Generation of Random Fields for Object …2)/IJETTCS-2015-10...2015/10/12 · shadow removal, Inner-outer outline profile line (IOOPL) Generation of Random Fields for Object Recognization

International Journal of Emerging Trends & Technology in Computer Science (IJETTCS) Web Site: www.ijettcs.org Email: [email protected]

Volume 4, Issue 5(2), September - October 2015 ISSN 2278-6856

Volume 4, Issue 5(2), September – October 2015 Page 8

Abstract Object Recognization plays an important role in Computer Vision. In the literature many methods have existed but no method is suitable for object recognization with shadow, noise, and low contrasted images. In this paper, a new binarization technique has been proposed to recognize the object even with shadow, noise and low contrasted images. This technique contains three steps. First step is preprocessing, second step is Binarization and third step is Recognization. Preprocessing can be implemented using normalization, edge perseverance. In Binarization has been done by using phase congruency model and the then object recognization is done by optimum parameters. The experimental results show that the proposed method is more efficient for generation of random fields for object recognization than the existing IOOPL method. Keywords: Binarization, Segmentation, Thresholding, Object Recognization 1. INTRODUCTION Segmentation and Object Recognization plays an important role in Computer Vision. It is often used to partition an image into separate regions, which ideally corresponds to real world objects. It is a major step towards content analysis and image understanding. S.Lee, et al in the paper[1], presented a new multistage method using hierarchical clustering for unsupervised image classification based on Markov Random Field, but the spatial contextual information of MRF may result in incorrect classification for images with compound textual patterns. F.Salzenstein and Christophe Collet in their paper [2] presented the comparison of statistical models fuzzy markov random fields and fuzzy markov chains for multispectral image segmentation. Fuzzy markov chain model was developed in an unsupervised way and is described by a joint density compound on neighbored pixels and characterized by using the parametric (P-FMC) and nonparametric (NP-FMC) approach. These approaches model with a higher power behave efficiently on synthetic and astronomical images. Peter Orbanz and Joachim M. Buhmann [3], in their paper, presented a nonparametric Bayesian model for histogram clustering which automatically determines the number of segments when spatial smoothness constraints on the class assignments are enforced by a Markov Random Field. T.Hyung Kim, et al [4] in their paper, presented a wavelet-based texture segmentation method using

multilayer perceptron networks and Markov Random Fields in a multi-scale Bayesian framework. C.Naga Raju, et al [5] in their paper presented a high resolution image classification by generating binary random fields based on fuzzy semantic rules by means of descriptors such as form, texture and relations between objects and sub-objects. C.Naga Raju et al [6] in their paper presented a technique for the removal of narrow bands of misclassified pixels near boundaries by using generating random fields based on Markov Random Field theory. C.Nagaraju et al [7] proposed a new texture segmentation method using compound MRF, in which the label MRF and boundary MRF are coupled with gray level watershed method to improve the segmentation performance. C. Naga Raju, et al [8], presented a model-based approach to the automatic extraction of linear features in the image in two steps, in first step, utilizes local information related to the geometry and radiometry of the structures to be extracted using a series of morphological filtering stages. In second step, a segment linking process is carried out incorporating contextual, a priori knowledge about the object shape, with the use of Markov Random Field Theory. W.Zhou, et al [9], presented a comparison study of three methods for land cover classification of shades areas from high spatial resolution imagery in an urban environment. Method 1 combines spectral information in shaded areas with spatial information for shadow classification. Method 2 applies a shadow restoration technique, the linear-correlation correction method to create a “shadow-free” image before the classification. Method 3 uses multisource data fusion to aid in classification of shadows. Ch.Li, et al [10], proposed a new method by using independent component analysis (ICA) algorithm, grayscale histogram, RGB channels, HIS space transformation and multi-threshold retinex to achieve the shadow detection and compensation. H.Song, et al [11], proposed a novel shadow detection algorithm based on the morphological filtering and a novel shadow reconstruction algorithm based on the example learning method using MRF. H.Zhang, et al [12], proposed an object-oriented shadow detection and removal method. In this method, shadow features are taken into consideration during image segmentation, and then according to the statistical features of the images, suspected shadows are extracted and for shadow removal, Inner-outer outline profile line (IOOPL)

Generation of Random Fields for Object Recognization using Binarization Technique

P. Rambabu1,C. Naga Raju2

1Research Scholar, Department of Computer Science & Engineering,

Jawaharlal Nehru Technological University Kakinada, Kakinada.

