Department of Electronics and Electrical Egineering

Browse

Now showing 1 - 20 of 232

Efficient watershed algorithms for image segmentation and related prototype architectures
(2004) Rambabu, C
Image segmentation is concerned with decomposing a given image into its constituent regions or objects. It is an important preliminary step in diverse fields of application like object recognition, image compression, medical image processing and biological analysis methods. Among the existing segmentation algorithm, watershed transformation is deemed to be a powerful tool for image segmentation owing to the simplicity of its formulation and implementation and its ability to identify the important closed contours of a given image. However, the time complexity of the majority of the watershed transform algorithms is quite high making their real- time application difficult. In particular, real time processes like moving object segmentation, road traffic monitoring and analysis of steel fracture demand fast computation of watershed transformation for image segmentation. At the same time a dedicated hardware architecture for implementing watershed algorithms would give rise to faster results as compared to a software program executed on a general purpose processor.
Vision based dynamic hand gesture recognition for human computer interaction
(2006) Bhuyan, Manas Kamal
With the increased interest in human-computer interaction (HCI), there has been rapid growth of research related to gesture recognition in recent years. Hand gesture recognition from visual images forms an important part of this research. This thesis reports on our research on recognition of dynamic hand gestures having different spatio-temporal characteristics. We develop methods to recognize dynamic hand gestures with (1) local hand motion only where only the fingers and the palm move without any movement of the whole hand, (2) global motion only where the hand as a whole move in space to make different gestures, and (3) both local and global motions where the fingers and palm create different hand poses as the arm traverses along a trajectory in space. We use the concept of object-based video abstraction for segmenting the frames into video object planes (VOPs) with each VOP corresponding to one semantically meaningful hand pose or position in space. A gesture is represented as a sequence of key video object planes (key VOPs) that correspond to significantly different VOPs in the vide sequence.
New developments in median filters
(2006) Singh, Khumanthem Manglem
The median filter is a nonlinear filter that outputs the median of the data inside a moving window of predetermined length and has been traditionally used to remove impulse noise from images. The filter is known for its ability to preserve edges while smoothing a noisy signal. One of the major drawbacks of the median filter is that it tends to alter pixels undisturbed by noise and disturb features like thin lines, because it is implemented invariantly across the image. One way to circumvent this problem is to detect impulse noise prior to the filtering operation. In this work, two new types of median filters that employ impulse detection mechanisms prior to filtering operations are proposed. These are the rank-conditioning and thresholding (RCT) median filters and the noncausal linear prediction error based median filters. The rank-conditioned and threshold median filters are based on simultaneous application of rank-conditioning and the thresholding of the deviation of the center pixel from healthy pixels in the window. Simultaneous application of these conditions is aimed at reducing the probability of a healthy pixel being detected as noisy. Rank-conditioning is kept fixed as one of the conditions - the rank of the centre pixel inside the sliding window does not lie in the trimming range if it is an impulse. Three different thresholding conditions are investigated. In RCT-I filter, the absolute difference of the centre pixel from the median is compared against a pre-defined threshold. The RCT-II filter employs the thresholding scheme of Chen and Wu's adaptive centre-weighted median filter. The thresholding scheme of signal-dependent rank-ordered mean filter is used in RCT-III filter. The corrupted pixel is replaced by the rank-ordered mean. ..
Motion- coherent video watermarking
(2007) Vinod, P.
One of the challenging security issues in video watermarking is the inter-frame collusion attacks. These attacks exploit the inherent redundancy in the video frames or in the watermarks to produce an unwatermarked copy of the video. The frame temporal filtering(FTF) is the basic inter-frame collusion attack in which temporal low-pass filtering is applied to the watermarked frames to remove temporally uncorrelated watermarks.
