Multispectral Sentinel-2 images are a valuable source of Earth observation data, however spatial resolution of their spectral bands limited to 10 m, 20 m, and 60 m ground sampling distance remains insufficient in many cases. This problem can be addressed with super-resolution, aimed at reconstructing a high-resolution image from a low-resolution observation. For Sentinel-2, spectral information fusion allows for enhancing the 20 m and 60 m bands to the 10 m resolution. Also, there were attempts to combine multitemporal stacks of individual Sentinel-2 bands, however these two approaches have not been combined so far. In this paper, we introduce DeepSent—a new deep network for super-resolving multitemporal series of multispectral Sentinel-2 images. It is underpinned with information fusion performed simultaneously in the spectral and temporal dimensions to generate an enlarged multispectral image. In our extensive experimental study, we demonstrate that our solution outperforms other state-of-the-art techniques that realize either multitemporal or multispectral data fusion. Furthermore, we show that the advantage of DeepSent results from how these two fusion types are combined in a single architecture, which is superior to performing such fusion in a sequential manner. Importantly, we have applied our method to super-resolve real-world Sentinel-2 images, enhancing the spatial resolution of all the spectral bands to 3.3 m nominal ground sampling distance, and we compare the outcome with very high-resolution WorldView-2 images. We will publish our implementation upon paper acceptance, and we expect it will increase the possibilities of exploiting super-resolved Sentinel-2 images in real-life applications.

Sea surface temperature (SST) can be measured from space using infrared sensors on Earth-observing satellites. However, the tradeoff between spatial resolution and swath size (and hence revisit time) means that SST products derived from remote sensing measurements commonly only have a moderate resolution (>1 km). In this article, we adapt the design of a super-resolution neural network architecture [specifically very deep super-resolution (VDSR)] to enhance the resolution of both top-of-atmosphere thermal images of sea regions and bottom-of-atmosphere SST images by a factor of 5. When tested on an unseen dataset, the trained neural network yields thermal images that have an RMSE$2-3\times $smaller than interpolation, with a 6–9 dB improvement in PSNR. A major contribution of the proposed neural network architecture is that it fuses optical and thermal images to propagate the high-resolution information present in the optical image to the restored thermal image. To illustrate the potential benefits of using super-resolution (SR) in the context of oceanography, we present super-resolved SST images of a gyre and an ocean front, revealing details and features otherwise poorly resolved by moderate resolution satellite images.

Controlling the distribution of sensitive content such as pornography has become paramount with the ever-growing accessibility to the internet. Manual filtering of such large volumes of data is practically impossible, thus, the automatic detection of said material is sought after by Law Enforcement Agencies (LEAs) and has been tackled in various manners. However, the sorting of flagged pornographic documents is still done manually using scales that describe hierarchical degrees of content severity. In this paper, we address pornography detection by creating a model capable of locating and labelling sexual organs in images and extend this model to perform image classification to provide the user with one of 19 semantically meaningful descriptors of the content. Generating these descriptors serves as a proof of concept before approaching LEAs to work with illegal CSA material and scales such as COPINE. After creating our own custom sexual organ object detection dataset for the task at hand, we achieved an object detection mean average precision score of 63.63% and a top-3 classification accuracy of 87.78%.

Convolutional Neural Networks (CNNs) have achieved impressive results across many super-resolution (SR) and image restoration tasks. While many such networks can upscale low-resolution (LR) images using just the raw pixel-level information, the ill-posed nature of SR can make it difficult to accurately super-resolve an image which has undergone multiple different degradations. Additional information (metadata) describing the degradation process (such as the blur kernel applied, compression level, etc.) can guide networks to super-resolve LR images with higher fidelity to the original source. Previous attempts at informing SR networks with degradation parameters have indeed been able to improve performance in a number of scenarios. However, due to the fully-convolutional nature of many SR networks, most of these metadata fusion methods either require a complete architectural change, or necessitate the addition of significant extra complexity. Thus, these approaches are difficult to introduce into arbitrary SR networks without considerable design alterations. In this paper, we introduce meta-attention, a simple mechanism which allows any SR CNN to exploit the information available in relevant degradation parameters. The mechanism functions by translating the metadata into a channel attention vector, which in turn selectively modulates the network's feature maps. Incorporating meta-attention into SR networks is straightforward, as it requires no specific type of architecture to function correctly. Extensive testing has shown that meta-attention can consistently improve the pixel-level accuracy of state-of-the-art (SOTA) networks when provided with relevant degradation metadata. Despite average memory/runtime overheads of less than ~2.6%/0.025 seconds for the datasets and models considered, meta-attention improves the performance for both PSNR and SSIM; for PSNR, the gain on blurred/downsampled (X4) images is of 0.2969 dB (on average) and 0.3320 dB for SOTA general and face SR models, respectively. The coding framework used for this paper is available at: https://github.com/um-dsrg/Super-Resolution-Meta-Attention-Networks.