2Associate Professor and Head, Department of Computer Science & Engineering, YSR Engineering College of Y U University, Proddatur.




matching is used. [13-16] provides different shadow detection and shadow removal techniques. J. Sauvola, et al [17], proposed a method for adaptive document image binarization, where the page is considered as a collection of subcomponents such as text, background and picture. The problem caused by noise, illumination and many source type-related degradations are addressed. B. Gatos, et al [18], addressed the advances in document image binarization using established evaluation performance measures. L. Laskov [19] presented an approach to document line segmentation based on a wavelet transform of the horizontal projective profile of the document image. K. Ntirogiannis, et al [20] addressed many challenges in handwritten document image binarization and proposed a new combination of global and local adaptive binarization method at connected component level that aims an improved an overall performance. B. Su, et al [21], proposed a novel document image binarization technique by using adaptive image contrast that addressed the segmentation of text from badly degraded document images. H. Z. Nafchi, et al [22], proposed a phase-based binarization model for ancient document images. [23-25], provided the information on different image binarization techniques. 2. EXISTING METHOD In existing system, generation of random fields for object reorganization is a two step process, Firstly, shadow features are taken into consideration during image segmentation, and then, according to the statistical features of the images, suspected shadows are extracted. Then secondly, for shadow removal, Inner-outer outline profile line (IOOPL) matching is used. 2.1 SHADOW DETECTION For shadow detection, a properly set threshold can separate shadow from nonshadow without too many pixels being misclassified. There are several different methods to find the threshold to accurately separate shadow and non-shadow areas. Bimodal histogram splitting provides a feasible way to find the threshold for shadow detection, and the mean of the two peaks is adopted as the threshold. H.Zhang et.al [12] attained the threshold according to the histogram of the original image and then finds the suspected shadow objects by comparing the threshold and grayscale average of each object obtained in segmentation.

H.Zhang et.al [12] chose the grayscale value with the minimum frequency in the neighborhood of the mean of the two peaks as the threshold

is the average grayscale value of the image; is the

left peak of the shadow in the histogram; is the threshold; is the neighborhood of , where

; and is the frequency of , where

(3) Where R is region, IR is the Inner-Region and OR is Outer-Region. Here random variables for object belongs any one of these regions. 2.2 SHADOW REMOVAL Shadows are removed by using the homogeneous sections obtained by line pair matching. Calculate the radiation parameter according to the homogeneous points of each object and then applies the relative radiation correction to each object. Let us assume that the inner homogeneous sections reflect the overall radiation of the single shadow. After obtaining the correction coefficient, all points of the shadow are corrected according to

Where stands for the pixel gray scale of the shadow after correction, stands for the pixel gray scale of the shadow before correction, and and are the coefficients of the minimum and maximum method or mean variance method calculated with the homogeneous points of the object, respectively. Where is calculated as

is calculated as

Where is the grayscale average of the inner homogeneous section at the waveband , is the grayscale average of the outer homogeneous sections at the waveband , is the standard deviation of inner homogeneous section at the corresponding waveband, and

is the standard deviation of the outer homogeneous sections at the corresponding waveband. By using this method, object was recognized for low complex images only. For high complexed images, the object is recognized partially and it also contains noise. Although image segmentation considering shadows can have better segmentation results, insufficient segmentation still exists. For example, a bird in water and its shadow cannot be separated. Also, parts of the shadow from low trees cannot be separated from the leaves. These problems are answered in our proposed method.




3. PROPOSED METHOD The generation of random fields for object recognization for the input image by using binarization method is contains three steps: 1)Preprocessing 2) Binarization and 3) Object Recognization as shown in Fig.2.

Fig 2: Flow chart of the proposed method

3.1 PREPROCESSING In the preprocessing step, To obtain a binarized image in rough form, we use a denoised image instead of the original image. The denoising process without changing the phase value, it determines the noise threshold at each scale and shrinking the magnitudes of the filter response vector appropriately. The most important part of denoising is automatic estimation of these thresholds using the statistics of the smallest filter respone. To estimate the distribution of the noise amplitude, these statistics are used. Then the noise distribution of other filter scales can be estimated proportionally. Rayleigh distribution can be used for modeling the distribution of noise energy

Where denotes the parameter of the Rayleigh distribution. The mean , the standard deviation , and the median of the Rayleigh distribution can be expressed based on .

