The split-inference paradigm divides an artificial intelligence (AI) model into two parts. This necessitates the transfer of intermediate feature data between the two halves. Here, effective compression of the feature data becomes vital. In this paper, we employ Z-score normalization to efficiently recover the compressed feature data at the decoder side. To examine the efficacy of our method, the proposed method is integrated into the latest Feature Coding for Machines (FCM) codec standard under development by the Moving Picture Experts Group (MPEG). Our method supersedes the existing scaling method used by the current standard under development. It both reduces the overhead bits and improves the end-task accuracy. To further reduce the overhead in certain circumstances, we also propose a simplified method. Experiments show that using our proposed method shows 17.09% reduction in bitrate on average across different tasks and up to 65.69% for object tracking without sacrificing the task accuracy.

As numerous edge devices start implementing intelligent components, the challenges of energy consumption, bandwidth efficiency, and privacy gain significance. One proposed solution relies on the paradigm of split inference, which optimizes the delegation of the computational load between edge and remote devices. We developed and implemented the standard-compliant split inference system with an encoder and decoder capable of real-time streaming and processing. Our system outperforms state-of-the-art video compression implementations by an average of 83% bitrate reduction, while preserving privacy. We demonstrate the system's real-time performance on consumer devices, with interactive visualizations of object detection and segmentation, incorporating real-time metrics. Demo video: https://youtu.be/bmCbUo_ZWWU

Wireless Sensor Networks are vulnerable to a plethora of different fault types and external attacks after their deployment. We focus on sensor networks used in healthcare applications for vital sign collection from remotely monitored patients. These types of personal area networks must be robust and resilient to sensor failures as their capabilities encompass highly critical systems. Our objective is to propose an anomaly detection algorithm for medical wireless sensor networks. Our proposed approach firstly classifies instances of sensed patient attributes as normal and abnormal. Once we detect an abnormal instance, we use regression prediction to discern between a faulty sensor reading and a patient entering into a critical state. Our experimental results on real patient datasets show that our proposed approach is able to quickly detect patient anomalies and sensor faults with high detection accuracy while maintaining a low false alarm ratio.

We present a method that employs human visual system characteristics for perceptually lossless optimization of video coding. Algorithm is robust and applicable for any modern hybrid coder. We decided to evaluate it using the most recent standard (HEVC/H.265) in order to show potential benefits for future applications. Experiments show savings of up to 8% in bitrate without any loss in quality as confirmed by two sets of subjective experiments using DSCQS and ACJ methodology.

Today, world is rapidly turning to high definition multimedia. From engineering and programming point of view, this usually means more computation is needed and more memory space is required to achieve these higher qualities. In this paper we explore the use of parallelization opportunities in graphics processors to accelerate video encoding. We evaluate the NVIDIA CUDA[1] toolkit and evaluate the performance of motion estimation in video encoding. The main goal of this paper is to evaluate the capabilities of NVIDIA/CUDA and develop a process for implementing video/multimedia applications. We have discovered that the difference in performance when CUDA is not used properly can be over 100x. We show how we were able to use CUDA capabilities to reduce the motion estimation time from 7000 milli seconds to 70 milli seconds.

Video identification or copy detection is a challenging problem and is becoming increasingly important with the growing popularity of online video services. The problem addressed in this paper is the identification of a given video clip in a given set of videos. For a given query video, the system returns all the instance of the video in the data set. This identification system uses video signatures based on video tomography. A robust and low complexity video signature is designed. The signatures are generated for video shots and not individual frames. This results in a compact signature of 64 bytes per video segment/ shot. The signatures are matched using simple Euclidean distance metric. Both signature generation and matching have a very low complexity. The system was evaluated using the dataset from video copy detection evaluations organized at the CIVR 2007. The results show that the proposed signatures outperform the results reported in the CIVR competition. The results also show that the signature is robust to common video transformations.

A novel encryption model for digital videos is presented. The model relies on the encryption-compression duality of certain types of permutations acting on video frames. In essence, the proposed encryption process preserves the spatial correlation and, as such, can be applied prior to the compression stage of a spatial-only video encoder. Several algorithmic modes of the proposed model targeted for different application requirements are presented and analyzed in terms of security and performance. Experimental results are generated for a number of standard benchmark sequences showing that the proposed method, in addition to providing confidentiality, preserves or improves the compression ratio. 1

This paper presents a novel background modeling and subtraction approach for video object segmentation. A neural network (NN) architecture is proposed to form an unsupervised Bayesian classifier for this application domain. The constructed classifier efficiently handles the segmentation in natural-scene sequences with complex background motion and changes in illumination. The weights of the proposed NN serve as a model of the background and are temporally updated to reflect the observed statistics of background. The segmentation performance of the proposed NN is qualitatively and quantitatively examined and compared to two extant probabilistic object segmentation algorithms, based on a previously published test pool containing diverse surveillance-related sequences. The proposed algorithm is parallelized on a subpixel level and designed to enable efficient hardware implementation.