Light field imaging has recently known a regain of interest due to the availability of practical light field capturing systems that offer a wide range of applications in the field of computer vision. However, capturing high-resolution light fields remains technologically challenging since the increase in angular resolution is often accompanied by a significant reduction in spatial resolution. This paper describes a learning-based spatial light field super-resolution method that allows the restoration of the entire light field with consistency across all angular views. The algorithm first uses optical flow to align the light field and then reduces its angular dimension using low-rank approximation. We then consider the linearly independent columns of the resulting low-rank model as an embedding, which is restored using a deep convolutional neural network (DCNN). The super-resolved embedding is then used to reconstruct the remaining views. The original disparities are restored using inverse warping where missing pixels are approximated using a novel light field inpainting algorithm. Experimental results show that the proposed method outperforms existing light field super-resolution algorithms, achieving PSNR gains of 0.23 dB over the second best performing method. The performance is shown to be further improved using iterative back-projection as a post-processing step.

This paper describes a lossless light field compression algorithm that exploits the redundancies present in all four dimensions of the light field using a 4D wavelet transform. We first group spatially collocated pixels from each view to form a hyper-pixel view. The redundancy present in each hyper-pixel is decorrelated using a 2D Haar wavelet transform. This process removes the angular correlations present in the light field. Nevertheless, each angular view is still spatially correlated. The 2D Cohen-Daubachies-Feaveau 5/3 wavelet transform is then employed to suppress the spatial redundacy of the light field. The resulting coefficients of the decorrelated light fields are then encoded using an arithmetic coding scheme. Simulation results demonstrate that the proposed lossless light field compression algorithm outperforms the lossless versions of JPEG, JPEG2000 and HEVC, and achieves a 47% bitrate reduction compared to the state of the art approach.

Sketches obtained from eyewitness descriptions of criminals have proven to be useful in apprehending criminals, particularly when there is a lack of evidence. Automated methods to identify subjects depicted in sketches have been proposed in the literature, but their performance is still unsatisfactory when using software-generated sketches and when tested using extensive galleries with a large amount of subjects. Despite the success of deep learning in several applications including face recognition, little work has been done in applying it for face photograph-sketch recognition. This is mainly a consequence of the need to ensure robust training of deep networks by using a large number of images, yet limited quantities are publicly available. Moreover, most algorithms have not been designed to operate on software-generated face composite sketches which are used by numerous law enforcement agencies worldwide. This paper aims to tackle these issues with the following contributions: 1) a very deep convolutional neural network is utilised to determine the identity of a subject in a composite sketch by comparing it to face photographs and is trained by applying transfer learning to a state-of-the-art model pretrained for face photograph recognition; 2) a 3-D morphable model is used to synthesise both photographs and sketches to augment the available training data, an approach that is shown to significantly aid performance; and 3) the UoM-SGFS database is extended to contain twice the number of subjects, now having 1200 sketches of 600 subjects. An extensive evaluation of popular and state-of-the-art algorithms is also performed due to the lack of such information in the literature, where it is demonstrated that the proposed approach comprehensively outperforms state-of-the-art methods on all publicly available composite sketch datasets.

