Prof. P. Nagabhushan, Director IIIT-Allahabad, India
Histo and Interval type of Data Space Every sample is multi-featured , and the generic understanding that every feature is characterized by an unique numeric value is an ideal expectation.The pragmatic aspects that every feature could be composed of several observations,could be due to temporal assimilation,could be descriptive;and that even a sample itself need not be an individual but could be a 'collective unit' , would make the data representation , complex from algorithmic perspective,though philosophically that is the most natural or in other words the most generic. "Edwin Diday" phrases such generic data analysis as 'Symbolic Data Analysis-SDA'. From Big-Data perspective clustering/classification analysis of such Symbolic samples requires an effective way to compute Distance between such samples. A simpler way of visualizing such Symbolic features is to represent them as histograms and a generic way to compute distances between Histo-Symbolic Objects will be presented. Further, if time permits, a model for dimensionality reduction of interval type of Symbolic features will also be presented.
Brief Bio Prof. P. Nagabhushan is presently working in the capacity of Director at IIIT-Allahabad, India. His key areas of specialization include Pattern Recognition, Digital Image Analysis and Processing, Document Image Analysis and Knowledge Mining.
Prof. Dharmendra Singh, IIT-Roorkee, India
Efficient Application of Image Analysis for adaptive Land Cover Classification Image analysis is defined as a methodical operation or sequence of operations performed on data representative of an observed image with the intent of quantifying the characteristics of the image, identifying variations and structure in the image, or altering the image in a way that facilitates its interpretation. The applications of digital image analysis are continuously enlarging through all areas of science and industry, including medicine, ecology, chemistry, engineering, military science, natural resources research, remote sensing of the earth’s surface from the atmosphere and space, navigation, and security protection. One of the important relevance of image analysis is in the procedure of land cover monitoring through images captured at different time intervals. This talk will be focused on the development of adaptive land cover classification algorithm for low/high resolution satellite images using efficient use of image analysis.
Brief Bio Prof. Dharmendra Singh is a Professor in the Department of Electronics and Communication Engineering at the Indian Institute of Technology, Roorkee, India. His research areas include Wave Propagation, Optical and Microwave Remote Sensing, Polarimetry and Interferometric applications, Image Fusion, Image Processing/Image Analysis, ICT, IoT, Cloud Based Solution, Microwave Imaging, Through Wall Imaging, Ground Penetrating Radar, Terahertz Imaging, Radar Absorbing Materials, Stealth Application and Radar Cross Section Reduction.
Brief Bio Dr. R. Balasubramanian is an Associate Professor in the Department of Computer Science and Engineering at the Indian Institute of Technology, Roorkee, India. His research interests include Fractional Transform Theory, Wavelet Analysis, Biometrics, Content Based Video Retrieval, Video Skimming and Summarization, Medical Imaging, Long-Range Imaging, Hyperspectral Imaging and Volume Graphics.
Dr. R.D. Garg, IIT-Roorkee, India
Brief Bio Dr. R.D. Garg is an Associate Professor in the Department of Civil Engineering at the Indian Institute of Technology, Roorkee, India. His research interests include Geomatics Engineering, Land Surveying, Remote Sensing, GIS, GPS, Digital Image Processing, SAR Interferometry and GPR.
Dr. Sanjeev Kumar, IIT-Roorkee, India
Chaos-based techniques for Image Encryption In this talk, we will have a depth understanding of chaotic maps together with their applications in image and video encryptions. The advantages of chaos-based encryption techniques will be discussed over the standard cryptographic algorithms in visual data encryption. In particular, several recent techniques in this direction will be explained together with some of our recent work. The application of fractional order neural network will be explained in order to have a secure image encryption. Finally, some Matlab demonstration will be given to have a coding idea of these techniques.
Brief Bio Dr. Sanjeev Kumar is an Associate Professor in the Department of Mathematics at the Indian Institute of Technology, Roorkee, India. His research interests include Inverse Problems in Imaging and Control, Computer Vision, Mathematical Imaging and Machine Learning.
Dr. Aveek Brahmachari, Senior Staff Engineer, Stryker Global Technology Center, Gurugram, India
Multi-view 3D Reconstruction This talk would be about reconstruction of rigid 3D scenes captured by multiple cameras or the same camera from multiple locations. The basics of formation of images and the underlying camera model and coordinate systems would be explained. In the problem, both the 3D camera geometry and the correspondences in the images captured are unknown. The camera geometry and the correspondences are recovered simultaneously and these estimates are used to finally generate a refined 3D model.
