S. Veni

An application of image processing techniques for detection of diseases on brinjal leaves using k-means clustering method

This work presents a method for identifying plant leaf disease and an approach for careful detection of diseases. The goal of proposed work is to diagnose the disease of brinjal leaf using image processing and artificial neural techniques. The diseases on the brinjal are critical issue which makes the sharp decrease in the production of brinjal. The study of interest is the leaf rather than whole brinjal plant because about 85-95 % of diseases occurred on the brinjal leaf like, Bacterial Wilt, Cercospora Leaf Spot, Tobacco mosaic virus (TMV). The methodology to detect brinjal leaf disease in this work includes K-means clustering algorithm for segmentation and Neural-network for classification. The proposed detection model based artiifical neural networks are very effective in recognizing leaf diseases.

Hexagonal Pixel Grid Modeling for Edge Detection and Design of Cellular Architecture for Binary Image Skeletonization

Digital images can be represented by rectangular pixel grid model. Yet an alternate model paradigm using a hexagonal pixel grid can be used to discretize and process images which are more suitable for computer vision modeling. The merits of using hexagonal lattice are superior symmetry, definite neighborhood and fewer samples are needed compared to a rectangular lattice. This paper elucidates the sub sampling procedure needed to obtain the hexagonally sampled image from the conventional rectangularly sampled image. Two image processing operations namely edge detection and image skeletonization were done on hexagonal lattice and also rectangular lattice for comparison. The algorithm used for the edge detection of sub sampled images is based on CLAP (cellular logic array processor) algorithm. Image Skeletonization was done using iterative thinning method which is better suited for VLSI Implementation. The paper further deals with the design and implementation of a cellular processor array (CPA) that executes binary image skeletonization on a hexagonal lattice. The implementation shows better results compared to the existing methods

A multimodal emotion recognition system from video

Emotion recognition (ER) systems finds applications in many fields like call centres, humanoid Roberts and robotic pets, telecommunication, psychiatry, behavioral science, educational softwares, etc., In this work, the speech and facial features extracted from the video data is explored to recognize the emotions. Since both these features are compliment to each other, on combining them will result in higher performance. The features used for emotion recognition from video data are geometric and appearance based while prosodic and spectral features are employed for speech signal. Support Vector Machine (SVM) classifier is used to capture the emotion specific information. The basic aim of this work is to explore the capability of speech and facial features to provide the emotion specific information.

Hardware implementation of CNN

The requisite properties of analog CNN components, like the Gilbert multiplier, Operational transconductance amplifier, and the current mirror, were separately estimated. Interconnect for a single cell was analyzed , and extended for a 3 /spl times/3 CNN, that has been implemented. A programmable integration time-constant and a template programmability is found possible. It is also seen that implementation is possible at very low power levels, typically 124 uW. The network considered in this design is a continuous-time rectangular type CNN with r = 1. In this paper the network was implemented using analog VLSI techniques and their performance was verified using cadence spectre IC5. The designed CNN could be used for the applications such as image processing, solution of partial differential equation, modelling of nonlinear phenomenon, physical system simulation, etc.

Detection of pollution content in an urban area using landsat 8 data

Pollution control is a challenging task in current scenario. The very first step to control pollution is to detect the sources of pollution. The urban areas are more polluted than rural due to the high population density. The pollutants considered in this paper are aerosol and asbestos sheets. The source of asbestos are building roofs which are mainly in urban area and that of aerosol is combustion of coal. The conventional image processing techniques failed to detect the pollutant in urban environment which can be performed well using multispectral imaging. Since each object has different temperatures using the TIR (Thermal Infrared) bands of Landsat 8 data, the urban objects are classified using the land surface temperature map. The presence of asbestos sheets is detected by change in intensity of images with respect to Band 7 (Short Wave Infrared) and Band 9 (Cirrus). Aerosol is comprised of components that cause air pollution. In this work, the PM10 value is considered as one of the measures to identify the concentration of particulate matters in specific area.