This paper studies the problem of low rank approximation of light fields for compression. A homography-based approximation method is proposed which jointly searches for homographies to align the different views of the light field together with the low rank approximation matrices. We first consider a global homography per view and show that depending on the variance of the disparity across views, the global homography is not sufficient to well-align the entire images. In a second step, we thus consider multiple homographies, one per region, the region being extracted using depth information. We first show the benefit of the joint optimization of the homographies together with the low-rank approximation. The resulting compact representation is then compressed using HEVC and the results are compared with those obtained by directly applying HEVC on the light field views re-structured as a video sequence. The experiments using different data sets show substantial PSNR-rate gain of our method, especially for real light fields.

Application of ocular biometrics in mobile and at a distance environments still has several open challenges, with the lack quality and resolution being an evident issue that can severely affects performance. In this paper, we evaluate two trained image reconstruction algorithms in the context of smart-phone biometrics. They are based on the use of coupled dictionaries to learn the mapping relations between low and high resolution images. In addition, reconstruction is made in local overlapped image patches, where up-scaling functions are modelled separately for each patch, allowing to better preserve local details. The experimental setup is complemented with a database of 560 images captured with two different smart-phones, and two iris comparators employed for verification experiments. We show that the trained approaches are substantially superior to bilinear or bicubic interpolations at very low resolutions (images of 13×13 pixels). Under such challenging conditions, an EER of ~7% can be achieved using individual comparators, which is further pushed down to 4-6% after the fusion of the two systems.

Numerous methods that automatically identify subjects depicted in sketches as described by eyewitnesses have been implemented, but their performance often degrades when using real-world forensic sketches and extended galleries that mimic law enforcement mug-shot galleries. Moreover, little work has been done to apply deep learning for face photo-sketch recognition despite its success in numerous application domains including traditional face recognition. This is primarily due to the limited number of sketch images available, which are insufficient to robustly train large networks. This letter aims to tackle these issues with the following contributions: 1) a state-of-the-art model pre-trained for face photo recognition is tuned for face photo-sketch recognition by applying transfer learning, 2) a three-dimensional morphable model is used to synthesise new images and artificially expand the training data, allowing the network to prevent over-fitting and learn better features, 3) multiple synthetic sketches are also used in the testing stage to improve performance, and 4) fusion of the proposed method with a state-of-the-art algorithm is shown to further boost performance. An extensive evaluation of several popular and state-of-the-art algorithms is also performed using publicly available datasets, thereby serving as a benchmark for future algorithms. Compared to a leading method, the proposed framework is shown to reduce the error rate by 80.7% for viewed sketches and lowers the mean retrieval rank by 32.5% for real-world forensic sketches.

Most face super-resolution methods assume that low- and high-resolution manifolds have similar local geometrical structure; hence, learn local models on the low-resolution manifold (e.g., sparse or locally linear embedding models), which are then applied on the high-resolution manifold. However, the low-resolution manifold is distorted by the one-to-many relationship between low- and high-resolution patches. This paper presents the Linear Model of Coupled Sparse Support (LM-CSS) method, which learns linear models based on the local geometrical structure on the high-resolution manifold rather than on the low-resolution manifold. For this, in a first step, the low-resolution patch is used to derive a globally optimal estimate of the high-resolution patch. The approximated solution is shown to be close in the Euclidean space to the ground truth, but is generally smooth and lacks the texture details needed by the state-of-the-art face recognizers. Unlike existing methods, the sparse support that best estimates the first approximated solution is found on the high-resolution manifold. The derived support is then used to extract the atoms from the coupled low- and high-resolution dictionaries that are most suitable to learn an up-scaling function for every facial region. The proposed solution was also extended to compute face super-resolution of non-frontal images. Extensive experimental results conducted on a total of 1830 facial images show that the proposed method outperforms seven face super-resolution and a state-of-the-art cross-resolution face recognition method in terms of both quality and recognition.