Wide Baseline Image Matching This part of the talk would focus on the problem of estimation of correspondences and camera geometry when the images have undergone high transformation and/or occlusion among other reasons leading to lower intersection of the images in 3D space.
Brief Bio Dr. Aveek Brahmachari is currently a Senior Staff Engineer in Stryker Global Technology Center, Gurugram, India. Previously, he has worked in Defense Research and Development Organization, Samsung and Mobileum in various engineering and scientific roles. His research interests and works are broadly focused in the field of Computer Vision and Machine Learning.
Dr. Biplab Banerjee, IIT-Roorkee, India
Deep Generative Models Currently, the generative models have found increasing attention in the deep learning community. In this talk, I will introduce the generic generative learning models. Then I will discuss about two popular DGMs: variational autoencoder and GAN. Finally, highlights on current research will be provided.
Brief Bio Dr. Biplab Banerjee is an Assistant Professor in the Department of Computer Science and Engineering at the Indian Institute of Technology, Roorkee, India. His research interests include Computer Vision, Machine Learning, Action Recognition, Zero-Shot Learning, Parts based Object Recognition and Target Recognition.
Dr. Navneet Kumar Gupta, IIT-Roorkee, India
Brief Bio Dr. Navneet Kumar Gupta is designated as a System Programmer in the Institute's Computer Center at Indian Institute of Technology, Roorkee, India. His research interests include High Performance Computing (HPC) through cluster solutions and grid technology, servers and storage devices.
Cross-Language Framework for Low-Resource Handwritten Image Analysis Handwritten image analysis has long been an active research area because of its complexity and challenges due to a variety of handwritten styles. Dataset is a necessary and important resource to develop any recognition/word-spotting system for benchmarking. It has been observed that the availability of training data is not uniformly distributed for all scripts. Hence, analysis of low-resource handwritten scripts is difficult as sufficient training data is not available and it is often expensive for collecting data of such scripts. In this talk, a cross language framework will be presented. The framework exploits a large resource of dataset for training and uses it for analyzing data of low-resource scripts.
Brief Bio Dr. P. P. Roy is an Assistant Professor in the Department of Computer Science and Engineering at the Indian Institute of Technology, Roorkee, India. His research interests encompass the areas of Pattern Recognition, Multilingual Text Recognition, Biometrics, Computer Vision, Image Segmentation, Bio-Signal Analysis, EEG based Pattern Analysis, Machine Learning and Temporal Data Analysis.
Biometric recognition systems and their applications Biometrics is the study of anatomical and behavioral features of living beings, for the purpose of their automatic identification. In this modern digital era, biometric based frameworks have successfully replaced traditional token based access granting mechanisms like passwords and PIN numbers. One of the most notable instances of such schemes includes the Government of India's Aadhar project wherein face, fingerprint, and iris features of approximately 430 million individuals have been successfully extracted and stored. This talk would focus on introducing the basic architecture of biometric recognition systems and their associated application domains. We would also discuss about some of the challenges associated with designing such systems including their security issues.
Brief Bio Dr. Debanjan Sadhya is a Post Doctoral Fellow in the Department of Computer Science and Engineering, Indian Institute of Technology,Roorkee, India. Currently he is working under Dr. R. Balasubramanian on the project Privacy Preserving Data Deduplication in Cloud. His primary research interests relate to the broad areas of Privacy, Biometrics and Information Security. More specifically, he works in various domains related to privacy preservation of Micro-databases and Biometric templates.
Developing Deep Learning-Based Computer Vision Applications Using MATLAB Deep learning is a machine learning technique that teaches computer to do what comes naturally to humans. Humans by far are most complex and accurate systems around us, with unique capability to learn from experience. Recently, deep learning based techniques have evolved to an extent that it is capable to solve vision applications such as classification, detection including other tasks with higher accuracy compared to humans. Deep learning-based models requires huge amount of data to train and deployment is challenging given the resource constraints. This session focuses on how MATLAB can assist from algorithm development to deployment on hardware for deep learning algorithms.
Topics covered during this session include :
Building deep learning-based image classification models from scratch and using transfer learning
Visualization of deep neural network architectures
Accelerating Ground Truth Labeling of Images and Video
Deployment of algorithms on embedded hardware