MULTIMEDIA LAB

Memorability Predictons using Deep Learning Techniques

In recent times, with huge advancement of artificial intelligence, computational intelligence has been emerging to make more appealing media interfaces like a smart web page, attractive advertisement, cover page of books, etc. To this end, it is required to have some object metrics which essentially describe some subjective media properties. Image or object memorability is such a metric which describes a subjective property of an image. Latest research works show that it is not an incomprehensible concept: variation in remembering images is consistent across viewers. It suggests that independent of a viewers’ contexts and biases, some images are intrinsically more memorable than others. This research work proposes few visual factors which play a major role in determining memorability at object and image levels and few deep learning based memorability prediction models are proposed to predict object and image memorability scores individually.

3D Video Watermarking

Due to enormous advancement of Internet technology and display devices 3D video becomes popular in recent times. To ensure secure media transmission, efficient authentication schemes for such 3D video sequence is a requirement. In recent past, watermarking is being regarded as a popular DRM tool for video authentication. It has been observed that video watermarking becomes a challenging task in presence of advanced auto-stereoscopic display devices and MVD (Multi-view Video plus Depth) based encoding technique in case of 3D video. On basis of this, a research problem formulated to propose a robust watermarking and authentication method for 3D video sequences.

Robust Video Watermarking against Content Adaptation Attack

With the enormous advent of computer, internet and multimedia technologies, multimedia content are easily accessible over the internet. This makes it easy for attackers to copy and tamper these valuable contents.Digital watermarking has been researched and used for copyright and content authentication successfully for many years. A scalable encoder encodes a video to its highest quality and the bitstream is adapted according to the requirement of various communication channels and user-end display devices. This content adaptation is done by scaling the resolution, quality and frame rate of the video. These content adaptations can corrupt content protection data. Content adaptation can be considered as potential attack to the watermark. In this area, different existing video watermarking scheme against content adaptation attacks are implimented and the limitations of these schemes are analyzed. On the basis of these limitations, a research problem has been formulated to propose a robust watermarking scheme against content adaptation attacks.

Compressed Domain Video Watermarking for H.264/AVC

In recent past, with the increasing threats of video piracy, video ownership authentication through watermarking become an important research issue. To protect video communication, digital video watermarking is considered as a promising solution. The performance of the video watermarking often evaluated with parameters such as robustness, visual quality of the watermarked signal, security, transparency, payload, increase in bit rate, blindness, resistance of a given set of attacks, need of fully decompression for embedding etc. However, most of these parameters are inter-dependent, often conflicting and are chosen based on the application. H.264 is considered as the most efficient compression video compression standard, builds on the concepts of MPEG-2 and MPEG-4. In this module of work, different compressed domain watermarking schemes based on P-frames that can provide better visual quality, handle drift compensation, minimize synchronization error, resisting frame drop and alter attack, and restricted increase in video bit rate are proposed. A motion coherent region detection method is described in compressed domain

Watermarking Schemes for High Resolution Video Stream

High Efficiency Video Coding (HEVC) is the newest standard created by JCT-VC team. The main goal of this standard is to compress high resolution video with 50% less bit rate than the previous standard H.264/AVC, maintaining same visual quality. Due to the less bit rate maintaining same visual quality and the growth of popularity of HD and beyond HD video, it is expected that HEVC will replace previous standard codec. The main goals are analyze the applicability of the H.264/AVC watermarking schemes over HEVC video stream by comparing the inherent difference between H.264/AVC and HEVC codec architecture and implimenting new schemes for High Resolution Video watermarking.

Resource Allocation Strategies for Multimedia Services in Wireless Networks

Recent advances in wireless broadband technology have spurred an ever-increasing demand for diverse data rate traffic flows with varied quality of Service (QoS) requirements ranging from real-time voice over IP (VoIP), streaming video / audio, online gaming etc. to soft real-time flows like telnet, web-browsing etc. and non-real-time best-effort data downloads. Such a dynamic system requires intelligent low overhead radio resource allocation mechanisms to efficiently serve a variety of heterogeneous user equipments (mobile phones, laptops, tablets etc.) such that the QoS demands of all flows are met while simultaneously satisfying other practical constraints / objectives like limited power budget, maximizing spectral efficiency in the face of ever changing user mobility and network dynamics, graceful degradation in times of overload etc. This research thus intends to investigate into the theoretical and practical aspects of scheduling mechanisms which multiplexes radio resources among different traffic flows in both time and frequency on the downlink and develop new efficient low overhead algorithms suited to the diverse applications mentioned above, possibly with provable theoretical bounds on their scheduling optimality.