Preface

The advancement of digital media brings with it opportunities. The internet boom of this millennium allows digital data to travel around the world in real time. Through progress in technology and digital devices, the volume of digital data is increasing exponentially: it is predicted that by 2025, the world will have 163 ZB (zettabytes) of data - almost 10 times more than exists today. With this explosion in the volume of digitized volume, appropriate techniques for data processing and analysis have become a challenging proposition for scientists and researchers. The goal of multidimensional data processing is either to extract information or patterns from data in large databases or to understand the nature of the process that produced the data. As a result, such data processing is an open field for research. Scientists and researchers are investing significant effort in discovering novel and efficient methods for storing and archiving data. Newer, higher-dimensional data structures have been created for this purpose. In addition, a huge amount of bandwidth is required for accurate transmission of such voluminous data. So, efforts are also underway to evolve suitable transmission mechanisms as well as security for data.

Recent applications like hyperspectral remote sensing of images and working with medical images deal with huge amounts of data. Hyperspectral images are particularly useful for object classification due to their rich content information. This large dimensionality also creates issues in practice (the curse of dimensionality), such as the Hough phenomena. Efficient segmentation, recognition, and analysis of multidimensional data, such as hyperspectral images, medical images, social media, and audio signals, remain a challenge. Image segmentation is a fundamental process in many image, video, and computer vision applications. It is a critical step in content analysis and image understanding. In the literature, several image segmentation techniques, such as gray-level thresholding, interactive pixel classification, neural network-based approaches, edge detection, and fuzzy rule-based segmentation, have been reported. In addition to increasing storage capacity, researchers are also applying intelligent data-processing techniques to reduce the computational complexity of algorithms while increasing their efficiency. Computational intelligence techniques like evolutionary algorithms, fuzzy sets, rough sets, classification tools, and deep learning tools are used extensively to successfully achieve these goals.

Digital data produced through data-processing algorithms has fundamental advantages of transportability, proficiency, and accuracy; but on the other hand, the data thus produced brings in several redundancies. To solve this challenging problem with data transmission in network surroundings, research on information security and forensics provides efficient solutions that can shield the privacy, reliability, and accessibility of digital information from malicious intentions.

This volume comprises 11 chapters on the various facets of multimodal data processing, ranging from cryptography to sensors and communication data analysis.

Chapter 1 introduces the concepts of multimodal data processing, with an emphasis on issues and challenges in this domain. The chapter also elucidates the different application areas of multimodal data processing.

Digital information science has emerged to seek a copyright protection solution for digital content disseminated through communication channels. A review of the related literature suggests that most of the domain methods have poor capacity control and are vulnerable to attacks. Chapter 2 presents a casting analogy and performance investigation of the proposed transform domain representative data-hiding system using a digital modulation technique. A watermark is constructed using a Boolean operation on the author signature data with an adaptive classifier that approximates the frequency masking characteristics of the visual system.

In Chapter 3, the authors present a digital image watermarking technique based on biometrics and implement it in hardware using a field-programmable gate array (FPGA). This scheme is focused on the covariance saliency method. For extreme security and individual authentication, biometrics such as the iris are introduced. This technique hides biometric information in a cover image so efficiently that the robustness and imperceptibility of the cover image are less likely to be affected and the image is not distorted (as proven during several attacks). A hardware implementation of this algorithm is also provided for the sake of self-sufficiency.

In Chapter 4, an invisible, spatial domain-based image watermarking scheme is demonstrated. One of the most traditional spatial techniques is simple least significant bits (LSB) replacement, which offers high data transparency for embedded information. However, only a small amount of data can be hidden in the case of single-bit (preferably LSB) replacement; consequently, the payload capacity is much less. Additionally, the data sustainability or robustness of the watermark is decreased, as most attacks affect the LSB plane. Thus, the proposed logic follows an adaptive LSB replacement technique where multiple bits are replaced from each pixel to implant the watermark. This adaptive bit replacement is performed in such a way that both the payload and the signal-to-noise ratio can be increased up to a certain level. Intelligent image clustering is utilized to obtain an optimized result.

Video content summarization is a popular research area. Everyday storage of video data is becoming increasingly important and popular. In the process, it is growing into big data. Summarization is an effective technique to obtain video content from large video data. In addition, indexing and browsing are required for large video data. One of the effective techniques for video summarization is based on keyframes: important video frames. In Chapter 5, the authors propose a keyframe-based video summarization technique using a dense captioning model. Initially, video data is taken as input to the model. The model generates region captioning as output, which is converted into a chunk of sentences after applying the clustering technique. This chunk of sentences is summarized to obtain video summary output.

In the modern era, self-driving cars are the most attention-grabbing development in the autonomous vehicle industry. Until now, Google and Tesla have been the most the encouraging participants in this industry. However, no one has yet achieved fully autonomous driving. Chapter 6 is based on autonomous driving in self-driving cars and focuses on fully autonomous driving in any situation. This driving achievement is possible due to the use of reinforcement learning and modern algorithms created for autonomous driving.

The Internet of Things (IoT) is an important means of connecting smart devices called sensors through a physical or cloud network. It amasses a large amount of data from these devices. However, an interoperability problem occurs when integrating data from different sensors or devices because the sensors' data sets are not compatible with each other. The process of data fusion in the IoT network has to be homogeneous and consistent, so the control of data is an important feature in this field. The IoT provides opportunities for data fusion in computer-based systems to improve operational performance, increase common dimensionality, and reduce ambiguity. Chapter 7 introduces an evolutionary study of multimodal data fusion in the smart environment. It examines data fusion motivations for the IoT with a specific focus on using algorithms (such as probabilistic techniques, artificial intelligence algorithms, and theory of belief methods) and particular IoT environments (centralized, distributed, hybrid, or blockchain systems).

Chapter 8 illustrates new, fast, adaptive, optimized blind channel estimation for a cyclic prefix-aided, space-time block-coded multiple input-multiple output orthogonal frequency division multiplexing (STBC-MIMO-OFDM) system. The bottleneck of earlier blind channel estimation techniques was due to high complexity and low convergence. Also, accurate transmission of multimodal data such as hyperspectral images, medical images in the healthcare sector, massive data in social media, and audiovisual signals is still under research. To overcome this problem, a modified flower pollination algorithm (MFPA) has been implemented to optimize data. The optimized MFPA provides good bit error rate (BER) and symbol error rate (SER) performance compared to the traditional flower pollination algorithm (FPA).

In recent times, the radio spectrum has been revealed as a limited resource due to the advent of various state-of-the-art wireless applications. Spectrum regulators initially adopted a static spectrum allocation (SSA) strategy to serve wireless applications on a non-interfering basis. However, although this strategy ensures the least interference between wireless applications, it is a bottleneck to serving huge numbers of spectrum users. The SSA strategy allocates frequency bands to licensed users. In Chapter 9, spectrum sensing is performed using a filter bank approach, which is a specific type of periodogram obtained from received data. The power spectral density of the received signal is estimated from a finite number of observations. In the present research, the Capon method is applied, which uses one bandpass filter to calculate an estimated spectrum value. This filter is designed to be selective based on the received data. Finally, binary...