Yogesh H. Dandawate

An automated approach for classification of plant diseases towards development of futuristic Decision Support System in Indian perspective

The major cause for decrease in the quality and amount of agricultural productivity is plant diseases. Farmers encounter great difficulties in detecting and controlling plant diseases. Thus, it is of great importance to diagnose the plant diseases at early stages so that appropriate and timely action can be taken by the farmers to avoid further losses. The project focuses on the approach based on image processing for detection of diseases of soybean plants. The soybean images are captured using mobile camera having resolution greater than 2 mega pixels. The purpose of the proposed project is to provide inputs for the Decision Support System (DSS), which is developed for providing advice to the farmers as and when require over mobile internet. Our proposed work classifies the images of soybean leaves as healthy and diseased using Support Vector Machine (SVM). The algorithm comprises of four major steps: image acquisition, extracting the leaf from complex background, statistical analysis and classification. The pre-processing step includes conversion from RGB to HSV (Hue Saturation Value) color space. For extracting the region of interest (ROI) from the original image, multi-thresholding is used. The color based and cluster based methods are used for segmentation. The algorithm uses Scale Invariant Feature Transform (SIFT) technique which automatically recognizes the plant species based on the leaf shape. The SVM classifier proves its ability in automatic and accurate classification of images. Finally, it can be concluded from the experimental results that this approach can classify the leaves with an average accuracy of 93.79%. The proposed system will enable the farmers to get advice from the agricultural experts with minimal efforts.

Performance comparision of color image compression based on enhanced Vector quantizer designed using different color spaces

This paper presents compression of color images using vector quantization (VQ). Being vector quantization is lossy compression technique, the quality of the decompressed images is degraded. In order to achieve trade off between quality of compression along with good compression ratio, the vector quantizer must be designed optimally. In this paper, we have designed an enhanced vector quantizer (codebook) by selective training of self-organized feature maps network with different images for quality improvement. The performance is analyzed using various quality measures along with conventionally used PSNR. The vector quantizer is designed for different color spaces such as RGB, HSI, HSV, YCbCr and performance is tested for quality determination. Finally, performance is compared with JPEG in terms of quality and file sizes.

Classification of Indian Classical Instruments Using Spectral and Principal Component Analysis Based Cepstrum Features

In applications such as music information and database retrieval systems, classification of musical instruments plays an important role. The proposed work presents automatic classification of Indian Classical instruments based on spectral and MFCC features using well trained back propogation neural network classifier. Musical instruments such as Harmonium, Santo or and Tabla are considered for an experimentation. The spectral features such as amplitude and spectral range along with Mel Frequency Cepstrum Coefficients are considered as features. Being features are not distinguished, classification is done using non parametric classifiers such as neural networks. Being number of cepstrum coefficients are large important coefficients are selected using Principal Component Analysis. It has been observed that using 42 samples for training and 18 for testing, back propogation neural network provides accuracy of 98%. The present work can be extended for more number of Hindustani and Carnitic classical musical Instruments.

Image and Video Compression: Fundamentals, Techniques, and Applications

Image and video signals require large transmission bandwidth and storage, leading to high costs. The data must be compressed without a loss or with a small loss of quality. Thus, efficient image and video compression algorithms play a significant role in the storage and transmission of data. Image and Video Compression: Fundamentals, Techniques, and Applications explains the major techniques for image and video compression and demonstrates their practical implementation using MATLAB programs. Designed for students, researchers, and practicing engineers, the book presents both basic principles and real practical applications. In an accessible way, the book covers basic schemes for image and video compression, including lossless techniques and wavelet- and vector quantization-based image compression and digital video compression. The MATLAB programs enable readers to gain hands-on experience with the techniques. The authors provide quality metrics used to evaluate the performance of the compression algorithms. They also introduce the modern technique of compressed sensing, which retains the most important part of the signal while it is being sensed.

Fusion based Multimodal Biometric cryptosystem

Security is main concern in Automated Teller Machines today. Multimodal Biometrics are more secure as compared to Unimodal Biometrics as even if one trait fails the other is present. The futuristic application of Multimodal Biometrics could be in Automated Teller Machines (ATM). The proposed work involves capturing of three biometric traits of a person namely face, fingerprint and palm vein by designed hardware later these are preprocessed and fused together for cryptography. Palm vein is chosen as a biometric trait as no two palm veins match unless they are of the same person also palm has a good vascular pattern making it a good identifying factor for an individual as compared to other biometric traits. Further security is achieved in the system by fusion of palm vein with face and fingerprint. Cryptosystem is used in order to get security. The images captured by the designed hardware are preprocessed using Image enhancement techniques and Features are extracted by Curvelet Transform, Gabor Filter and Principal Component analysis. The feature vectors are fused at feature level using Euclidean distance and later matched.