By using the expected value of the magnitude response of the smallest filter scale, the median , and all the other parameters of the Rayleigh distribution can be estimated.

Which results in an estimation of

Suppose, denotes the mean and denotes the variance of the Rayleigh distribution, the noise threshold can be computed using

Noise responses from the smallest scale filter pair estimated and a noise threshold is obtained for each orientation. This noise response distribution is used to estimate the noise amplitude distribution of other filter scales using some constant. Finally, based on the noise thresholds obtained, the magnitudes of the filter response vector shrink appropriately, and they do so by soft thresholding, while leaving the phase unchanged. Normalized denoised image is obtained by applying a linear image transform on the denoised image and then Otsu’s method is applied on this normalized denoised image. Otsu [27] suggested minimizing the weighted sum of within-class variance of the object and background pixels to establish an optimum threshold. Minimization of within-class variance is equivalent to maximization of between-class variance. This method gives satisfactory results for bimodal histogram images. To measure the thresholding performance, a criterion measure is introduced by Otsu:

Where

Is the between-class variance which can be simplified to

The optimal threshold is given by maximizing , or equivalently maximizing , since is independent of

.

Because of the denoising method attempts to shrink the amplitude information of the noise component, Otsu’s method can remove noisy and degraded parts of images. The problem with this approach is that it misses weak edges, which means that we cannot get the exact output. To solve this problem, binarized image and edge map obtained using the canny operator [26] are combined. Canny operator is applied on the original image and for combination those edges without any reference in the aforementioned binarized image are removed. Then compute a convex hull image of the combined image. 3.2 BINARIZATION During this process, grayscale images were converted into binary images, which mean that every pixel in the image is converted to binary values (“0” and “1”). The maximum moment of phase congruency covariance feature of the phase congruency is used in the binarization process. Here, The backgrounds from the potential foreground parts are separated by using . This step works well even in badly degraded images, by using noise modeling method, majority of badly degraded background pixels are rejected. Generation of random variables for image is done by using the maximum moment of phase congruency covariance , can be defined as




The map is quantifying the edge strength. values are in the range , where a large value means a strong edge. Phase congruency is calculated as

Where is phase deviation function, is the local magnitude, is the noise threshold, is the phase congruency weighting mean function, In equation (19), is defined as

Where is the gradient vector and is the magnitude of the sobel operator and are calculated as

Then in equation (19), the noise threshold is of the form

Where is the number of standard deviations to be used to reject noise is estimated as

Where is a constant, is the binarization result of Ostu’s method on the input image. is the preprocessing step output. Here, the minimum probable value for is 2. Then in equation (19), is phase congruency weighting mean function, is constructed by applying a sigmoid function to the filter response spread

where is the filter response spread cut-off value, is a gain factor. Where is the fractional measure of spread, is calculated as

Where denotes the total number of filter scales. denotes the amplitude of the filter pair with minimum response. 3.3 OBJECT RECOGNIZATION After end of this step, the structure of object is determined. However, the image is still noisy, and edges have not been accurately binarized. Also, the binarization output is affected by some types of degradation. 4. QUALITY PARAMETERS 4.1 JACCARD INDEX The similarity measure is the Jaccard Index known as Jaccard similarity coefficient, very popular and frequently used as similarity indices for binary data. The area of overlap is calculated between the binary image and its corresponding gold standard image as shown in equation

If the threshold object and corresponding gold standard image (associated ground truth image) are exactly similar then the measure is 100 and measure 0 represents they are totally dissimilar, however the higher the measure indicates more similarity. 4.2 PEAK SIGNAL NOISE RATIO (PSNR) PSNR looks at how many pixels in the test image differ from the ground truth image values, and by how much. This metric is based more directly on the image difference and is calculated by

Where the Mean Square Error (MSE) is calculated from

A higher PSNR indicates a better match. 4.3 MEAN Mean is defined as the average gray levels of the image. The gray level image the Mean is calculated as

Where is the gray level image and and are number of rows and number of columns of resulting image. 4.4 STANDARD DEVIATION The standard deviation of gray level image is calculated as

