Advances in Signal Processing and Intelligent Recognition Systems

We describe ‘Tidzam’, an application of deep learning that leverages a dense, multimodal sensor network installed at a large wetland restoration performed at Tidmarsh, a 600-acre former industrial-scale cranberry farm in Southern Massachusetts. Wildlife acoustic monitoring is a crucial metric during post-restoration evaluation of the processes, as well as a challenge in such a noisy outdoor environment. This article presents the entire Tidzam system, which has been designed in order to identify in real-time the ambient sounds of weather conditions as well as sonic events such as insects, small animals and local bird species from microphones deployed on the site. This experiment provides insight on the usage of deep learning technology in a real deployment. The originality of this work concerns the system’s ability to construct its own database from local audio sampling under the supervision of human visitors and bird experts.

Plant Leaf retains many characteristics that can be used to achieve automatic plant classification, Leaf also contains other traits like shape, size, color, texture but these are not enough to distinguish one plant species from another. Extraction of prominent leaf features involves enhancement and feature normalization. The leaf venation may explore as a promising feature due to its correlation similarity between intra-classes. Recently numerous plant classification systems have been proposed. Most of the plant recognition systems rely on single feature but automatic plant leaf classification system that uses single feature often faces several limitations in terms of accuracy. The limitations of system can be overcome by building multimodal plant classification systems that fabricates multiple features together using feature level fusion as well as score level fusion. This paper presents score level fusion of LBP (Local Binary Patterns) and GIST (Global descriptors) features towards building more robust automatic plant classification system. The results shows that, the score level fusion has contributed towards efficient plant classification with the 87.22% genuine accept rate (GAR) for GIST features, LBP features with 78.39% GAR and GIST + LBP scores 89.23% of GAR were observed.

Poultry industry needs a system to automate the process of identifying the hen is infected or not. We have proposed a system that uses IoT and sensors to analyze and identify the infected hen. This reduces the cost of labor and increase the accuracy of the identification process. In this paper we discuss about the overall system, audio and video analysis methods and comparing the results using Matlab. The process of sick identification has been optimized using the Matlab results.

A key component in effective kitchen management is inventory control. Keeping track of the kitchen inventory leads to more informed planning and decision-making. With technology advancing in a fast pace and everything around us becoming automated, people prefer to monitor and perform their day-to-day activities by using the smart devices they carry everywhere rather than manually recording and monitoring things. Maintaining and keeping track of everyday common food inventory is becoming one of the major problems in various households, restaurants and food chains. Replenishing the containers at the right moment and also knowing the expiry of foods is a major concern. Working people and busy restaurants find it difficult to keep track because it requires human intervention at the right time. Through this, it is easy to keep an eye on potential problems related to waste and pilferage. In this project we propose an IOT (Internet of Things) based food inventory tracking system, which ensures real time monitoring of the kitchen inventory. The collected data can be analysed in real time to understand the daily or weekly consumption and also predict usage/consumption patterns. There is also provision to check the real time status, history of consumption through an android application. The system contains a Microcontroller, load cell and wireless Module, MQTT broker, a desktop application and an Android application through which real time inventory tracking is performed. The proposed solution is completely wireless and reliable for both domestic and commercial purposes.

Green revolution suggests that agriculture systems, such as farms turn into dynamic entities boosting animal production in an eco-friendly way. In this direction, we propose exploiting the acoustic modality for farm monitoring. Such information could be used in a stand-alone or complimentary mode to monitor the farm constantly and provide a great level of detail. To this end, we designed a scheme classifying the vocalizations produced by farm animals. We employed a feature set able to capture diverse characteristics of generalized sound events seen from different domain representations (time, frequency, and wavelet). These are modeled using state of the art generative and discriminative classification schemes. We performed extensive experiments on a publicly available dataset, where we report encouraging recognition rates.

Clickbaits are the articles containing catchy headlines which lure the reader to explore full content, but do not have any useful information. Detecting clickbaits solely by the headline without opening the link, can serve as a utility for users over internet. This can prevent their time from useless surfing caused by exploring clickbaits. In this paper Ant Colony Optimization, a Swarm Intelligence (SI) based technique has been used to detect clickbaits. In comparison with algorithms used in the past, this SI based technique provided a better accuracy and a human interpretable set of rules to classify clickbaits. A maximum accuracy of 96.93% with a set of 20 classification rules was obtained using the algorithm.

