This PDF file contains the front matter associated with SPIE Proceedings Volume 7882, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

This paper describes video coding technology proposal submitted by Qualcomm Inc. in response to a joint call for proposal (CfP) issued by ITU-T SG16 Q.6 (VCEG) and ISO/IEC JTC1/SC29/WG11 (MPEG) in January 2010. Proposed video codec follows a hybrid coding approach based on temporal prediction, followed by transform, quantization, and entropy coding of the residual. Some of its key features are extended block sizes (up to 64x64), recursive integer transforms, single pass switched interpolation filters with offsets (single pass SIFO), mode dependent directional transform (MDDT) for intra-coding, luma and chroma high precision filtering, geometry motion partitioning, adaptive motion vector resolution. It also incorporates internal bit-depth increase (IBDI), and modified quadtree based adaptive loop filtering (QALF). Simulation results are presented for a variety of bit rates, resolutions and coding configurations to demonstrate the high compression efficiency achieved by the proposed video codec at moderate level of encoding and decoding complexity. For random access hierarchical B configuration (HierB), the proposed video codec achieves an average BD-rate reduction of 30.88c/o compared to the H.264/AVC alpha anchor. For low delay hierarchical P (HierP) configuration, the proposed video codec achieves an average BD-rate reduction of 32.96c/o and 48.57c/o, compared to the H.264/AVC beta and gamma anchors, respectively.

Distributed Video Coding (DVC) is an emerging video coding paradigm for the systems that require encoders having low complexity that are supported by decoders having high complexity as would be required for, say, real-time video capture and streaming from one mobile phone to display on another. Under the assumption of an error-free transmission channel, the coding efficiency of current DVC systems is still below that of the latest conventional video codecs, such as H.264/AVC. To increase coding efficiency we propose in this paper that either every second Key frame or every Wyner-Ziv frame is downsampled by a factor of two in both dimensions prior to encoding and subsequent transmission. However, this would necessitate upsampling coupled with interpolation at the decoder. Simple interpolation (e.g., bilinear or FIR filter) would not suffice since high-frequency (HF) spatial image content would be missing. Instead, we propose the incorporation of a super-resolution (SR) technique that is based upon using example High Resolution images with content that are specific to the Low Resolution scene that needs its HF content to be recovered. The example-based scene-specific SR technique will add computational complexity to the decoder side of the DVC system, which is allowable within the DVC framework. Rate-distortion curves will show that this novel combination of SR with DVC improves the system performance by up to several decibels as measured by the PSNR, and can actually exceed the performance of an H.264/AVC codec, using GOP=IP, for some video sequences.

A novel cross-layer method is proposed for real-time transmission of standard compliant scalable video over a powerlimited multiple-input multiple-output (MIMO) system with channel state feedback. In the MIMO system, adaptive power allocation and antenna selection are utilized for creation of unequal bit error rate (BER) sub-channels. BER across all the sub-channels can be improved by reducing the channel throughput. In the proposed method, the scalable video is first divided into multiple video sub-streams of unequal importance by content-based partitioning and sorting of video layers. A novel technique is utilized to select the sub-stream data to be sent over the available MIMO sub-channels as to match the importance of the video data to both the channel BER and data transmission delay. Video packets that are delayed excessively are discarded at the transmitter. A trade-off exists between the losses in video peak signal-to-noise ratio (PSNR) resulting from discarded video packets at the transmitter, and gains in video PSNR due to lower channel BER. Simulation results show that the proposed method results in significantly improved performance compared with video transmission over constant BER channels with throughput equal to the video bit rate.

In this paper, we propose a scheme for the optimal allocation of power, source coding rate, and channel coding rate for each of the nodes of a wireless Direct Sequence Code Division Multiple Access (DS-CDMA) visual sensor network. The optimization is quality-driven, i.e. the received quality of the video that is transmitted by the nodes is optimized. The scheme takes into account the fact that the sensor nodes may be imaging scenes with varying levels of motion. Nodes that image low-motion scenes will require a lower source coding rate, so they will be able to allocate a greater portion of the total available bit rate to channel coding. Stronger channel coding will mean that such nodes will be able to transmit at lower power. This will both increase battery life and reduce interference to other nodes. Two optimization criteria are considered. One that minimizes the average video distortion of the nodes and one that minimizes the maximum distortion among the nodes. The transmission powers are allowed to take continuous values, whereas the source and channel coding rates can assume only discrete values. Thus, the resulting optimization problem lies in the field of mixed-integer optimization tasks and is solved using Particle Swarm Optimization. Our experimental results show the importance of considering the characteristics of the video sequences when determining the transmission power, source coding rate and channel coding rate for the nodes of the visual sensor network.