Feature analysis and compensation for speaker recognition under stressed condition
(2007) Raja, G. Senthil
This thesis documents our investigations on feature analysis and design of compensators for speaker recognition under stressed condition. Any condition that causes a speaker to vary his or her speech production from normal to neutral condition is called stressed speech condition. Stressed speech is induced by emotion, high workload, sleep deprivation, frustration and environmental noise. In stressed condition the characteristics of speech signal are different from that of normal or neutral condition. Due to changes in speech signal characteristics the performance of speaker recognition system may degrade under stressed speech conditions. In this work the problem of speaker recognition under stressed condition has been dealt with three broad approaches. First, six speech feature (mel- frequency cepstral coefficients (MFCC) linear prediction (LP) coefficients, linear predictions cepstral (LPCC), reflection coefficients (RC), arc-sin reflection coefficients (ARC) and log area ratio (LAR)) which are widely used for speaker recognition are analysed for evaluation of their characteristics under stressed condition. Statistical technics such as Probability density function (pdf) characteristics, F-ratio test and Kolmogorov-Smirnov (K-S) test are used for evaluation of speaker discrimination capability of the features under stressed condition. Two classifiers, Vector Quantisation (VQ) classifier and Gaussian Mixture Model (GMM) are used to evaluate the speaker recognition results with different speech features.
Limit cycle approach for identification and control of processes
(2007) Padhy, Prabin Kumar
The thesis is concerned with identification and control of single-input-single-output (SISO) and two-input-two-out (TITI) processes based on relay induced limit cycles. A relay based on-line identification method for stable and unstable SISO processes is proposed where the process model parameters are estimated from a single relay test. An ideal relay in parallel with a proportional-integral-derivative (PID) controller induces a limit cycle oscillation whose frequency and amplitude are used for the process identification. Firstly, the describing function (DF) analysis is used for the.......
Controller Design Methods for Linear systems with Emphasis on Integrating Processes
(2008) Ali, Ahmad
Certain new approaches for designing efficient controllers for linear systems, in particular, for integrating processes, are proposed in this thesis. The parameters of the proposed proportional integral- derivative (PID) controllers and its variants are optimized by minimizing the integral square error (ISE) using the so called bacterial foraging algorithm (BFA). Although ISE is a standard minimization criterion as well as BFA has certain advantages such as better solution quality and good convergence, both of them suffer from some drawbacks. To obtain the robust closed loop performance, the ISE criterion is initially modified by appending time parameters to the said criterion. Similarly, to ensure the convergence of BFA, an adaptive strategy for the run length vector along with two modifications in the standard BFA are also proposed. Next, the parameters of the PID and its variants are estimated using the proposed Dslope of the Nyquist curvesD technique. The approach of the said slope of the Nyquist curves method is also extended to design controllers for the modified Smith predictors. Finally, the parameters of the PID and its variants are derived analytically using the user-defined percentage overshoot and process model parameters...
Optimization of Planar Spiral Inductor and Design of Multilayer Pyramidal Inductor for Silicon Radio Frequency Integrated Circuits
(2009) Haobijam, Genemala
Silicon integrated passive devices have been gaining importance with the need to integrate more functionalities on a single chip to realize the complex systems on chip. .....