A decentralized detection method is proposed for revealing a radioactive nuclear source with unknown intensity and at unknown location, using a number of cheap radiation counters, to ensure public safety in smart cities. In the source present case, sensors nodes record an (unknown) emitted Poisson-distributed radiation count with a rate decreasing with the sensor-source distance (which is unknown), buried in a known Poisson background and Gaussian measurement noise. To model energy-constrained operations usually encountered in an Internet of Things (IoT) scenario, local one-bit quantizations are made at each sensor over a period of time. The sensor bits are collected via error-prone binary symmetric channels by the Fusion Center (FC), which has the task of achieving a better global inference. The considered model leads to a one-sided test with parameters of nuisance (i.e., the source position) observable solely in the case of $$\mathcal {H}_{1}$$ hypothesis. Aiming at reducing the higher complexity requirements induced by the generalized likelihood ratio test, Davies’ framework is exploited to design a generalized form of the locally optimum detection test and an optimization of sensor thresholds (resorting to a heuristic principle) is proposed. Simulation results verify the proposed approach.

Understanding students’ emotion, especially during the classroom time, can help to improve the positive classroom emotional climate towards promoting academic achievement. Unfortunately, most of the exisiting studies that try to understand the emotion of students have just utilized a questionnaire method to discover the link between the classroom emotional climate and academic achievement. Such methods do not reflect exactly the emotion of students in the real-time mode. There are also other studies that leverage hi-tech technologies (e.g. video camera, sensors, smartphones) to capture data generated by people themselves (e.g. physiological data, facial expression, body postures, human-smartphone interaction) to recognize emotion. Nonetheless, these methods build either a general model for all users or an individual model for a selected user leading to having a weak adaptive ability. In this paper, we introduce Wemotion, a smart-sensing system built by smartphones and wristbands that can not only detect students’ emotion in real-time mode and also evolve to continuously improve the accuracy during the life cycle. The system is evaluated by real data collected from volunteers and compared to several existing methods. The results show that the system works well and satisfies the purpose of our research.

Obstacle detection and collision avoidance are complicated particularly in drones where accuracy matters a lot to avoid collision between a vehicle and an object. This complication arises due to restricted number of heavy sensors like radar. To overcome the drawbacks of heavy sensor, light weight monocular cameras can be employed. Monocular cameras are capable of analyzing and computing depth by giving the three-dimensional representation. In the proposed method, key point features are extracted from each video frame using Computer vision algorithms like Harris corner detector and Scale Invariant Feature Transform algorithm (SIFT). Then by using Brute Force Matching (BFM), key points of consecutive frames are matched. As drone move towards obstacle, size of obstacle increases i.e. convex hull size around key point increases which shows obstacle is detected.

The monitoring condition of electrical machine is an important parameter for maintenance of industrial process operation levels. In this paper, an investigation based on learning machine classifiers to proper classify machine multiple faults i.e. stator short-circuits, broken rotor bars and bearings in three phase induction motors driven by different inverters models is proposed. Experimental tests were performed in 2 different motors, running at steady state, operating under variable speed and torque variation resulting in 2967 samples. The main concept of proposed approach is to apply the three phase current amplitudes to immediately detect motor operating conditions. The high dimensionality of the input vectors in the algorithms was solved through the discretization of the current data, which allows the reduction the classification complexity providing a optimized waveform in comparison with the original one. The results show that it is possible to classify accurately these faults.

Apart from Illumination, Pose and Expression variations, low dimension is also a primary concern that spiflicate the performance of face recognition system. This work distils to applying v-Hog Tensor discriminant analysis on small sized face image to yield good result in terms of correctness rate. Firstly the face image is mapped on to w-quintuple Colorspace to effectively interpret information existing in the image. Further discriminant features are extracted out of Tensor plane to bore on the confounded image due to reduction of image size. To exhibit the beauty of the feature, v-Hog [22] is adopted. The obtained features are further mapped to a lower dimension space for efficient face recognition. In this work the effect of fSVD [17] with bias is also considered to fortify the recognition system. Finally, for classification five different similarity measures are used to obtain an average correctness rate.