The Transverse Field Detector (TFD) is a filter-less and demosaicking-less color sensitive device that easily allows the design of more than three color acquisition channels at each pixel site. The separation of light into different wavelength bands is based on the generation of transverse electric fields inside the device depleted region, and exploits the properties of the Silicon absorption coefficient. In this work we propose such a device for the joint capture of visible and near infrared (NIR) radiation, for possible applications in videoconferencing and 3D imaging. In these applications the detector is used in combination with suitably generated NIR structured light. The information of the fourth acquisition channel, mainly capturing NIR signals, can be used both for sampling NIR light intensity and for subtracting unwanted NIR crosstalk from visible channels thus avoiding the need for the IR-blocking filter. Together with the presentation of a 4-channel sensor, a suitable algorithm for the processing of signals generated in the visible and infrared bands is described. The goal of the algorithm is to minimize the crosstalk of NIR radiation inside the visible channels and, simultaneously, to maintain good color reproduction and noise performance for the sensor, while holding a good sensitivity of the NIR channel up to 900 nm. The analysis indicates that the algorithm reduces the crosstalk of infrared signals inside R, G and B channels from 31%, 12% and 5% respectively to less than 2%. Concerning noise propagation, the worst coefficient of the color conversion matrix (CCM) is -2.1, comparable to those obtained for CCM of Bayer Color Filter Arrays.

This study aims at the robust automatic detection of buildings with a gable roof in varying rural areas from very-high-resolution aerial images. The originality of our approach resides in a custom-made design extracting key features close to modeling, such as e.g. roof ridges and gutters. In this way, we allow a large freedom in roof appearances. The proposed method is based on a combination of two hypotheses. First, it exploits the physical properties of gable roofs and detects straight line-segments within non-vegetated and non-farmland areas, as possibilities of occurring roof-ridges. Second, for each of these candidate roof-ridges, the likely roof-gutter positions are estimated for both sides of the line segment, resulting in a set of possible roof configurations. These hypotheses are validated based on the analysis of size, shadow, color and edge information, where for each roof-ridge candidate the optimal configuration is selected. Roof configurations with unlikely properties are rejected and afterwards ridges with overlapping configurations are fused. Experiments conducted on a set of 200 images covering various rural regions, with a large variation in both building appearance and surroundings, show that the algorithm is able to detect 75% of the buildings with a precision of 69.4%. We consider this as a reasonably good result, since the computing is fully unconstrained, numerous buildings were occluded by trees and because there is a significant appearance difference between the considered test images.

We study the ability to derive meaningful information from decompressed imaging spectrometer data. Each hyperspectral band is linearly predicted by a previously transmitted band. The residual, formed by subtracting the prediction from the original data is compressed either with a near-lossless bit plane coder or with the lossy JPEG2000 algorithm. We investigate the effects of these two compression methods on hyperspectral image processing using whole- and mixed-pixel analysis techniques. Surprisingly, the lossy coder outperforms near-lossless method in terms of its impact on final hyperspectral data applications.

Superresolution is a sophisticated strategy to enhance image quality of both low and high resolution video, performing tasks like artifact reduction, scaling and sharpness enhancement in one algorithm, all of them reconstructing high frequency components (above Nyquist frequency) in some way. Especially recursive superresolution algorithms can fulfill high quality aspects because they control the video output using a feed-back loop and adapt the result in the next iteration. In addition to excellent output quality, temporal recursive methods are very hardware efficient and therefore even attractive for real-time video processing. A very promising approach is the utilization of Kalman filters as proposed by Farsiu et al. Reliable motion estimation is crucial for the performance of superresolution. Therefore, robust global motion models are mainly used, but this also limits the application of superresolution algorithm. Thus, handling sequences with complex object motion is essential for a wider field of application. Hence, this paper proposes improvements by extending the Kalman filter approach using motion adaptive variance estimation and segmentation techniques. Experiments confirm the potential of our proposal for ideal and real video sequences with complex motion and further compare its performance to state-of-the-art methods like trainable filters.

In 2007, Li and Cox showed that their scheme called Perceptual-QIM (P-QIM) was one of the solutions the most successful in order to watermark multi-bits in an image by a quantization approach. Our research led us to take some of their ideas and brought new proposals. This paper presents a new scheme named Hyper-Cube. In addition to re-express the mechanisms of watermarking from a dierent angle and to give a clear framework, we propose two improvements: the computation of the modied Watson slacks on a neighborhood, and the use of a cleverly integrated error correcting code. Additionally, we experimentally show that the addition of the JPEG quantization table for setting the size of lattices do not reduce performances. This demonstrate that the scheme may easily be integrated in a joint watermarking-compression scheme. Given the obtained results, we can conclude that the Hyper-Cube watermarking scheme is currently one of the most successful technique when one wants to watermark an image using quantization-based approaches.