Cardiovascular Signal Compression Using a New Wavelet Energy Based Diagnostic Distortion Measure
(2009) Manikandan, M. Sabarimalai
This thesis presents an adaptive wavelet based compression method for cardiovascular signals. Generally, the performance of the lossy compression system depends on the methodologies used for compression and the quality measure used for evaluation of distortion. However, in order to introduce a closed loop rate or quality control, one needs an adequate distortion measure since it plays an important role in rate-distortion optimization technique used for finding optimal coding parameters. Generally, compression method with two-stage design employs global wavelet thresholding followed by fixed linear quantization approach. But this may introduce a severe signal distortion since a subband consists of wavelet coefficients with great magnitude differences and exhibits varying dynamic range according to the signal characteristics. Therefore, first an adaptive wavelet compression approach based on the preprocessing step, multiresolution signal decomposition (MSD) technique, classification of wavelet coefficients, constraint threshold control zero-zone nearly uniform midtread quantization (TCZNUMQ), modified index coding (MIC) and Huffman coding schemes is proposed in this work. Generally, the amplitude distribution of wavelet coefficients of most ECG signals has sharp concentrations around zero in their distributions, and the relevant wavelet coefficients of the signal contents appear very close in a sequence order within a wavelet subband. The proposed approach exploits the above properties using TCZNUMQ and MIC schemes for achieving substantial improvements in compression performance. In this approach, the wavelet coefficients are classified into frames based on the statistics of subband coefficients for providing better quantization. The classified coefficients are then quantized using the constraint TCZNUMQ scheme in an adaptive manner. In this quantizer design, the zero-zone width is defined by the threshold parameter T for wavelet thresholding and the outer-zone width is chosen according to the distortion specification. A constraint on the TCZNUMQ scheme is studied to further reduce the computational cost of the conventional two-stage scheme. Since indexes or location of the nonzero wavelet coefficients are in sequence order, the modified index coding scheme is used to compress the integer significance map by exploiting the redundancy among the indexes. The performance of compression method is tested using the well-known MIT-BIH arrhythmia (mita) database which contains varying characteristics of various ECG signals and different noises. The effect of noise filtering is one of the features in the wavelet transform based ECG signal compression. In such a case, smoothing of low-level background noise of the ECG signal causes a large percentage root mean square difference (PRD) value but no clinical feature distortion and, conversely, a small average distortion may severely deteriorate clinical performance if the error is concentrated in the regions of significant features. Moreover, noise present in the input decreases compression rate of the coder since the coder spends extra bits on approximating the noise for a user-specified PRD with a desired accuracy.
Adaptive Network based Fuzzy Inference System (ANFIS) as a Tool for System Identification with Special Emphasis on Training Data Minimization
(2009) Buragohain, Mrinal
Nearly two decades back nonlinear system identification consisted of several ad-hoc approaches which were restricted to a very limited class of systems. However, with the advent of the various soft computing methodologies like neural networks, the fuzzy logic and the genetic algorithm combined with modern structure optimization techniques, a wider class of systems can be handled at present. Complex systems may be of diverse characteristics and nature. These systems may be linear or nonlinear, continuous or discrete, time varying or time invariant, static or dynamic, short term or long term, central or distributed, predictable or unpredictable, ill or well defined. System outputs may be measurable or unmeasurable. Models of real systems are of fundamental importance in virtually all disciplines and hence there is a strong demand for advanced modeling, identification and controlling schemes. This is because models help in system analysis which in turn help to get a better understanding of the system for predicting or simulating a systemDs behavior. Also, system models facilitate application and validation of advanced techniques for controller design. Development of new processes and analysis of the existing ones along with their optimization, supervision, fault detection, and component diagnosis are all based on the models of the systems. As most of the real world systems are nonlinear in nature, an endeavor is made for modeling a nonlinear system in the present work. A linear system is considered to be a special case of the nonlinear system. The challenges involved in modeling, identification and control of a nonlinear system are too many and attempt has been made to tackle them by applying various soft computing methodologies. In most of the conventional soft computing methods the system modelling results are dependent on the number of training data used. It has been found that the modeling results improve as the number of training data increases. But in many complex systems the number of available training data are less and the generation of new data is also not cost effective. In such a scenario the system has to be modelled with the available data. The proposed modeling scheme hasD The results obtained by applying the proposed technique are comparable and in some cases superior to those obtained by using the conventional neuro-fuzzy model. D Comparable or superior results are obtained with this proposed model even though the number of data pairs used for system modeling here are less as compared to that used in the conventional methods. D It resulted in reduction of the number of computations involved. As the experiments were performed by using reduced number of specifically chosen data, the number of computations required to be performed also came down. been devised keeping such a possibility in mind. The results obtained by applying this proposed model are compared with the results obtained by using various statistical and genetic algorithm based fuzzy models and finally the relative merits and demerits involved with the respective models are discussed. The work embodied in the present thesis is concerned with optimal design of the conventionally existing soft computing based system models. The statistics based Full factorial design (FFD) and the V-fold cross validation technique are applied to...