Stroke is one of the principal reasons for adult impairment worldwide. Retinal fundus images are analyzed for the detection of various cardiovascular diseases like Stroke. Stroke is mainly characterized by soft and hard exudates, artery or vein occlusion and alterations in retinal vasculature. In this research work, Histogram of Oriented Gradients (HoG) has been implemented to extract features from the region of interest of retinal fundus images. This innovative method is assessed for the computer aided diagnosis of normal healthy and abnormal images of stroke patients. A comparative analysis has been made between the extracted HoG features and Haralick features. HoG features extracted from the region of interest, when given to a Naïve Bayes classifier provides an accuracy of 93% and a Receiver Operating Characteristic (ROC) curve area of 0.979.

With the widespread adoption and use of RFID tags, a valid scenario is one in which an RFID-tagged object includes several components that each have their own individual RFID tags. Under such a context, each of the components are bound to be included in or excluded from the main object over its lifetime. In order for only the tags that are a part of the main object to be authenticated by the main object, there is a need for a secure protocol that ensures that no other tag has access to the shared secrets among the main object and the component objects. Moreover, there is also a need to address relay attacks by adversaries under such scenarios. Existing authentication protocols address relay attacks through round-trip distance measurements in such inclusion/exclusion scenarios. While this works in principle, distance-bounding approaches are not always reliable. We consider another approach for inclusion/exclusion scenarios and develop a protocol sketch for this context. We also provide related security analysis.

Display system finds vast application in most of the fields. The digital display discussed in this paper uses Raspberry pi, Led dot matrix, shift registers (74HC595) has their main components. The digital display system uses IOT platform (i.e. IBM WATSON) as a communication interface between raspberry pi and the led dot matrix. The user types the message that needs to be displayed into the display (led dot matrix) using web application/android application which can be accessed by the user, only if the authentication process is completed. Then this message becomes input to the raspberry pi (controller) which acts as a gateway in this IOT based application. Generally all micro-controller development boards have 20–30 General Purpose Input Output (GPIO) pins which are insufficient, because it cannot be interfaced with the larger displays. So to solve this problem, the Serial In Parallel Out (SIPO) technique is used. The shift register IC uses this technique and takes one serial input of 8 bits and converts it into 8 different outputs of one bit, which can be connected to the led dot matrix. So the overall system built is user friendly, cost effective and consumes low power as it uses LED display. This digital display is a perfect replacement for posters and trivial noticeboards.

Wireless sensor network swiftly emerges as one of the most prominent technology for industrial and human security application. The paper presents hybrid routing protocol for monitoring of radiation at nuclear plant using Wireless Sensor Network technology. The proposed protocol uses hybrid clustering for deployment of sensor nodes in the given area. There are two types of heterogeneous sensor nodes deployed in different region based on energy level for efficient monitoring of radiation. The super node sense the radiation level in their region continuously and transmit the sensed data to base station using fuzzy based clustering technique along with node scheduling. The normal nodes are used for routing purpose along with super nodes. The proposed protocol yields the optimum energy consumption and improve the performance and network lifetime for monitoring the nuclear plant surrounding.

In this paper, we propose to address the issue of lane prediction on unstructured roads, i.e. roads where lane markings are not available. Lane prediction has received considerable attention in the last decade towards the development of ADAS (Advanced driver assistance systems) system. We consider lane prediction as a vision based problem and propose a learning based framework for lane prediction. We pre-process the data using adaptive thresholding to estimate ROI (Region of Interest) in an image. We develop a variant of Bayesian learning using the evidence based Cascaded Dempster Scafer Combination Rule to categorize the road and non-road sectors of the region of interest. We also propose to post-process the data with improved morphological operations to remove the outliers. Lane prediction finds applications in pothole detection, autonomous driving etc. We demonstrate the results on real datasets captured in different scenarios. We compare the results with different state-of-art techniques of lane prediction to validate the efficiency of proposed algorithm.

Virtual assistant is boon for everyone in this new era of 21st century. It has paved way for a new technology where we can ask questions to machine and can interact with IVAs as people do with humans. This new technology attracted almost whole world in many ways like smart phones, laptops, computers etc. Some of the significant VPs are like Siri, Google Assistant, Cortana, and Alexa. Voice recognition, contextual understanding and human interaction are the issues which are not solved yet in this IVAs. So, to solve those issues 100 users participated a survey for this research and shared their experiences. All users’ task was to ask questions from the survey to all personal assistants and from their experiences this research paper came up with the actual results. According to that results many services were covered by these assistants but still there are some improvements required in voice recognition, contextual understanding and hand free interaction. After addressing these improvements in IVAs will definitely increased its use is the main goal for this research paper.