Where is the gray level value of the image, and are the number of rows and number of columns of the resulting image and is the mean of the image. 5. RESULT ANALYSIS Few experiments are conducted to demonstrate the effectiveness and robustness of our proposed method. The proposed method is tested and compared with existing method over the 100 images dataset. The binarization performance is evaluated by using the Peak Signal to Noise Ration (PSNR), Jaccard Index (JI), Mean ( ) and Standard Deviation ( ). The dataset images contains degradations such as shadows, low contrast, noise, fog covered, variable background intensity, bleed-through, smear, smudge etc. which appear frequently in the image. We quantitatively compare our proposed method with existing IOOPL technique on dataset. Result Analysis of our proposed method is divided into two subsections. These are subjective evaluation and experimental evaluation. 5.1 SUBJECTIVE EVALUATION In this section, we present a subjective evaluation of our proposed method. We have run our method on 100 images. For result analysis we present only 20 images. In each case, we compare the results of the proposed method with the existing IOOPL method. Fig. 7 shows the original image, result of existing IOOPL method and result of the proposed binarization method. The first row to sixth row show low complex images in which the foreground – bird are made up of multiple




colors and textures and these images contains shadow, noise, low contrast and fog covering in the background. In these cases, our proposed binarization method works well and it recognizes the object perfectly, while the existing IOOPL method fails to recognize the object perfectly and it makes some mistakes, it includes some noise and shadows in the segmentation. The examples in the rows form 7 to 15 rows show medium complex images demonstrate situations in which the background is composed of shadows, noise, low contrast and fog covering in the background and foreground is bird, horse, man and elephant. The proposed binarization method, in these cases shows a few more artifacts than those in the previous rows. In particular the segmentation of elephant in Img 12 is not perfect. In all the cases, segmentations are quite good compared to the results of the existing IOOPL method, which containing more noise, shadows in the background. The example images in the rows from 16 to 20 rows show high complex information, which contains background information like trees, water, low contrast and shadows and foreground information like bird. In these cases, segmentation using proposed binarization method is quite good compared to the results of the existing IOOPL method, which contain more noise, shadow in the background of the segmented image. By observing the results of the existing and proposed methods, our proposed binarization method perfectly recognize the object by removing the noise, shadows even in low contrast and fog covered in the background compared to the existing IOOPL method. 5.2. EXPERIMENTAL EVALUATION Our method is evaluated with respect to the quality of the provided segmentation which illustrates that the proposed method is comparable in quality with other existing IOOPL technique. In this section we provided four evaluation methodologies. These are 1) Jaccard Index (JI) 2) Peak Signal to Noise Ration (PSNR), 3) Mean ( ) and 4) Standard Deviation ( ). From the experiments for each test image we obtain Jaccard Index (JI) computed from equation (27) is compared with standard image by different method including existing IOOPL method and proposed method is shown in Table 1. Fig. 3 shows the comparison of Jaccard Index (JI) of two different methods including existing IOOPL method and proposed method. The proposed method gives the optimum threshold value compared to the existing method. From the experiments for each test image we obtain Peak Signal to Noise Ration (PSNR) computed from equation (28) is compared with standard image by different method including existing IOOPL method and proposed method is shown in Table 2. Fig. 4 shows the comparison of Peak Signal to Noise Ration (PSNR) of two different methods including existing IOOPL method and proposed method. The proposed method gives the optimum threshold value compared to the existing method.

From the experiments for each test image we obtain Mean ( ) computed from equation (30) is compared with standard image by different method including existing IOOPL method and Proposed method is shown in Table 3. Fig. 5 shows the comparison of Mean ( ) of two different methods Table 1: Comparisons of the Jaccard Index for Different

Methods

Table 2: Comparisons of the PSNR for Different Methods

Table 3: Comparisons of Mean for the Different Methods




Table 4: Comparisons of Standard Deviation for the Different Methods

Fig. 7: From left to right, Original images and

corresponding results for the existing and proposed algorithm and histogram of the original image.