Combined Temporal and Spectral processing methods for speech Enhancement
(2009) Krishnamoorthy, P.
This thesis proposes a combined temporal and spectral processing(TSP) approach for the enhancement fo degraded speech. There major sources of degradation, namely, background noise, reverberation and speech from the competing speakers are considered in this work. ..
Limited Data Speaker Recognition
(2009) Jayanna, H S
This work demonstrates the task of recognizing the speaker with the constraint of limited data. The performance of the speaker recognition system depends on the techniques employed in the analysis, feature extraction, modelling and testing stages. Existing limited data speaker recognition techniques mostly concentrate on modelling techniques to improve the performance. It is also possible to improve the performance using efficient techniques for speech analysis, feature extraction, modelling and testing. We have developed techniques for each stage of the speaker recognition system to improve the performance. In the analysis stage, speech signal is analyzed using Single Frame Size and Rate (SFSR), Multiple Frame Size (MFS), Multiple Frame Rate (MFR) and Multiple Frame Size and Rate (MFSR) analysis techniques. For this study, theMel Frequency Cepstral Coefficients (MFCC) are used as features and Vector Quantization (VQ) as modeling technique. In order to verify the effectiveness of various techniques, we carried out initial experiment using 3 sec training and testing data for a set of first 30 speakers taken from the YOHO database. The SFSR, MFS, MFR and MFSR provide 70%, 73%, 87% and 90% identification performance, respectively. The experiments are later extended for other data sizes and databases. The same trend is observed in these experiments also. Thus this study demonstrates that the performance of speaker recognition under limited data condition can be improved using MFSR analysis technique. In the feature extraction stage, different feature extraction techniques like MFCC, Delta MFCC, Delta-Delta MFCC, Linear Prediction Residual (LPR), Linear Pre- diction Residual Phase (LPRP) and their combinations are explored. For this study, SFSR is used as analysis technique and VQ as modeling technique. The combina- tion of MFCC, Delta MFCC, Delta-Delta MFCC, LPR and LPRP features provided 83% performance against 70% for only MFCC in the initial experiment. The same trend is observed for other data sizes and databases. This infers that the combi- nation of features is effective for improving the performance of speaker recognition under limited data condition. In the modeling stage, experimental evaluation of the modelling techniques like Crisp Vector Quantization (CVQ), Fuzzy Vector Quantization (FVQ), Self-Organizing Map (SOM), Learning Vector Quantization (LVQ), GaussianMixtureModel (GMM) and Gaussian Mixture Model-Universal Background Model (GMM-UBM) is made. In addition, the combined classifiers evaluation is also made on the basis of exper- imental knowledge of individual modeling techniques. This includes 1) LVQ and FVQ, 2) LVQ and GMM, and 3) LVQ and GMM-UBM classifiers. In this study, SFSR analysis technique is used for extracting the MFCC features. The combined LVQ and GMM-UBM classifier provides 87% performance against 70% for CVQ in the initial experiment. The same experimental set up is extended for other data sizes and databases and observed the same trend. This study demonstrates that the combined LVQ and GMM-UBM modelling can be used for speaker recognition to improve its performance under limited data condition. The above studies are made independently. That is, proposed technique is used in the respective stage and the existing techniques in the remaining stages of the speaker recognition system. To analyze the effectiveness of proposed techniques, integrated systems are...