Existing face hallucination methods are optimized to super-resolve uncompressed images and are not able to handle the distortions caused by compression. This work presents a new dictionary construction method which jointly models both distortions caused by down-sampling and compression. The resulting dictionaries are then used to make three face super-resolution methods more robust to compression. Experimental results show that the proposed dictionary construction method generates dictionaries which are more representative of the low-quality face image being restored and makes the extended face hallucination methods more robust to compression. These experiments demonstrate that the proposed robust face hallucination methods can achieve Peak Signal-to-Noise Ratio (PSNR) gains between 2-4.48dB and recognition improvement between 2.9-8.1% compared with the low-quality image and outperforming traditional super-resolution methods in most cases.

This paper presents a depth coding strategy that employs K-means clustering to segment the sequence of depth images into K clusters. The resulting clusters are losslessly compressed and transmitted as supplemental enhancement information to aid the decoder in predicting macroblocks containing depth discontinuities. This method further employs an in-loop boundary reconstruction filter to reduce distortions at the edges. The proposed algorithm was integrated within both H.264/AVC and H.264/MVC video coding standards. Simulation results demonstrate that the proposed scheme outperforms the state of the art depth coding schemes, where rendered Peak Signal to Noise Ratio (PSNR) gains between 0.1 dB and 0.5 dB were observed.

Distributed video coding (DVC) is a paradigm that can shift most of the computational intensive tasks from the encoder to the decoder. This allows for the design of low complexity encoders that can be deployed in devices equipped with limited resources. However, the compression efficiency obtained using practical DVC codecs are still distant from those of traditional predictive video coding schemes such as H.264/AVC. One of the limitations of the existing DVC architectures is that they consider the correlation noise to be randomly distributed across the whole video frame. This paper shows that the Syner–Ziv (WZ) values that are closer to the endpoints of the quantization intervals have a higher probability of producing incorrect side information (SI) predictions. Thus, through this knowledge, rate-adaptive low-density parity-check accumulate codes that provide a higher level of protection to the unreliable SI bits can be exploited. Experimental results show that the proposed scheme can reduce the WZ bit-rate up to 13% relative to the state-of-the-art DISCOVER architecture when considering interpolation techniques and by up to 0.9 dB for extrapolation techniques.

Despite theorems predicting that Distributed Video Coding can achieve the same performance as traditional predictive video coding schemes, the coding efficiency of practical architectures is still far from these bounds. This is attributed to the poor Side Information (SI) estimated at the decoder and to the inability of the channel codes to recover the source at the Slepian-Wolf (SW) limits. This paper tackles the latter issue by recovering the SI bit-planes starting from the most unreliable bit of each coefficient. Most of the mismatch in SI is thus accumulated within the first decoded bit-planes, leaving the last bit-planes with very few or no mismatch. Low-Density Parity-Check Accumulate (LDPCA) codes can then benefit from such compact distribution of correlation noise since they offer a smaller percentage error, from the SW bounds, when mismatch is accumulated in few higher entropy bit-planes. Furthermore, with this setup, most of the last bit-planes can be recovered very effectively using just 8-bit or 16-bit Cyclic Redundancy Codes. Experimental results show that the proposed scheme can reduce the Wyner-Ziv bit-rates by up to 21% compared to the DISCOVER codec.

The Distributed Video Coding (DVC) paradigm can theoretically reach the same coding efficiencies of predictive block-based video coding schemes, like H.264/AVC. However, current DVC architectures are still far from this ideal performance. This is mainly attributed to inaccuracies in the Side Information (SI) predicted at the decoder. The work in this paper presents a coding scheme which tries to avoid mismatch in the SI predictions caused by small variations in light intensity. Using the appropriate rounding operator for every coefficient, the proposed method significantly reduces the correlation noise between the Wyner-Ziv (WZ) frame and the corresponding SI, achieving higher coding efficiencies. Experimental results demonstrate that the average Peak Signal-to-Noise Ratio (PSNR) is improved by up to 0.56dB relative to the DISCOVER codec.

Motion and disparity estimations are employed in Multi-view Video Coding (MVC) to remove redundancies present between temporal and different viewpoint frames, respectively, in both the color and the depth multi-view videos. These constitute the major computational expensive tasks of the video encoder, as iterative search for the optimal mode and its appropriate compensation vectors is employed to reduce the Rate-Distortion Optimization (RDO) cost function. This paper proposes a solution to limit the number of modes that are tested for RDO to encode the inter-view predicted views. The decision is based on the encoded information obtained from the corresponding Macroblock in the Base view, identified accurately by using the multi-view geometry together with the depth data. Results show that this geometric technique manages to reduce about 70% of the estimation's computational time and can also be used with fast geometric estimations to reduce up to 95% of the original encoding time. These gains are obtained with little degradation on the multi-view video quality for both color and depth MVC.