In this paper, we describe a Trellis Coded Quantization (TCQ)-based quantization and watermarking technique in the framework of JPEG2000 still image compression. Furthermore, we investigate the design of a novel joint compression and watermarking scheme based on a hybrid TCQ module which can perform at the same time quantization and watermark embedding. The watermark extraction process can be achieved both during and after image decompression. Another advantage is the lower complexity of the system because the quantization stage is used for both compression and watermarking purposes. Experimental results have demonstrated that the proposed joint scheme successfully survives JPEG2000 compression with minimal degradation of the image quality. We also studied the robustness of the scheme against gaussian filtering attack, gaussian noise attack, valumetric attack and jpeg attack.

We present a novel method for robust indexing and retrieval of multiple motion trajectories obtained from a multi-camera system. Motion trajectories describe the motion information by recording the objects' coordinates in the video sequence. We generate a four-dimensional tensor representation of multiple motion trajectories from multiple cameras. We subsequently rely on high-order singular value decomposition (HOSVD) for compact representation and dimensionality reduction of the tensor. We show that HOSVD-based representation provides a robust framework that can be used for a unified representation of the HOSVD of all subtensors. We thus demonstrate analytically and experimentally that the proposed HOSVD-based representation can handle flexible query structure consisting of an arbitrary number of objects and cameras. Simulation results are finally used to illustrate the superior performance of the proposed approach to multiple trajectory indexing and retrieval from multi-camera systems compared to the use of a single camera.

Many practical scenarios such as video tracking in lossy environment require a robust accurate tracking algorithm with dropped frames. A novel robust approach is proposed for visual tracking in the first part of this paper in the presence of frame loss with the Bayesian Importance Sampling framework based on first-order hidden Markov model (HMM). The graphical methods are firstly used to provide an exact solution for estimation using first-order hidden Markov model (HMM) with dropped frames. We subsequently rely on Sequential Importance Sampling to derive the first-order particle filtering algorithm with missing frames. In the second part of the paper, we promote this result and present that graphical methods can also be used to provide an exact solution to particle filtering with missing frames for an mth-order hidden Markov model (HMM) and cycle-free graphs. The resulting algorithm requires a small number of particles for efficient tracking. Experimental results demonstrate the superiority and robustness of the proposed approach to the standard methods, yet the additional computational time required is negligible.

This paper proposes an affine image registration technique formulated in a hypothesis-test framework. The technique is based on a mapping algorithm which matches curves and junctions (edge corners) to achieve registration. A similarity metric based on partitioned (short) curves is utilized to narrow the space of possible junction mappings (hypotheses) and the transformation matrix of junction and curve mappings are analyzed (testing) to find the best affine registration of two images. Experimental results show that the proposed algorithms can effectively align images.

In dependent stereo image compression, the aim is to minimize the bitrate of disparity map and that of residual image. Traditionally, focus has been paid on either disparity map or residual image. In this paper, we compute an optimal disparity map (in terms of bitrates) by jointly exploiting the trade-off between the disparity map and the residual image. Firstly, the dense disparity map is obtained using existing optical flow technique. Secondly, the dense disparity map is quantized using a RD framework. Consequently, the resulting bitrate of the disparity map decreases significantly at the cost of a slight increase of the bitrate of the residual image. As a result, the overall bitrate attains minimum value. The proposed scheme is compatible and can be integrated in JPEG2000 framework.

In this paper we present a new method for object tracking initialization using background subtraction. We propose an effective scheme for updating a background model adaptively in dynamic scenes. Unlike the traditional methods that use the same "learning rate" for the entire frame or sequence, our method assigns a learning rate for each pixel according to two parameters. The first parameter depends on the difference between the pixel intensities of the background model and the current frame. The second parameter depends on the duration of the pixel being classified as a background pixel. We also introduce a method to detect sudden illumination changes and segment moving objects during these changes. Experimental results show significant improvements in moving object detection in dynamic scenes such as waving tree leaves and sudden illumination change, and it has a much lower computational cost compared to Gaussian mixture model.