Modelling of multi-antenna wireless channels and relay based communication systems
(2009) Paul, Babu Sena
Different issues related to modelling of multi-antenna wireless channels and relay based communication systems have been investigated in this thesis. The effects of array geometry and orientation of arrays in aMIMO system have been investigated in the frame work of geometrically based one ring scattering model for a macrocellular scenario. A technique based on scattering parameters has been developed to find the channel matrix for a MIMO system, modelled geometrically from a microwave perspective, employing suitably terminated antennas acting as scatterers. The performance of a 2D2 MIMO system has also been evaluated taking mutual coupling between the antenna elements into consideration. Using geometrical based single bounce modelling, characteristics of mobile-to-mobile communication channel has been studied and analytical expressions has been derived for probability density function of angle of arrival and time of arrival of signals in terms of model parameters. A virtual MIMO system in the form of two-hop relay channels constituting two diversity paths has been investigated for bit error rate performance with selection and maximal ratio combining techniques applied at the receiver. The works reported in this thesis are expected to contribute towards the better understanding of certain issues related to modelling of multi-antenna wireless communication channels and relay based systems...
Design of Robust Fuzzy Controllers for Uncertain Nonlinear Systems
(2010) Senthil, Kumar D
This thesis deals with the analysis and design of robust fuzzy controllers for uncertain nonlinear systems using Takagi-Sugeno (T-S) model based approach. A T-S fuzzy model is used here to approximate the uncertain nonlinear systems where the nominal model and uncertain terms of the consequent parts of the fuzzy model are identified by a linear programming approach and then they are expressed in a form suitable for robust fuzzy controller design. With the derived T-S fuzzy model, various types of robust fuzzy controllers are designed that guarantee not only stability but also satisfy the specified performance criteria of the closed-loop control system. The first type of T-S controller is a robust fuzzy guaranteed cost controller for trajectory tracking in uncertain nonlinear systems. The fixed Lyapunov function based approach is used to develop the robust controller and the design conditions are derived as a problem of solving a set of linear matrix inequalities (LMIs). Next, this research work focuses on robust stabilization, robust H1 stabilization and robust H1 tracking control of uncertain nonlinear systems by using a richer class of Lyapunov function called parametric Lyapunov function. This parametric Lyapunov function based approach attempts to reduce the conservatism associated in the controller design for nonlinear systems with slowly varying uncertainties. The design conditions are derived as matrix inequality involving parametric uncertainties and then they are reduced to finite dimensional matrix inequalities by using the multiconvexity concept. These matrix inequalities are then solved by an iterative LMI based algorithm. Finally, the results of standard state-space T-S fuzzy system with parametric Lyapunov function based approach are extended to synthesize the robust controller for application to uncertain fuzzy descriptor systems...
Perceptual Hashing for Wavelet-Based Scalably-Coded Video
(2010) Saikia, Navajit
A perceptual hash function for video extracts a Dxed-length binary string called the perceptual hash on the basis of the perceptual content of the video. Besides being sensitive to the content diDerences in videos, a perceptual hash function should be robust against the content-preserving operations on the videos. Recent developments in the Deld of scalable video coding (SVC) demands the robustness of the perceptual hash against the scalability features of SVC. The 3D discrete wavelet transform (3D-DWT) is a way of achieving scalable coding, wherein the inherent multi-resolution structure of the 3D-DWT is exploited. This thesis deals with content-based representation and hashing of video using the 3D-DWT for the use in the wavelet-based SVC (WSVC). This thesis Drst considers extracting representative frames for video using the 3D-DWT. It ex- amines the representation of the content of a video at the group-of-frames (GOF) level by the bands of the 3D-DWT decomposition. The spatio-temporal low-pass band at the full level of temporal and an intermediate level of spatial decomposition of a GOF is used for representing the content of the GOF. Experimental results show the eDectiveness of the band in representing the content of the GOF. Two perceptual hash functions are extracted from the perceptually-representative spatio-temporal low-pass band. For this purpose, the band is divided into perceptual blocks that are sensitive to local contents of the GOF. The Drst hash function derives a hash of the GOF by binarising the wavelet coeDcients in each perceptual block. The similarity between two GOFs is measured in terms of the maximum Hamming distance between the hashes of the corresponding perceptual blocks. Experi- mental results show that the hash function is robust against the scalability features of WSVC and other content-preserving operations, and sensitive to content diDerences at the frame and GOF levels. The hash function has limitations of a large hash size and weak confusion and diDusion properties. The second hash function computes a compact hash by binarising the forward and backward cumulative averages of the local means of the perceptual blocks in the spatio-temporal low-pass band. Experimental results show the robustness of the hash functions against the scalability features of WSVC and other content-preserving operations, and the sensitivity to the content diDerences at the frame and GOF levels. This hash function is shown to have good diDusion and confusion properties....
