].

[20].

   
109 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

(IJCSIS) International Journal of Computer Science and Information Security, Vol. 8, No. 9, December 2010

[21].

[22].

[23]. [24].

[25].

[26].

Proceedings of SPIE Conference on Medical Imaging, 2009. A. Solberg and A. Jain, “Texture fusion and feature selection applied to SAR imagery,” IEEE Transactions on Geoscience and Remote Sensing, vol. 35, no. 2, pp. 475–479, 3 1997. D. Geman, C. Geman, C. Graffigne, and D. Pong, “Boundary detection by constrained optimization,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 12, no. 7, pp. 609–628, Jul 1990. R.M. Haralick, “Statistical and structural approaches to texture,” Proc. IEEE, vol. 67, no. 5, pp. 786–804, May 1979. M. Abdou, M. Tayel, “Automated Biomedical System for Image Segmentation and Transmission,” The International Journal of Robotics and Automation, Vol. 23, Issue 1, 2008. M. Abdou, “An Introduced Blocked-Wavelet Algorithm for Signal and Time Improvements of Images with Multiple Regions,” CGIM 2008, 10th IASTED International Conference on Computer Graphics and Imaging, Innsbruck, Austria, pp.34- 39, Feb. 2008. H. Woehrle, E. Hutchison, S.Ozkar2, G. Finke “Analysis of Nanoparticle Transmission Electron Microscopy Data Using a PublicDomain Image-Processing Program, Image” © Tubitak ,Turk J Chem vol. (30) P.p. 113,(2006).

____________________________________________________________________ 

 

Dr. Bayumy A. B. Youssef is a researcher in computer graphics department, Information technology institute, Mubarak City for Scientific Research and Technological Applications, Alexandria, Egypt. He received Bs.c. and Ms.c. degrees from Mechanical Engineering department, faculty of Engineering, Cairo University, Egypt. He received his Ph.D degree from the Information Technology department, institute of graduate studies and research, Alexandria University 2004. His research interests include Numerical modeling and Simulation, Visualization of Scientific Data, Mesh generation product, Investigating Conservation and Restoration of Culture Heritage and (2D & 3D) Image Processing (Segmentation, Restoration, &registration) __________________________________________________ Walaa M. Sheta is an associate professor of Computer graphics in Informatics Research Institute at Mubarak city for Scientific Research (MUCSAT) since 2006. During 20012006 he has worked as Assistant professor at MUCSAT. He holds a visiting professor position at University of Louisville in US and University of Salford in UK. He advised approximately 20 master’s and doctoral graduates, his research contributions and consulting spans the areas of real-time computer graphics, Human computer Interaction, Distributed Virtual Environment and 3D image processing. He participated and led many national and multinational research funded projects. He received M.Sc. and PhD in Information Technology from University of Alexandria, in 1993 and 2000, respectively. He received B.Sc. from Faculty of Science, University of Alexandria in 1989

AUTHORS PROFILE _________________________________________
Dr.  Mohamed  Abd­ElRahman  Abdou  is  an  assistant  professor  in  Communications  Engineering.  He  obtained  his  PhD  in  2006  when  he  started  a  tenure  track  as  a  researcher  at  Mubarak  City  for  Sciences  &  Technology.  In  2008  he  joined  Pharos  University  in  Alexandria  as  an  assistant  professor in the EE department. His  research interests are in the area of  image  and  signal  processing,  biomedical  engineering,  and  knowledge  based  systems  (fuzzy  control‐  Neural  Network‐  cellular automata). He published his first international book in  February 2010 in medical image processing.  His is  a reviewer  of several international journals: ELSEVIER, IASTED, WSEAS               

   
110 http://sites.google.com/site/ijcsis/ ISSN 1947-5500

Sign up to vote on this title
UsefulNot useful

Master Your Semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master Your Semester with a Special Offer from Scribd & The New York Times

Cancel anytime.