Hyperspectral remote sensing image analysis has always been a challenging task and hence there are several techniques employed for exploring the images. Recent approaches include visualizing hyperspectral images as third order tensors and processing using various tensor decomposition methods. This paper focuses on behavioural analysis of hyperspectral images processed with various decompositions. The experiments includes processing raw hyperspectral image and pre-processed hyperspectral image with tensor decomposition methods such as, Multilinear Singular Value Decomposition and Low Multilinear Rank Approximation technique. The results are projected based on relative reconstruction error, classification and pixel reflectance spectrums. The analysis provides correlated experimental results, which emphasizes the need of pre-processing for hyperspectral images and the trend followed by the tensor decomposition methods.

A 2 × 6 planar coaxial cavity horn antenna array has been proposed for the transmitter module of the MELISSA GB-SAR system [7]. This system is installed in Italy for monitoring of land deformations leading to landslides and is required to work round-the-clock for continuous monitoring. Failure of even a single antenna element in the transmitting or receiving module of this system could alter the radiation pattern of the system and could prove to be hazardous. This article performs a comparative analysis of the Genetic algorithm and Particle Swarm Optimization algorithm to recover the failed element in the 2 × 6 antenna array. The results of MatLab simulation prove that both the GA and PSO algorithms converge well to auto-recover the failed element.

State of Charge (SoC) is the important criterion which reflects the actual battery usage. So, the State of Charge (SoC) has to be precisely estimated for improving the life and the rate of utilization of the battery. During normal operation of the battery, parameters like charge and discharge efficiency, temperature, etc., tend to affect the accurate estimation of SoC. In this paper, for estimating battery SoC with higher accuracy, Strong Tracking Cubature Kalman Filtering (STCKF) is used and the battery model parameters are identified using the method of Recursive Least Square (RLS). Simulation results indicate, STCKF estimates the SoC values as that of Ampere-Hour (AH) method with very minimal error and the dynamically modeled battery parameter values follows the same discharge characteristics as that of real batteries.

Artificial intelligence methods are widely used in different interdisciplinary areas. The paper is devoted to application the method of machine learning and data mining to construction a neuromorphic fault dictionary (NFD) for testing and fault diagnostics in analog/mixed-signal integrated circuits. The main issues of constructing a NFD from the big data point of view are considered. The method of reducing a set of essential characteristics based on the principal component analysis and approach to a cut down the training set using entropy estimation are proposed. The metrics used for estimating the classification quality are specified based on the confusion matrix. The case study results for analog filters are demonstrated and discussed. Experimental results for both cases demonstrate the essential reduction of initial training set and saving of time on the NFD training with high fault coverage up to 100%. The proposed method and approach can be used according to the design-for-testability flow for analog/mixed-signal integrated circuits.

Hyperspectral images are remotely sensed high dimension images, which capture a scene at different spectral wavelengths. There is a high correlation between the bands of these images. For an efficient classification and processing, the high data volume of the images need to be reduced. This paper analyzes the effect of dimensionality reduction on hyperspectral image classification using vectorized convolution neural network (VCNN), Grand Unified Recursive Least Squares (GURLS) and Support Vector Machines (SVM). Inorder to analyze the effect of dimensionality reduction, the network is trained with dimensionally reduced hyperspectral data for VCNN, GURLS and SVM. The experimental results shows that, one-sixth of the total number of available bands are the maximum possible reduction in feature dimension for Salinas-A and one-third of the total available bands are for Indianpines dataset that results in comparable classification accuracy.

Seizure is defined as a sudden synchronous activity of a group of neurons resulting in an electric surge in the brain. Epilepsy is a brain disorder indicated by repeated seizures. Around 10 million people in India are suffering from epilepsy. Electroencephalogram (EEG) signal being low cost and non-invasive in nature can be used effectively for seizure detection. The present work focuses on developing an efficient epileptic seizure detection system using intracranial EEG signals. Dual tree complex wavelet transform is used to decompose the signal into various sub-frequency bands. Probability features are used to extract efficient indicators for seizure and healthy classes. Discriminant correlation analysis is used to increase the difference between different classes as well as reduce the difference between same classes. The fused feature set is clustered using fuzzy c means clustering algorithm. Hidden Markov model discriminates the seizure class with healthy class with good efficiency. Maximum accuracy of 98.57% is achieved for seizure detection with very low execution time.

