Philipp Schaber

Saliency detection for stereoscopic video

We present a novel system for automatically detecting salient image regions in stereoscopic videos. Our proposed algorithm considers information based on three dimensions: salient colors in individual frames, salient information derived from camera and object motion, and depth saliency. These three components are dynamically combined into one final saliency map based on the reliability of the individual saliency detectors. Such a combination allows using more efficient algorithms even if the quality of one detector degrades. For example, we use a computationally efficient stereo correspondence algorithm that might cause noisy disparity maps for certain scenarios. In this case, however, a more reliable saliency detection algorithm such as the image saliency is preferred. To evaluate the quality of the saliency detection, we created modified versions of stereoscopic videos with the non-salient regions blurred. Having users rate the quality of these videos, the results show that most users do not detect the blurred regions and that the automatic saliency detection is very reliable.

CamMark: a camcorder copy simulation as watermarking benchmark for digital video

In 1998, Petitcolas et al. proposed StirMark [14] as a benchmark for image watermarking schemes. The main idea was to introduce a re-sampling process that mimics the analog process of printing and scanning a watermarked image. For digital video, the corresponding concept is a camcorder copy, where a video displayed on a screen is (digitally) recorded using a video camera. As most commercial video streaming systems (VOD, IPTV) and offline distribution (Blu-ray, HDDs for cinemas) are strongly protected by means of DRM, filming a display is actually a relevant use case and a requirement for robust video watermarking systems to survive. We therefore present a tool to simulate content re-acquisition with a camcorder. Our goal is to support watermark development by enabling automated test cases for such camcorder copy attacks, as well as to provide a benchmark for robust video watermarking. Manually creating camcorder copies is a cumbersome process, and even more problematic, it is hardly reproducible with the same setup. By re-sampling each video frame, we simulate the typical artifacts of a camcorder copy: geometric modifications (aspect ratio changes, cropping, perspective and lens distortion), temporal modifications (unsynchronized frame rates and the resulting frame blending), sub-sampling (rescaling, filtering, Bayer color array filter), and histogram changes (AGC, AWB). We also support simulating camera movement (e.g., a hand-held camera) and background insertion.

Semi-automatic registration of videos for improved watermark detection

Virtually every video watermarking technology can benefit from comparison with the original content. For non-blind schemes it is fundamental; for others it is an improvement to increase the watermarks signal-to-noise ratio by subtracting the content that is often noise to the detector. A direct frame-by-frame comparison of the videos is not possible due to the fact that illegal copies of videos usually differ significantly from their originals caused by different spatial resolution or frame rates, geometric distortions from capturing, or targeted attacks. In this paper, we present a software tool that enables the semi-automatic temporal and spatial synchronization of frames and pixels of two similar videos. This process is called registration. We put our focus on utilizing human capabilities with the smallest possible effort, to allow a high overall performance and precision of the registration. An efficient graphical user interface supports the users and visualizes the results of all steps. In addition, we specifically distinguish digitally reproduced copies from analog (camcorded) copies in which two or more frames are blended into a new frame.

CamMark: Analyzing, Modeling, and Simulating Artifacts in Camcorder Copies

To support the development of any system that includes the generation and evaluation of camcorder copies, as well as to provide a common benchmark for robustness against camcorder copies, we present a tool to simulate digital video re-acquisition using a digital video camera. By resampling each video frame, we simulate the typical artifacts occurring in a camcorder copy: geometric modifications (aspect ratio changes, cropping, perspective and lens distortion), temporal sampling artifacts (due to different frame rates, shutter speeds, rolling shutters, or playback), spatial and color subsampling (rescaling, filtering, Bayer color filter array), and processing steps (automatic gain control, automatic white balance). We also support the simulation of camera movement (e.g., a hand-held camera) and background insertion. Furthermore, we allow for an easy setup and calibration of all the simulated artifacts, using sample/reference pairs of images and videos. Specifically temporal subsampling effects are analyzed in detail to create realistic frame blending artifacts in the simulated copies. We carefully evaluated our entire camcorder simulation system and found that the models we developed describe and match the real artifacts quite well.