Detection and classification of diseases of Grape plant using opposite colour Local Binary Pattern feature and machine learning for automated Decision Support System

Plant diseases cause major economic and production losses as well as curtailment in both quantity and quality of agricultural production. Now a days, for supervising large field of crops there is been increased demand for plant leaf disease detection system. The critical issue here is to monitor the health of the plants and detection of the respective diseases. Studies show that most of the plant disease can be diagnosed from the properties of the leaf. Thus leaf based disease analysis for plants is an exciting new domain. The technique proposed for identification of plant disease through the leaf texture analysis and pattern recognition. In this work we focus on Grapes plant leaf disease detection system. The system takes a single leaf of a plant as an input and segmentation is performed after background removal. The segmented leaf image is then analyzed through high pass filter to detect the diseased part of the leaf. The segmented leaf texture is retrieved using unique fractal based texture feature. Fractal based features are locally invariant in nature and therefore provides a good texture model. The texture of every independent disease will be different. The extracted texture pattern is then classified using multiclass SVM. The work classifies focus on major diseases commonly observed in Grapes plant which are downy mildew & black rot. The proposed approach avails advice of agricultural experts easily to farmers with the accuracy of 96.6%.

Three dimensional image reconstruction using interpolation of distance and image registration

Three dimensional view of objects is becoming of utmost importance. In day to day technology and entertainment applications, instead of using traditional multi-view technology. We are proposing estimation of distance co-ordinates using distances obtained by ifm O3D200 three dimensional camera at lower resolution of image pixels. The camera uses laser technology with time of flight principle. The same scene is imaged using high resolution 2D camera. The matching of certain feature points in 3D & 2D camera scene and interpolating distance a 3D view can be reconstructed by simple triangulation process. The main features used are edges linked used Hough transform for matching and image registration. It has been observed that bilinear and bicubic interpolation gives better results eventhough the resolution of 3D camera is very less. We further applied technique for computation of volume in water tank and trench, which was very close to the actual measurement.

Palm vein pattern-based biometric recognition system

Palm vein-based biometric authentication system aims to recognise individuals from their unique palm vein structure which is next to impossible to duplicate owing to the fact that palm veins are present in the subsurface of the skin and not apparent under visual light. The aim of the proposed work is to develop a low cost but efficient system for acquiring images of the veins, processing these images and matching using various algorithms. Images have been acquired using a web camera and an infrared LED illumination that highlights the veins. The region of interest ROI is extracted from the images and then processed. Three techniques for matching are proposed. Principal component analysis PCA, 2D-wavelet based feature and template designed exclusively for palm vein ROI is applied over ROI for matching. The accuracy of the each algorithm is deduced to compare the three algorithms. The highest accuracy achieved is 93.54% using template matching technique.

Handwritten Devanagari character recognition using wavelet based feature extraction and classification scheme

This paper gives a new approach for recognition of handwritten Devanagari characters. Twenty handwritten characters from 100 people resulting 2000 characters are used for the experimentation. The handwritten characters written of paper is scanned, preprocessed and on every individual characters wavelet transform is applied so as to get decomposed images of characters. Statistical parameters are computed over the decomposition to form feature vector. The feature vectors serve as input to back propagation neural networks for classification into one of 20 classes and based classes they are recognized. The accuracy obtained is around 70 percent over large number of samples.

Rivers and Coastlines Detection in Multispectral Satellite Images Using Level Set Method and Modified Chan Vese Algorithm

The paper discusses wide variety of ways in which multispectral satellite images are being utilized in coastline and river detection. Flooding is a major problem in which causes distraction to the natural resources. River detection in satellite images is useful in flood monitoring, tracing sedimentation along the river bank and tracking dry outs of the major rivers. Coastline detection is an important for coastline zone monitoring, extraction and analysis of coastline changes which are caused by gradual washing out of sand or by abrupt natural calamity. The proposed work presents an approach for detecting rivers and coastlines over water bodies by the Level Set (LS) Approach and Chan Vese (CV) algorithm. CV approach was initially designed for the medical imaging. In the proposed work CV method is modified with respect to the contour smoothening parameters and time step which further improves the algorithm accuracy for the river and coastline detection. Based on the experimental results we compared LS segmentation method with the modified CV model both subjectively and objectively. For objective analysis measures like Dice coefficient, computation time and Hausdorff Distance are used.