This paper explores the challenges opportunities in developing and deploying 3D TV services. The 3D TV services can be seen as a general case of the multi-view video that has been receiving significant attention lately. The keys to a successful 3D TV experience are the availability of content, the ease of use, the quality of experience, and the cost of deployment. Recent technological advances have made possible experimental systems that can be used to evaluate the 3D TV services. We have developed a 3D TV prototype and have currently conducting our first user study to evaluate the quality and experience. These experiences have allowed us to identify challenges and opportunities in developing 3D TV services

In order to embed a watermark into a binary document image, some subset of image pixels needs to be modified. This modification will cause a document image distortion. Careful selection of image pixels can make distortion appear less visible. We propose a new binary document image pixel scoring method, the structural distortion measure, whose objective is to identify image pixels whose modification, as part of a watermark embedding process, will minimize document image visible distortion

The wireless networks are notoriously error prone and all errors cannot be prevented in real-time communications. The problem of error correction becomes even more challenging in mobile multicast/broadcast applications. The mobile devices are being equipped with multiple modems that could work simultaneously; for example, devices with both GSM (WAN) and WLAN networks such as WiFi. These multi-modal devices can use the second network to improve their error resilience. We propose a P2P approach to establish and utilize an error recovery channel on a secondary network for multi-user video applications. The mobile devices within the vicinity can utilize the WLAN network to form a P2P network for error recovery purpose. We developed and evaluated three error recovery models for error recovery over secondary networks. The proposed models balance the response time, bandwidth utilization, fairness, and unnecessary data received

Low bitrates enabled by the H.264/AVC standard come at the cost of significant decoding complexity. The problem is exacerbated when multi-view video coding (MVC), presently being standardized by the MPEG committee, is considered. Increasing the number of reference views will improve the prediction but also increase the encoding and decoding complexity. In this paper we present an alternative to transform coding and arithmetic coding used to code motion compensation residuals in H.264/AVC. Proposed approach is sorting permutation based, resulting in significantly decreased decoding complexity. We evaluate the efficiency of the coding and present initial results obtained for some sequences in published literature

There is a renewed interest in the 3DTV research primarily due to the advances in low cost 3D display technologies. The two views required for 3DTV can be compressed using standard video compression techniques. MPEG-2 is widely used in digital TV applications today and H.264/MPEG-4 AVC is expected to be the leading video technology standard for digital video in the near future. The compression gains and quality of 3DTV will vary depending on the video coding standard used. While inter-view prediction will likely improve the compression efficiency, new approaches such as asymmetric view coding are necessary to greatly reduce bandwidth requirements for 3DTV. This demonstration will show the quality of 3D video experience on autostereoscopic displays using H.264 and MPEG-2. We will also show the benefits of using asymmetric view coding as well as the role of eye dominance on 3DTV experience.

Recent advances in video compression and 3D displays have necessitated a further understanding and development of 3D video coding algorithms. The emergence of low cost autostereoscopic displays is expected to drive the growth of 3DTV services. This paper discusses key issues that affect the quality of 3D video experience on autostereoscopic displays. The characteristics of the human visual system can be exploited to compress individual stereo views at different qualities without affecting the perceptual quality of the 3D video. The H.264/AVC video coding algorithm was used to compress each view. We examine the bounds of asymmetric stereo view compression and its relationship to eye-dominance based on a user study. This paper also presents the design and development of a modular video player with stereoscopic and multi-view capabilities including a discussion of useful tools for accelerating the development and enhancing flexibility. The experimental results indicate that eye-dominance influences 3D perception and as a result will impact the coding efficiency of 3D video.

Multimedia Broadcast Multicast Service (MBMS) is currently being standardized by the 3GPP. The MBMS services will enable new class of applications with efficient use of spectrum. MBMS also poses a new set of requirements on video coding and delivery. In this paper we present a short overview of MBMS and a video codec designed for MBMS. The proposed video codec addresses the two key issues in video delivery over MBMS: 1) error resilience and 2) resource adaptive playback. We show that the proposed codec is simple to implement on low-resource terminals such as mobile phones and outperforms MPEG-4 SP coding under lossy conditions.

A recently proposed Chaotic-Key Based Algorithm (CKBA) has been shown to be unavoidably susceptible to chosen/known-plaintext attacks and ciphertext-only attacks. In this paper we enhance the CKBA algorithm three-fold: 1) we change the 1-D chaotic Logistic map to a piecewise linear chaotic map (PWLCM) to improve the balance property, 2) we increase the key size to 128 bits, and 3) we add two more cryptographic primitives and extend the scheme to operate on multiple rounds so that the chosen/knownplaintext attacks are no longer possible. The new cipher has much stronger security and its performance characteristics remain very good.

The emerging H.264 video coding standard offers a flexible coding algorithm for use in a range of applications from broadcast TV to low bitrate mobile applications. The flexibility comes with complexity. The H.264 video coding algorithm is complex and requires substantial amount of resources for encoding and decoding. The resources required to decode an H.264 video varies depending upon the encoding options used. Understanding the resources required to play the video at the receiver is especially important for mobile devices with limited amount of resources. In this paper, we present an overview of the H.264 video coding standard, complexity analysis of H.264 video, and its implications for resource management and adaptive coding for mobile devices. We examine high level bitstream characteristics and its suitability for estimating the bitstream complexity and receiver resource consumption. We show that high level metrics can be used to estimate receiver resource consumption with reasonable accuracy.

An effective content-based visual image retrieval system is presented. This system consists of two main components: visual content extraction and indexing, and query engine. Each image in the image database is represented by its visual features: color and spatial information. The system uses a color label histogram with only thirteen bins to extract the color information from an image in the image database. A unique unsupervised segmentation algorithm combined with the wavelet technique generates the spatial feature of an image automatically. The resulting feature vectors are relatively low in dimensions compared to those in other systems. The query engine employs a color filter and a spatial filter to dramatically reduce the search range. As a result, queue processing is speeded up. The experimental results demonstrate that our system is capable of retrieving images that belong to the same category.

In this paper, we describe a real-time video compression algorithm, based on Three-Dimensional Discrete Cosine Transform. The algorithm, called XYZ video compression, takes advantage of statistical behavior of video data both in spatial and temporal domains. We develop an algorithm for determining optimal quantizers for the three-dimensional DCT. We also add human visual factors into the optimal quantizers. The obtained results show a significant superiority of the proposed algorithm compared to the MPEG standard in term of the duality of the video, compression ratios, and its real-time implementation.