Fig 3: Comparisons of Jaccard Index for the Different

Methods

Fig 4: Comparisons of PSNR for the Different Methods

Fig 5: Comparisons of Mean for the Different Methods

Fig 6: Comparisons of Standard Deviation for the

Different Methods

including existing IOOPL method and proposed method. The proposed method gives the optimum threshold value compared to the existing method. From the experiments for each test image we obtain Standard Deviation (σ) computed from equation (31) is compared with standard image by different method including existing IOOPL method and Proposed method is shown in Table 4. Fig. 6 shows the comparison of Standard Deviation (σ) of two different methods including existing IOOPL method and proposed method. The proposed method gives the optimum threshold value compared to the existing method. The proposed method confirms the qualitative enhancement over the existing method. 6. CONCLUSION In this paper, a new method for generation of random fields for object recognization is proposed. The proposedtechnique is very useful for object recognization even the image contains noise, shadows, low contrast and fog covered background. The proposed method effectively removes the shadows in the image, noise in the image and identified the objects even for low contrast and fog




covered images. We evaluated performance of the proposed method by using 4 different performance measures like Jaccard Index, PSNR, Standard Deviation and Mean on image dataset. This method gives better results compared with shadow detection and removal technique using IOOPL. Even though this method produces better results, it fails remove noise in the images Img 10, Img 11, Img 16,Img 18, Img 19 and Img 20 and still some shadows are present in Img 9 and Img 15. In future work, we can overcome these problems and find the perfect object recognization. References [1]. Sanghoon Lee and Melba M.Crawford,

“Unsupervised Multistage Image Classification Using Hierarchical Clustering With a Bayesian Similarity Measure”, IEEE Transaction on Image Processing , Vol.14, No.3, pp312-320, March 2005

[2]. Fabin Salzenstein and Christophe Collet, “Fuzzy Markov Random Fields versus Chains for Multispectral Image Segmentation”, IEEE Transaction on Pattern Analysis and Machine Intelligence, Vol.28, No.11, pp 1753-1767, November 2006

[3]. Peter Orbanz and Joachim M.Buhmann, “Nonparametric Bayesian Image Segmentation”, International Journal of Computer Vision, Vol 77, pp25-45, 2008

[4]. T Hyung Kim, Kyu Eom and Y Shin Kim, “Multiscale Bayesian texture segmentation using neural networks and markov random fields” Neural Computers & Application,Vol 18,pp141-155, 2009

[5]. Naga Raju and L.S.S. Reddy, “Generation of Entropy Based Binary Random Fields for Image Boundary Detection Based on Fuzzy Semantic Rules”, IJCSNS, Vol.8, No.5, pp.297-300, May 2008

[6]. Naga Raju and L.S.S. Reddy, “Abnormality Detection By Generating Random Fields Based on Markov Random Field Theory”, IJCSNS, Vol.8, No.5, pp.260-263, May 2008

[7]. C. Naga Raju, P.S.V. Srinivasa Rao and G. Girija Rani, “A Novel Method for Generation of Random Fields for Boundary Detection and Classification of Images”, IJCSNS, Vol.8, No.7, pp.331-337, July 2008

[8]. C. Naga Raju , L. S. S .Reddy and D. Ramesh Babu, “Automatic Extraction of Linear Features Using Morphological Operations and Markov Random Fields”, International Journal of Information Theory and Knowledge Management, Vol.2, No.1, pp.5-8, June 2009.

[9]. W. Zhou, G. Huang, A. Troy and M. L. Cadenasso, “Object-based land cover classification of shaded areas in high spatial resolution imagery of urban areas: A comparison study”, Remote Sensing of Environment, Vol. 113, pp.1769-1777, 2009

[10]. Ch. Li, J. Yin, J. Zhao and F. Ye, “Detection and Compensation of Shadows based on ICA Algorithm in Remote Sensing Image”, International Journal of Advancements in Computing Technology, Vol.3, No.7, pp.46-54, August 2011

[11]. H. Song , B. Huang and K. Zhang, “Shadow Detection and Reconstruction in High-Resolution Satellite Images via Morphological Filtering and Example-Based Learning”, IEEE Transactions on Geoscience and Remote Sensing, accepted May 2013.,pp 1-10.

[12]. H. Zhang, K. Sun and W. Li, “ Object-Oriented Shadow Detection and Removal From Urban High-Resolution Remote Sensing Images”, IEEE Transactions on Geoscience and Remote Sensing, accepted January 2014, pp. 1-10

[13]. G. Gurav, M.B. Limkar and Sanjay M. Hundiwale, “Study of Different Shadow Detection and Removal Algorithm”, International Journal of Research in Electronics and Communication Technology, Vol.1, Issue 2, pp.26-29, April-June 2014.

[14]. Nagarathna B M, and Santhosh S, “Survey on Different Methods for Object-Orientation Based Shadow Detection and Removal Method from Urban High Resolution Elevated Objects”, ISRASE-2014, pp. 1-4.