The persistent improvements witnessed in multimedia production have considerably augmented users demand for immersive 3D systems. Expedient implementation of this technology however, entails the need for significant reduction in the amount of information required for representation. Depth image-based rendering algorithms have considerably reduced the number of images necessary for 3D scene reconstruction, nevertheless the compression of depth maps still poses several challenges due to the peculiar nature of the data. To this end, this paper proposes a novel depth representation methodology that exploits the intrinsic correlation present between colour intensity and depth images of a natural scene. A segmentation-based approach is implemented which decreases the amount of information necessary for transmission by a factor of 24 with respect to conventional JPEG algorithms whilst maintaining a quasi identical reconstruction quality of the 3D views.

Distributed Video Coding (DVC) is a coding paradigm where video statistics are exploited, partially or totally, at the decoder. The performance of such a codec depends on the accuracy of the soft-input information estimated at the decoder, which is affected by the quality of the side information (SI) and the dependency model. This paper studies the discrepancies between the bit planes of the Wyner-Ziv (WZ) frames and the corresponding bit planes of the SI. The relationship between these discrepancies is then exploited to predict the locations where the bit plane of the SI is expected to differ from that of the original WZ frame. This information is then used to derive more accurate soft-input values that achieve better compression efficiencies. Simulation results demonstrate that a WZ bit-rate reduction of 9.4% is achieved for a given video quality.

The Multi-view Video Coding (MVC) technique provides significantly better coding efficiency compared to simulcast transmission of different camera view-points. This is done by exploiting both motion and disparity compensation techniques to compress the different view-point videos. This paper proposes a more efficient MVC technique that transmits the optimally selected compensation replacements with respect to more accurate compensation vector predictors obtained using the multi-view geometry and the depth information. In this technique, the SKIP mode is also modified to adapt its compensation direction from both the temporal and the view-point reference frames. Experimental results show that the proposed MVC technique gives an average gain in video quality of about 0.5dB, for the interview predicted view-points.

Disparity estimation is used in Multi-view Video Coding (MVC) to remove the inter-view redundancies present in both color and depth multi-view video sequences. The standard H.264/MVC achieves high compression efficiency by deriving the optimal disparity vector through the exhaustive calculation of the Rate-Distortion cost function for all the possible search points. This makes disparity estimation highly computational expensive. This paper proposes an efficient technique that exploits both the multi-view and the epipolar geometries to determine the optimal search area, resulting in a reduction of search points and thus computations. Simulation results show that this technique can save up to 95% of the computational cost for disparity estimation, with negligible loss in coding efficiency for both the color and the depth multi-view video coding.

Distributed Video Coding (DVC) is a coding paradigm which shifts the major computational intensive tasks from the encoder to the decoder. Temporal correlation is exploited at the decoder by predicting the Wyner-Ziv (WZ) frames from the adjacent key frames. Compression is then achieved by transmitting just the parity information required to correct the predicted frame and recover the original frame. This paper proposes an algorithm which identifies most of the unreliable bits in the predicted bit planes, by considering the discrepancies in the previously decoded bit plane. The design of the used Low Density Parity Check (LDPC) codes is then biased to provide better protection to the unreliable bits. Simulation results show that, for the same target quality, the proposed scheme can reduce the WZ bit rates by up to 7% compared to traditional schemes.