Modeling Temporal Effects in Re-captured Video

The re-capturing of video content poses significant challenges to algorithms in the fields of video forensics, watermarking, and near-duplicate detection. Using a camera to record a video from a display introduces a variety of artifacts, such as geometric distortions, luminance transformations, and temporal aliasing. A deep understanding of the causes and effects of such phenomena is required for their simulation, and for making the affected algorithms more robust. In this paper, we provide a detailed model of the temporal effects in re-captured video. Such effects typically result in the re-captured frames being a blend of the original videos source frames, where the specific blend ratios are difficult to predict. Our proposed parametric model captures the temporal artifacts introduced by interactions between the video renderer, display device, and camera. The validity of our model is demonstrated through experiments with real re-captured videos containing specially marked frames.

Robust digital watermarking in videos based on geometric transformations

In the efforts to fight piracy of high-valued media content, forensic digital watermarking as a passive content security scheme is a potential alternative to current, restrictive approaches like DRM. In this paper, we present a novel watermarking scheme for videos based on affine geometric transformations. Frames can be modified in an imperceptible manner by applying a small, global rotation, translation, or zooming, which can be detected later on by comparison with the originals. To compensate geometric distortions that have been introduced while a video travels down legal as well as illegal distribution chains, a spatio-temporal synchronization is performed using our video registration toolkit application. To evaluate our approach, we compare it with several other schemes regarding the robustness against common attacks, including camcorder capture.

Real‐time, ray casting‐based scatter dose estimation for c‐arm x‐ray system

Objectives Dosimetric control of staff exposure during interventional procedures under fluoroscopy is of high relevance. In this paper, a novel ray casting approximation of radiation transport is presented and the potential and limitation vs. a full Monte Carlo transport and dose measurements are discussed. Method The x‐ray source of a Siemens Axiom Artix C‐arm is modeled by a virtual source model using single Gaussian‐shaped source. A Geant4‐based Monte Carlo simulation determines the radiation transport from the source to compute scatter from the patient, the table, the ceiling and the floor. A phase space around these scatterers stores all photon information. Only those photons are traced that hit a surface of phantom that represents medical staff in the treatment room, no indirect scattering is considered; and a complete dose deposition on the surface is calculated. To evaluate the accuracy of the approximation, both experimental measurements using Thermoluminescent dosimeters (TLDs) and a Geant4‐based Monte Carlo simulation of dose depositing for different tube angulations of the C‐arm from cranial‐caudal angle 0° and from LAO (Left Anterior Oblique) 0°‐90° are realized. Since the measurements were performed on both sides of the table, using the symmetry of the setup, RAO (Right Anterior Oblique) measurements were not necessary. Results The Geant4‐Monte Carlo simulation agreed within 3% with the measured data, which is within the accuracy of measurement and simulation. The ray casting approximation has been compared to TLD measurements and the achieved percentage difference was −7% for data from tube angulations 45°‐90° and −29% from tube angulations 0°‐45° on the side of the x‐ray source, whereas on the opposite side of the x‐ray source, the difference was −83.8% and −75%, respectively. Ray casting approximation for only LAO 90° was compared to a Monte Carlo simulation, where the percentage differences were between 0.5–3% on the side of the x‐ray source where the highest dose usually detected was mainly from primary scattering (photons), whereas percentage differences between 2.8–20% are found on the side opposite to the x‐ray source, where the lowest doses were detected. Dose calculation time of our approach was 0.85 seconds. Conclusion The proposed approach yields a fast scatter dose estimation where we could run the Monte Carlo simulation only once for each x‐ray tube angulation to get the Phase Space Files (PSF) for being used later by our ray casting approach to calculate the dose from only photons which will hit an movable elliptical cylinder shaped phantom and getting an output file for the positions of those hits to be used for visualizing the scatter dose propagation on the phantom surface. With dose calculation times of less than one second, we are saving much time compared to using a Monte Carlo simulation instead. With our approach, larger deviations occur only in regions with very low doses, whereas it provides a high precision in high‐dose regions.

A Component System for Spontaneous Virtual Networks

In this paper, we propose a framework and a component system, which can be used to dynamically create modularized network services. A composer can automatically compose such services from a library of available components, taking into account dependencies among components and the possibility of sharing components by different services. In addition to group communication and publish/subscribe functionalities, which are implemented as components already, it is very easy to extend the library with third-party components or own implementations. The framework performs the calculation of a suitable service composition and its deployment among all participating nodes. Each composition can take into account different application requirements or incorporate new components without changing the host application. Since the system is also highly modularized itself, it allows to flexibly choose alternative composition or deployment algorithms, or develop new ones.