The paper presents bit-plane specific new measures to visualize the extensive statistical detail of an image. We compute the frequency of ones, maximum run length and correlation among rows (columns) in each bit-plane of an image. The computed measures give row-wise and column-wise structural detail at bit-plane level and help an interpreter to analyze given image deeply for its effective interpretation and understanding. In this paper, the application of these measures is shown in cryptography to statistically analyze the image ciphers. The simulation study shows that the proposed measures are very useful and can be applied in various image processing applications for pattern recognition and understanding of visual objects.

Cervical cancer is one of the most common cancer among women world-wide which can be cured if detected early. The current gold standard for cervical cancer diagnosis is clumsy and time consuming because it relies heavily on the subjective knowledge of the medical professionals which often results in false negatives and false positives. To reduce time and operational complexities associated with early diagnosis, we require a portable interactive diagnostic tool for early detection, particularly in developing countries where cervical cancer incidence and related mortality is high. Incorporation of digital colposcopy in place of manual diagnosis for cervical cancer screening can increase the precision and strongly reduce the chances of error in a time-specific manner. Thus, we propose a robust and interactive colposcopic image analysis and diagnostic tool, which can categorically process colposcopic images into Type I, Type II and Type III cervigrams and identify lesions in least amount of time. Furthermore, successful binning of diagnosed cervigrams into digital colposcopic library and incorporation of a set of specific parameters that are typically referred to for identification of transformation zone and SCJ (squamo columnar junction) with the help of open source Programming language - “R” is one of the major highlights of the application. The software has the ability to automatically identify and quantify the morphological features, color intensity, sensitivity and other parameters digitally which may improve and accelerate screening and early diagnosis, ultimately leading to timely treatment of cervical cancer.

There are various inbound web services which prescribe services to clients. Specialists are more engaged in making framework for proposal of web service (WS) which limit the intricacy of selection process and improve the quality of service (QOS) suggestion. Our work implements a framework which recommends web services using an analytical model based on the contextual information provided by the service providers. This system helps users obtain high quality service automatically. Adaptive work performs feature reduction, similarity and ranking of WS. The important feature reduction process helps identify attribute values with maximum accuracy which results in proper evaluation of data. Efficient selection of WS for service composition requires better methods which properly calculate the similar values. A similarity helps to identify the closest services as per the requirement in the process of service composition. Ultimately, the system automatically selects the set of web services with highest similarity scores from the optimized set of web service description.

Background: To bring it as a human interactive perceive color process, an automatic color model based segmentation of White Matter (WM), Gray Matter (GM), and Cerebrospinal Fluid (CSF) in Magnetic Resonance Brain images is proposed in this paper.Methods: Preprocessing process is done for the MRI brain images using wavelet based bivariate shrinkage method and Contour based Brain Segmentation method (CBSM). Then segmentation of brain tissues using Hue Saturation Value (HSV) color model Based Histogram Thresholding Technique (HSVBHTT) was applied. Normal and Alzheimer’s disease (AD) brain images obtained from Internet Brain Segmentation Repository (IBSR) and Minimal Interval Resonance Imaging in Alzheimer’s Disease (MIRIAD) datasets.Results and Conclusions: The results of proposed method was analyzed with similarity measures and quantitative measures like Jaccard (J), Dice (D), Sensitivity (S) and Specificity (SP) and compared with the manual segmented images which produced better results on segmenting WM, GM and CSF compared to other existing methods.

Estimating the accurate pitch marks for prosody modification is an essential step in the epoch based time and pitch scale (prosody) modification of a given speech. In epoch based prosody modification, the perceptual quality of the time and pitch scale modified speech depends on the accuracy with which glottal closure instants (epochs) are estimated. The objective of the present work is to improve the perceptual quality of the prosody modified speech by accurately estimating the epochs location. In the present work the effectiveness of variational mode decomposition (VMD) in spectral smoothing and wavelet synchrosqueezing transform (WSST) in time-frequency sharpening of a given signal is exploited for refining the zero frequency filtering (ZFF) method which is one of the simple and popular epoch extraction method. The proposed refinements to the ZFF method found to provide improved epoch estimation performance on emotive speech utterances where the conventional ZFF method show severe degradation due to rapid pitch variations. Improved mean opinion scores are obtained based on the subjective evaluation tests performed on the prosody modified speech with the epochs estimated using the refined ZFF method. The reason for improved perceptual quality in the prosody modified speech is the better identification accuracy of the estimated epochs using the proposed method as compared to the conventional ZFF method in the case of emotive speech signals.