Image Denoising using Sparse and over complete representations
(2011) Deka, Bhabesh
An image is corrupted by noise during its acquisition, transmission or storage. Noise degrades the image quality and interpretability. The aim of image denoising is to remove the distortion resulted by the noise while keeping the detail features in the image intact. The sparse denoising methods represent a class of denoising algorithms based on the principle that a natural signal like the image is sparse over some basis set but the noise is not. This has led to the widespread application of these methods in image denoising. Recent works in the literature focus on the use of sparse and overcomplete representations to develop state-of-the-art image denoising techniques. The thesis addresses the problem of image denoising by the sparse and overcomplete representations. One of the recent additions to the greedy solution to the sparse representation problem is the Bayesian Pursuit Algorithm (BPA). The thesis critically examines the algorithm and proposes a modiDcation by integrating a new initialization scheme and a stopping criterion to enhance its per- formance. The modiDed BPA converges to the true solution much faster than the standard BPA. It is also applied for the removal of additive white Gaussian noise from gray-scale images. Experimental results demonstrate that the modiDed BPA successfully removes the noise at low noise levels. The thesis proposes two sparse denoising techniques based on the overcomplete dictionary for the removal of the impulse noise and the speckle noise. The novel sparse reconstruction Dlter is proposed for the removal of the impulse noise based on the principle of signal recovery by compressed sensing. The impulse-free pixels in the image are identiDed by an impulse detection algorithm and are used to reconstruct the noisy pixels using the orthogonal matching pursuit (OMP) algorithm with an overcomplete dictionary. The Dlter performs better than selected state-of-the-art impulse denoising Dlters. The thesis explores removal of the speckle noise from medical ultrasound images using the sparse representation on an overcomplete dictionary. Many of the denoising algorithms including the ones based on the sparse representation of the signal require the noise to be additive, white and Gaussian. However, this assumption fails for speckle in the medical ultrasound images in the log-transform domain. The thesis adopts an existing technique for the decorrelation and the Gaussianization of the speckle in the log-transform domain and the resulting image is denoised by using sparse and overcomplete representations. Two denoising methods are applied on the preprocessed ultrasound images in the log-transform domain. They are: (1) the undecimated wavelet transform (UDWT) based soft thresholding and (2) the OMP-based sparse representation. Both the methods show improvements over the existing techniques for despeckling of medical ultrasound images...