[15]. Vishal G.M and P.U. Chati, “Shadow Detection Technique of Satellite Image for Shadow Removal”, International Journal of Advance Research in Computer Science and Management Studies, Vol.3, Issue.4, pp.176-179,April 2015

[16]. Harpreet Kaur and Navdeep Kaur, “Shadow Detection and Removal from Images – A Review”, International Journal of Computer Science and Information Technologies, Vol.6, Issue.2, pp.1755-1757, 2015.

[17]. J. Sauvola and M. Pietikainen, “Adaptive document image binarization”, The journal of the pattern recognition society, vol. 33, pp. 225-236, 2000

[18]. B. Gatos, K.Ntirogiannis and I. Pratikakis, “ICDAR 2009 Document Binarization Contest” 10th International Conference on Document Analysis and Recognization, pp.1375-1382, 2009

[19]. L. Laskov, “Application of wavelet decomposition to document line segmentation”, Serdica Journal of Computing, Vol. 6, pp.149-162, 2012

[20]. K. Ntirogiannis, B. Gatos and I. Pratikakis, “A combined approach for the binarization of handwritten document images”, Pattern Recognition Letters, 2012.

[21]. Bolan Su and S. Lu, “A Robust Document Image Binarization Technique for Degraded Document Images”, IEEE Transaction on Image Processing, 2012

[22]. H. Ziaei Nafchi, and R. F. Moghaddam, “Phase – Based Binarization of Ancient Document Images: Model and Applications”, IEEE Transactions on Image Processing, Vol. 23, No.7, pp.2916-2930, July 2014.

[23]. J. Kaur and Rajiv Mahajan, “A Review of Degraded Document Image Binarization Techniques”, International Journal of Advanced Research in Computer and Communication Engineering, Vol.3, Issue.5, pp. 6581-6586, May 2014




[24]. K. Ajitha and P. Kannan, “A survey on degraded document image binarization techniques”, International Journal of Advanced Research in Computer Engineering & Technology, Vol. 3, Issue. 11, pp. 3805-3808, November 2014

[25]. T. Kaur Gill, “Document Image Binarization Techniques – A Review”, International Journal of Computer Applications, Vol. 98, No. 12, pp.1-4, July 2014

[26]. J.Canny, “A computational approach to edge detection”, IEEE Trans. Pattern Anal. Mach. Intell., vol. 8, no. 6, pp. 679-698, Nov. 1986

[27]. N.Otsu “A Threshold Selection Method from Gray-Level Histogram”, IEEE Transactions on Systems, MAN, and Cybernetics, Vol. SMC-9, No. 1, January 1979.

AUTHORS

RAMBABU PEMULA, received his B.Tech Degree in Computer Science and Engineering from J.N.T.U, Hyderabad and M.Tech Degree in Software Engineering from J.N.T.U, Hyderabad.

Presently pursuing Ph.D in the area of Digital Image Processing in the Department of Computer Science and Engineering from J.N.T.University Kakinada. He is currently working as Assistant Professor and Head of the Department in Computer Science and Engineering, Nimra Institute of Engineering and Technology, Ongole, Andhra Pradesh. He got 8 years of teaching experience. He is member of various professional societies like IEEE, ACM and ISTE.

Dr. C. Naga Raju is currently working as Associate professor and Head of the Department of Computer Science and Engineering at YSR Engineering College of Yogi Vemana University,

Proddatur, YSR Kadapa District, Andhra Pradesh, India. He received his B.Tech Degree in Computer Science and Engineering from J.N.T.University, Anantapur, and M.Tech Degree in Computer Science from J.N.T.University Hyderabad and Ph.D in Digital Image Processing from J.N.T.University Hyderabad. He has got 18 years of teaching experience. He received research excellence award, teaching excellence award and Rayalaseema Vidhyaratna award for his credit. He wrote text book on C & Data Structures and Pattern Recognition. He has six Ph.D scholars. He has published fifty six research papers in various National and International Journals and about thirty research papers in various National and International Conferences. He has attended twenty seminars and workshops. He delivered 10 keynote addresses. He is member of various professional societies like IEEE, ISTE and CSI.

Documents

Generation of Random Fields for Object …2)/IJETTCS-2015-10...2015/10/12 · shadow removal, Inner-outer outline profile line (IOOPL) Generation of Random Fields for Object Recognization