A compact wideband high gain Ψ shaped patch antenna on circular ground plane is proposed. The patch is designed and fabricated on FR4 substrate with a thickness of 1.6 mm and is compact in size as 122.7 mm2 including ground plane. The SMA connector with a center radius of 0.36 mm is connected at a coordinate (x = 5 mm, y = 3.3125 mm) as a feeding line to give RF signal as input. A wide impedance bandwidth is obtained due to the slots on the radiating element and by placing shorting post near zero potential fields makes the structure compact. An impedance bandwidth is further enhanced by placing an inverted U-slot on ground plane. The proposed structure is resonating at five different frequencies 1.924 GHz, 2.88 GHz, 5.29 GHz, 8.58 GHz and 12.27 GHz yields an impedance bandwidth of 345 MHz, 92 MHz, 200 MHz and 4.77 GHz respectively. Reflection coefficient (S11) at 8.58 GHz & 12.27 GHz is −36.37 dB and −44.61 dB respectively. The proposed antenna is giving a maximum gain of 6.1 dB and has a stable radiation pattern with in the resonating band. The designed antenna is fabricated and is experimentally validated for the results. It reveals that the proposed antenna is suitable for WLAN and X band applications.

To optimize the Discrete Cosine Transform (DCT) in terms of size and to improve the quality of the image in image compression, in this paper, A 2D-DCT using Loeffler algorithm along with Canonical Signed Digit (CSD) and Canonical Sub expression Elimination (CSE) has been proposed. For fast computation 2-DCT/IDCT is executed by using 1D DCT row column method. The performance of the proposed architecture has been evaluated and compared with the other technique.

The human exposure assessment is dependent on the measurement parameter such as Specific Absorption Rate (SAR). SAR investigation highlights the biological impact of human health due to a radiation source. This paper proposes a novel investigatory method for SAR assessment on human tissues such as kidney, heart, lungs, trachea, and nerve. The radiation source is considered is Inverted F Antenna designed to operate at GSM frequency 900 MHz. The tissue structures are designed with equivalent properties. The experimental analysis of exposure on tissues is performed, and SAR is computed. The various response of tissues towards SAR is compared and analyzed with gain and operation characteristics.

In this paper, an investigation has been made to detect the self-similarity and stationarity nature of magnitude of occurred Earthquake by exploring the fractal pattern and the variation nature of frequency of the essential parameter, Magnitude of occurred earthquake across the different place of the world. The time series of magnitude (19.04.2005 to 07.11.2017), of occurred earthquakes, collected from U.S.G.S. have been analyzed for exposing the nature of scaling (fractality) and stationary behavior using different statistical methodologies. Three conventional methods namely Visibility Graph Analysis (VGA), Wavelet Variance Analysis (WVA) and Higuchi’s Fractal Dimension (HFD) are being used to compute the value of Hurst parameter. It has been perceived that the specified dataset reveals the anti-persistency and Short-Range Dependency (SRD) behavior. Binary based KPSS test and Time Frequency Representation based Smoothed Pseudo Wigner-Ville Distribution (SPWVD) test have been incorporated to explore the nature of stationarity/non-stationarity of that specified profile where the magnitude of earthquake displays the indication of non-stationarity character.

In multiple antenna systems like Multiple Input Single Output (MISO) or Multiple Input Multiple Output (MIMO), the transmitted signal gets distorted due to various factors such as channel fading, multipath propagation, mobility and transmission bandwidth. Using the concepts of differential signalling and permutation modulation which were developed for wireline channels, a new signalling method is proposed in this paper for MIMO channel configurations. The BER performance of proposed system is systematically evaluated under different channel conditions, which includes Additive White Gaussian Channel (AWGN) and multipath Rayleigh fading conditions under MISO and MIMO configurations. The proposed system gives a performance better than conventional Alamouti space time coding scheme, with the same order of computational complexity. It also performs within the theoretical upper bound plotted under AWGN conditions.

In this paper, we present an extensive study and quantitative evaluation of six segmentation techniques on images of Berkeley Segmentation Database. Image segmentation plays a vital role in many computer vision applications and benchmarking such algorithms may assist research community in present and future research efforts in the field of image segmentation. Color space models, Hybrid color space and wavelet, Gradient Magnitude Techniques, K – means, C-Means & Fuzzy C-Means (FCM) and Edison’s Mean – shift approaches are evaluated using at least six metrics with respect to ground-truth boundaries of entire images in BSD 300/500 dataset images. The results stated here gives useful insights to above mentioned approaches and its significance in aligning upcoming research avenues in image segmentation.