Statistocal Modeling of Lapped Transforma Coefficients and its Applications : Statistical Modeling of Lapped Transform Coefficients and its applications
(2011) Nath, Vijay Kumar
In the last two decades, statistical modeling in wavelet domain has been an active research area due to its multiresolution and time frequency localization properties. These properties have been effciently exploited in many image processing applications like image compression, denoising, deblocking, deblurring etc with considerable success. The main drawback of wavelet transform is that it is not very good in capturing the directional information present in natural images. The Lapped Transforms have been proposed to overcome the blocking artifacts of the DCT with increased coding gain. It was observed that the Lapped Transforms (LT) based methods are very good in preserving oscillatory components present in images like textures. This thesis deals with statistical modeling of LT coefcients of natural images and its applications. The thesis first considers an exhaustive study on the determination of a suitable statistical distribution that best approximates the block Lapped Orthogonal Transform (LOT) and Lapped Bi-orthogonal Transform (LBT) coefcients. The widely used Kolomogrov-Smirnov (KS) and Chisquare goodness of it tests indicate that the Generalized Gaussian is the most appropriate statistical distribution that best approximates the block LOT and LBT coefcients of natural images. Such a study is very useful in the design of optimal quantizers that may lead to minimum distortion. Employing the dyadic remapping feature of LTs, the LT coefcients can be rearranged into octave-like representation. The rearranged LT coefcients in various detailed subbands show highly non Gaussian statistics and can be modeled in a way similar to wavelet coefcients. The dyadic remapped LT coefcients when used in compression and denoising applications show performance comparable to that of wavelet based methods. This thesis next considers an exhaustive study on the determination of a suitable statistical distribution that best models the dyadic remapped LT coefcients. The experimental results indicate that the Generalized Gaussian distribution best approximates the dyadic remapped LT coe cients. Such a study plays an important role in developing more e cient algorithms for compression and denoising applications. The problem of reducing additive white Gaussian noise in LOT domain is considered in the thesis. The main motivation is that LOT has good energy compaction, it is robust to oversmoothing and the noise and the signal statistics can be modeled precisely in the LT domain. We propose three LT based image denoising methods based on statistical modeling of dyadic rearranged LT coe cients. The rst method uses a Bayesian minimum mean square error (MMSE) estimator based on modeling the global distribution of dyadic rearranged LT coe cients by Generalized Gaussian distribution. The second method assumes the local distribution of the rearranged LT coe cients to be Gaussian with spatially varying variance and applies local Wiener lter to reduce the noise. Based on the encouraging performance of single local Wiener ltering in LT domain, a doubly local Wiener ltering framework is developed in the same domain. The third method employs a maximum a posteriori (MAP) estimator which uses the local Laplace probability density function with local variance for the estimation of the noise free coe cients. Experiment....
Left Ventricular Wall Detection from MRI Scans using Random Walk
(2011) Dakua, Sarada Prasad
The left ventricle (LV) is one of four chambers (two atria and two ventricles) in the human heart. Because it supplies oxygenated blood to the entire body, it possesses a stance of great significance in the medical diagnosis. In order to know the problem against any chest pain, physicians first try to look to the shape of the LV. Therefore, its segmentation always remains as the first and foremost step. The accuracy of any segmentation is usually achieved through lessening the user interaction. A semi-automatic method for image segmentation that uses random walks was introduced in 2006. This approach requires user guidance to define the desired content to be extracted in the image. This is fast, effective and intuitive which can successfully perform segmentation irrespective of the image type. At the same time, its performance depends largely on the degree of homogeneity and separability of the objects present in the image. The lesser is the obscurity, unlike the cardiac magnetic resonance images, the better is its execution. Since myocardium (thick layer of cardiac muscle) is responsible for the contraction and relaxation of the ventricle, its inner (endocardium) and outer (epicardium) lining should be extracted simultaneously. The main contributions of this thesis are as follows: 1) In endocardial wall detection, Random Walk approach faces many challenges while applying this algorithm on ischemic cardiac magnetic resonance (CMR) images (as they are more obscure), one of them is initial seed(s) selection. Furthermore, the free parameter D in this algorithm does have the influence on its performance, which is usually decided by the user. In order to reduce the user interaction (because more user interaction introduces larger variability in the performance), attempts have been made to solve these two issues in our research and made the algorithm automatic, 2) The grey level distribution of myocardial muscle does not differ much from the outer surrounding muscles, therefore, Random Walk approach is little suitable for its meaningful segmentation. Instead, a modified active contour model that operates on the endocardial boundary is applied to section out the epicardium, 3) The methodologies adapted in endocardial wall detection do not produce satisfactory result, especially in case of short axis CMR images. After a thorough review of the nature of weighting function, three functions for the same purpose have been suggested to achieve better segmentations, and finally 4) To compare the performances of these weighting functions, an unsupervised technique has been proposed in the end..