Improved self-snake based anisotropic diffusion model for edge preserving image denoising using structure tensor

The performance of classifier algorithms used for predictive analytics highly dependent on quality of training data. This requirement demands the need for noise free data or images. The existing partial differential equation based diffusion models can remove noise present in an image but lacking in preserving thin lines, fine details and sharp corners. The classifier algorithms can able to make correct judgement to which class the image belongs to only if all edges are preserved properly during denoising process. To satisfy this requirement the authors proposed a new improved partial differential equation based diffusion algorithm for edge preserving image denoising. The proposed new anisotropic diffusion algorithm is an extension of self-snake diffusion filter which estimates edge and gradient directions as eigenvectors of a structure tensor matrix. The unique feature of this proposed anisotropic diffusion algorithm is diffusion rate at various parts of an image matches with the speed of level set flow. In the proposed algorithm an efficient edge indicator function dependent on the trace of the structure tensor matrix is used. The proposed model performs best in preserving thin lines, sharp corners and fine details since diffusion happens only along edges and diffusion is totally stopped across edges in this model. The additional edge-stopping term which is a vector dot product of derivative of an edge stopping function and derivative of an image computed along gradient and edge orthogonal directions is used in this model as shock filter which enables increased sharpness at all discontinuities. The performance of proposed diffusion algorithm is compared with other classical diffusion filters like conventional perona-malik diffusion, conventional self-snake diffusion methods.

Biorthogonal wavelet filters for compressed sensing ECG reconstruction

Abstract This paper elaborates the design details of a new set of bi orthogonal wavelet filters derived from double sided exponential splines. The designed wavelets are applied in compressed sensing (CS) scenario and results were quite promising. CS is a signal acquisition paradigm, which surpasses the traditional limit of Nyquist sampling. Increasing the reconstruction quality with minimum number of samples in CS is always challenging. We have addressed this challenging task of increasing the reconstruction quality within a minimum number of measurements in CS by developing this new set of biorthogonal wavelet filters. Biorthogonal wavelets have several advantages such as linear phase as compared to orthogonal wavelets. This wavelet which we prefer to call as dew1 (double exponential wavelet 1) is applied in CS based ECG reconstruction scenarios and experimented over 21 data records from MIT arrhythmia data base. A total of 950 experiments were conducted in three CS based methodologies for ECG reconstruction and the results were noted. Over all we were able to get nearly 30% improvement in the reconstruction quality. This paper elaborates the design of these bi orthogonal filters and its application in CS based ECG reconstruction scenario. Other than endorsing the results, we also aim to familiarize this newly designed wavelet so it can be further experimented in different domains.

Emotion Recognition from Videos Using Facial Expressions

In recent days, automatic emotion detection is a field of interest and is used in fields such as e-learning, robotic applications, human–computer interaction (HCI), surveillance, ATM monitoring, mood-based playlists/YouTube videos, psychological studies, medical fields like supporting blind and dumb people, for treating autism in children, entertainment, animation, etc., The proposed work describes detection of human emotions from a real-time video or image with the help of classification technique. The major part of human communication constitutes of facial expression, which is around 55% of the total communicated information. The basic facial expressions that are considered by the psychologists are: happiness, sadness, anger, fear, surprise, disgust, and neutral. The proposed work aims to classify a given video into one of the above emotions using efficient facial features extraction techniques and SVM classifier. The author’s contribution is to increase the efficiency in emotion recognition by implementing the above mentioned superior feature extraction and classification methods.

Hyperspectral crop classification using fusion of spectral, spatial features and vegetation indices: Approach to the big data challenge

The world keeps on generating massive amounts of data daily, which lead to the demands for finding new methods to deal with challenges related to ‘Big Data’. Automatic identification of crops is one of the major applications in the field of the hyperspectral image processing. The major curbs involved in crop classification are: i) Huge dimension of the data and ii) Spectral Similarity amongst crops. This paper proposes a new method of crop classification in which fusion of spectral, spatial and vegetation indices is used as the feature set to overcome the limitation of spectral similarity problem. Here the processing is done in two stages: dimensionality reduction and supervised classification. The dimensionality reduction is done using Principal Component Analysis (PCA) and Minimum Noise Transform (MNF) technique and the selected dimensions are classified using Support Vector Machine Classifier. The results obtained using the proposed technique show that on integrating the vegetation indices along with the spectral and spatial features have raised the accuracy to 98.0749% and helped achieve a kappa coefficient of 0.9769.

Big data challenges in airborne hyperspectral image for urban landuse classification

In recent years, it was a difficult task to classify a huge set of data due to the increasing population in urban places. As of now, satellite hyperspectral image provides information but this is not sufficient to classify data in urban areas. To develop the urban areas, accurate and timely information is necessary for the government. Hence, airborne hyperspectral data provides sufficient information for urban planning and disaster management. This paper, focuses on the following objectives: (i) To improve the classification accuracy in bigdata images (ii) To reduce the mixed pixels in residential buildings that are surrounded by small trees (iii) To bring down similar pixels of roads and parking lots. In this paper, 15 different classes were classified which are important for the growth in urban areas. The SVM classifier has more accuracy and better kappa coefficient compared with Neural Network (NN) and K-Means clustering. The Overall Accuracy (OA) has improved by 23.3.