Diabetic Maculopathy (DM) is one of the major problems of diabetes mellitus and it is one of the key reasons for the vision problem. It arises due to the leakage of blood from injured retinal veins. The development of DM is moderate and soundless and it is found in 10% of the world diabetic population. If diabetic maculopathy is not noticed in the underlying stage the effect this on macula is irreversible and can prompt vision loss. Therefore, screening of diabetic eye helps in finding diabetic maculopathy at the beginning stage which prevents the vision loss. In this review paper, the anatomy of the human eye and a brief overview of diabetes, diabetic retinopathy and diabetic maculopathy is presented. The literature review of various methods/techniques used for detection of DM in retinal fundus images and the performance metrics used to measure these methods are discussed in details. Issues involved in DM detection are also mentioned in this paper.

Image enhancement is the process of improving the quality of a digitally stored image so that it is more suitable for certain applications. Here a multiimage enhancement method is proposed which uses non linear wavelets. Though wavelet transform is a linear operation, nonlinear extensions can be made by combining it with mathematical morphology. Max-plus algebra based morphological wavelet transform is a type of non-linear wavelet. Unlike the linear wavelet, this wavelet transform uses maximum and minimum operations instead of linear analysis filters. It is used to maintain the edge information of reconstructed image. The original image is decomposed into scaled, horizontal, vertical and diagonal components using max-plus algebra based morphological wavelet transform. These components undergo bilinear interpolation and coefficients are combined by averaging. Then it undergoes inverse transformation to produce a high resolution image. This method results in less mathematical calculations than other existing methods. Performance comparison parameters are calculated for the output image and it is compared with the parameters obtained from other techniques.

In this paper, a comparison-free sorting algorithm is proposed for negative and positive elements which satisfies the conditions such as hardware complexity. The basic idea is to sort the array of input integer elements without performing any comparison related operations between the data. Sorting technique for negative and positive numbers are executed involving similar hardware. Therefore, it doesn’t require any complex hardware design. This avoids any usage of memory such as SRAM or any circuitry involving complex design as compared to that of the ones used in other conventional sorting practices. Instead the proposed work utilizes basic registers to store the binary elements. FSM module is proposed with a comparison-free sorting algorithm in order to reduce hardware complexity. All the designs are coded in VHDL and verified using ModelSim SE10.4d simulator. The conventional and the proposed design are synthesized using Vivado and Synopsys DC Design Compiler (90 nm CMOS technology). From the synthesis report, it is observed that proposed FSM with comparison-free sorting algorithm has reduction in power and area compared to the conventional design.

Digital Modeling of archaeological sites and the inherent cultural heritage has proved to be a tool of great importance not only for discovery of data but also for recovery and understanding of data from the archaeological site remains. The irreversible and destructive nature of excavation can be mitigated by using the techniques of 3D digital modeling using the available hardware and methodologies of computer vision and reconstruction. The widespread adoption of 3D technologies have facilitated not only timely and accurate recording methods but also storage and reusability of data that can be further used for collaborative and reconstructive tasks. This allows for preservation of the excavated objects, artefacts and landscapes and also for inter-disciplinary research. In this work we have put forth the numerous techniques that are used for 3D reconstruction of objects and artefacts, creation of virtual environments of heritage sites and underwater excavation incorporating not only unearth artefacts but also maps and location information to form a virtual and immersive environment.

Efficiency of a recognition algorithm depends upon availability of noiseless and classes of images with unique and differentiable features. Thinning plays an important part in building a recognition system for various types of images. It is mainly used in handwritten character recognition systems and Biometric systems. In Handwritten character recognition systems where there exist large variations within same class of images, thinning plays a vital role. Here curvelet based thinning algorithm is proposed which produces better results compared to other thinning algorithms. The capability of curvelet transform in estimation of directional features added to the strength of proposed work. Curvelet transform is combined with watershed algorithm, binarization and skeletonization to produce the final thinned output. The efficiency of the proposed work has been estimated in this work through detailed experimental analysis and comparisons with 7 efficient thinning algorithms. The measures used for comparison in this works are number of foreground pixels, PSNR (peak signal to noise ratio), MSE (mean square error), RMSE (root mean square error), reduction rate and thinness ratio.