Design of amplifiers and fabrication of high performance thin-film transistors using carbon nanotubes
(2011) Narasimhamurthy, K C
The work carried out in this thesis is divided into three parts. The first part discusses the study on carbon nanotubes (CNT) interconnects. The second part will be dealing with the modelling of CNT based transistors and design of analog circuits based on CNTs. The final part will present the fabricated thin film transistors based on single walled carbon nanotubes (SWCNTs). The study on the SWCNT interconnects is focused on the estimation of their magnetic inductance at various bias voltages. The analysis of magnetic inductance is carried out for the ground-signal-ground (GSG) configuration of SWCNT based interconnects having various dimensions and different percentage of metallic SWCNT (m-SWCNT) purities. The result indicates a variation in the loop inductance value as high as 34% for closely spaced semi-global interconnects. The study on the SWCNT transistor modelling aims to develop close-form equations for the drain current and drain to source voltage for CNT field effect transistor (CNFET) in terms of its dimensions. Although these proposed models are based on curve fitting method, they provide a quick first order numerical estimate of drain current and drain to source voltage of the CNFET.
Fast Mode Decision Methods in H.264/AVC Standard
(2011) Ganguly, Amrita
The H.264/AVC video coding standard oDers signiDcantly improved compression eD- ciency and Dexibility compared to previous standards. H.264/AVC provides better video at a lower bit rate and this translates into lesser storage requirement. H.264/AVC com- pression makes it possible to transmit HD television over a limited-capacity broadcast channel. H.264/AVC is a sophisticated compression method. However, for all its advan- tages, it must be acknowledged that it requires more processing power to encode video with H.264/AVC as compared to the previous standards. The standard is complex and its implementation is more challenging. The computationally expensive H.264/AVC coder can lead to higher encoding and decoding times. This thesis presents some new methods for reduction of the encoding time and the complexity of the H.264/ AVC encoder. A detailed study of the intra and inter prediction processes was carried out. Based on the observations gleaned from this study, a number of fast mode decision algorithms are proposed in the spatial domain and in the transform domain. Natural videos have objects with strong edges. Homogeneous region in a frame can be detected from the edges in it. This characteristic of the video is utilized for a new algorithm for intra and inter prediction process based on the edge histogram characteristic of the image. The reduction in complexity is achieved by selecting only a few coding modes from the total available modes depending upon the edge histogram characteristics and the homogeneity of the image. There are many regions in a video that are stationary. These regions generally get encoded in the SKIP mode. SKIP mode is simple to compute and if this mode can be detected prior to performing the inter prediction process, large computational saving is possible in the encoder. Thus, a low complexity algorithm is developed by predicting the \skipped"" macroblocks prior to motion estimation through early detection of the zero quantized coeDcients. A weighted prediction algorithm is next proposed that identiDes a macroblock based on certain parameters. Weights are assigned for these parameters and the encoding modes are selected accordingly. These algorithms are developed in the spatial domain. Transform domain based fast encoding algorithms are proposed where the statistics of the quantized transform coeDcients are utilized for the mode decision process. The energy of the transform coeDcients are usually concentrated in the low fequency compo- nents. An algorithm is proposed where the decision is taken from the study of the low frequency coeDcients of a block. Finally, a Subband/DCT based fast encoding algorithm is developed for H.264/AVC where the wavelet decomposition of the input image is uti- lized for the fast mode decision process. This concept is also used in a method for fast intra prediction in Scalable Video Coding. Simulation results show that the algorithms achieve signiDcant savings in complexity with a negligible loss in rate-distortion performance. Subjective evaluations show that these techniques result in similar perceptual quality when compared to a reference encoder JM12.4 of the H.264/AVC. These algorithms are likely to be useful in implementing real-time H.264/AVC standard encoders in low bitrate applications and for devices with restricted computational capability..

Browse

Browsing Department of Electronics and Electrical Egineering by Issue Date

Results Per Page

Sort Options