In this paper we propose a background estimation and update algorithm for cluttered video surveillance sequences in indoor scenarios. Taking inspiration from the sophisticated framework of the Beamlets, the implementation we propose here relies on the integration of the Radon transform in the processing chain, applied on a blockby- block basis. During the acquisition of the real-time video, the Radon transform is applied at each frame in order to extract the meaningful information in terms of edges and texture present in the block under analysis, providing with the goal of extracting a signature for each portion of the image plane. The acquired model is updated at each frame, thus achieving a reliable representation of the most relevant details that persist over time for each processed block. The algorithm is validated in typical surveillance contexts and presented in this paper using two video sequences. The first example is an indoor scene with a considerably static background, while the second video belongs to a more complex scenario which is part of the PETS benchmark sequences.

People tracking has to face many issues in video surveillance scenarios. One of the most challenging aspect is to re-identify people across different cameras. Humans, indeed, change appearance according to pose, clothes and illumination conditions and thus defining features that are able to robustly describe people moving in a camera network is a not trivial task. While color is widely exploited in the distinction and recognition of objects, most of the color descriptors proposed so far are not robust in complex applications such as video surveillance scenarios. A new color based feature is introduced in this paper to describe the color appearance of the subjects. For each target a probabilistic color histogram (PCH) is built by using a fuzzy K-Nearest Neighbors (KNN) classifier trained on an ad-hoc dataset and is used to match two corresponding appearances of the same person in different cameras of the network. The experimental results show that the defined descriptor is effective at discriminating and re-identifying people across two different video cameras regardless of the viewpoint change between the two views and outperforms state of the art appearance based techniques.

This paper presents a method to estimate the number of people in crowded scenes without using explicit object segmentation or tracking. The proposed method consists of three steps as follows: (1) extracting space-time interest points using eigenvalues of the local spatio-temporal gradient matrix, (2) generating crowd regions based on space-time interest points, and (3) estimating the crowd density based on the multiple regression. In experimental results, the efficiency and robustness of our proposed method are demonstrated by using PETS 2009 dataset.

The users that take part in a video multicast session often have different bandwidth capabilities, and also experience different loss rates in the network. The process of delivering video to these receivers appropriate to their available bandwidth is called heterogeneous multicast. We present a solution for this problem based on the combination of multiple description coding and network coding. The receivers feed back information on the number of linearly independent descriptions that they obtain, and this is used to optimize the rate allocation of multiple descriptions. Our simulations show significant improvement in performance compared to unirate multicast that uses network coding. Unlike the case for unirate multicast, users with higher bandwidth receive higher quality video with MD/PNC.

Nowadays, the 3D video system using the MVD (multi-view video plus depth) data format is being actively studied. The system has many advantages with respect to virtual view synthesis such as an auto-stereoscopic functionality, but compression of huge input data remains a problem. Therefore, efficient 3D data compression is extremely important in the system, and problems of low temporal consistency and viewpoint correlation should be resolved for efficient depth video coding. In this paper, we propose an object-adaptive depth compensated inter prediction method to resolve the problems where object-adaptive mean-depth difference between a current block, to be coded, and a reference block are compensated during inter prediction. In addition, unique properties of depth video are exploited to reduce side information required for signaling decoder to conduct the same process. To evaluate the coding performance, we have implemented the proposed method into MVC (multiview video coding) reference software, JMVC 8.2. Experimental results have demonstrated that our proposed method is especially efficient for depth videos estimated by DERS (depth estimation reference software) discussed in the MPEG 3DV coding group. The coding gain was up to 11.69% bit-saving, and it was even increased when we evaluated it on synthesized views of virtual viewpoints.

This paper presents an efficient depth map coding method based on color information in multi-view plus depth (MVD) system. As compared to the conventional depth map coding in which depth video is separately coded, the proposed scheme involves color information for depth map coding. In details, the proposed algorithm subsamples input depth data along temporal direction to reduce the bit-rate, and non-encoded depth frames are recovered at the decoder side guided by the motion information extracted from the decoded color video. The simulation results shows the high coding efficiency of the proposed scheme, and it also shows that recovered depth frame are not much different from the reconstructed one. Furthermore, it can even provide temporally consistent depth map which results in better subjective quality for view-interpolation.

In this work, a novel method for color video compression using key-frame based color transfer has been proposed. In this scheme, compression is achieved by discarding the color information of all but few selected frames. These selected frames are either the key frames (frame selected by a key frame selection algorithm) or the Intra coded (I) frames. The partially colored video is compressed using a standard encoder thereby achieving higher compression. In the proposed decoder, a standard decoder first generates the partially colored video sequence from the compressed input. A color transfer algorithm is then used for generating the fully colored video sequence. The complexity of the proposed decoder is close to a standard decoder, allowing its use in wide variety of applications like video broadcasting, video streaming, hand-held devices etc.