Real-time wireless conversational and broadcasting multimedia applications offer particular transmission challenges as reliable content delivery cannot be guaranteed. The undelivered and erroneous content causes significant degradation in quality of experience. The H.264/AVC standard includes several error resilient tools to mitigate this effect on video quality. However, the methods implemented by the standard are based on a packet-loss scenario, where corrupted slices are dropped and the lost information concealed. Partially damaged slices still contain valuable information that can be used to enhance the quality of the recovered video. This study presents a novel error recovery solution that relies on a joint source-channel decoder to recover only feasible slices. A major advantage of this decoder-based strategy is that it grants additional robustness while keeping the same transmission data rate. Simulation results show that the proposed approach manages to completely recover 30.79% of the corrupted slices. This provides frame-by-frame peak signal-to-noise ratio (PSNR) gains of up to 18.1 dB, a result which, to the knowledge of the authors, is superior to all other joint source-channel decoding methods found in literature. Furthermore, this error resilient strategy can be combined with other error resilient tools adopted by the standard to enhance their performance.

This paper presents a novel error-resilient strategy which employs a reversible watermarking technique to protect the H.264/AVC video content. The proposed scheme adopts reversible watermarking to embed an error detection codeword within every Macro block (MB). The watermark is then extracted at the decoder and used to detect the corrupted MBs to be concealed. The proposed scheme further manages to recover the original video content after watermark extraction, thus providing no loss in video quality. The simulation results demonstrate that the proposed approach provides a substantial gain of up to 2.6 dB in Peak Signal-to-Noise Ratio (PSNR) relative to the standard with a minimal increase in complexity.

Multi-view video coding exploits inter-view redundancies to compress the video streams and their associated depth information. These techniques utilize disparity estimation techniques to obtain disparity vectors (DVs) across different views. However, these methods contribute to the majority of the computational power needed for multi-view video encoding. This paper proposes a solution for fast disparity estimation based on multi-view geometry and depth information. A DV predictor is first calculated followed by an iterative or a fast search estimation process which finds the optimal DV in the search area dictated by the predictor. Simulation results demonstrate that this predictor is reliable enough to determine the area of the optimal DVs to allow a smaller search range. Furthermore, results show that the proposed approach achieves a speedup of 2.5 while still preserving the original rate-distortion performance.

This letter presents a hybrid error control and artifact detection (HECAD) mechanism which can be used to enhance the error resilient capabilities of the standard H.264/advanced video coding (AVC) codec. The proposed solution first exploits the residual source redundancy to recover the most likelihood H.264/AVC bitstream. If error recovery is unsuccessful, the residual corrupted slices are then passed through a pixel-level artifact detection mechanism to detect the visually impaired macroblocks to be concealed. The proposed HECAD algorithm achieves overall peak signal-to-noise ratio gains between 0.4 dB and 4.5 dB relative to the standard with no additional bandwidth requirement. The cost of this solution translates in a marginal increase in the complexity of the decoder. In addition, this method can be applied in conjunction with other error resilient strategies and scales well with different encoding configurations.

The H.264/AVC standard employs a number of error-resilient mechanisms to correct transmission errors. These methods assume a packet-loss scenario, where all the macroblocks (MBs) contained within a corrupted slice are dropped and concealed. However, most of the MBs contained within corrupted slices provide minimal (if any) visual distortions and therefore concealing them causes a superfluous drop in the quality of the recovered video content. This paper presents a novel error control mechanism which employs prior information and residual source-redundancy to recover the most-likelihood feasible H.264/AVC bitstream. Simulation results show that the algorithm recovers a number of corrupted sequences and achieves overall peak signal-to-noise ratio (PSNR) gains between 1 dB and 2 dB over the standard. The proposed solution is compatible with the H.264/AVC with no additional bandwidth requirements.

Ka-band satellite multimedia communication networks play important roles because of their capability to provide the required bandwidth in remote places of the globe. However, because of design complexity, in practice they suffer from poor design and performance degradation because of being practically forced to guarantee acceptable end-user satisfaction in conditions of extremely low bit error rates, which is emphasised with the vulnerability of compressed video content to transmission errors, often impossible to be applied during the service development phase. A novel discrete event simulation model is presented, which provides performance estimation for such systems based on subjective measurement and a better quality of experience. The authors show that the proposed model reduces implementation cost and is flexible to be used for different network topologies around the globe.

Wireless multimedia services are increasingly becoming popular boosting the need for better quality-of-experience (QoE) with minimal costs. The standard codecs employed by these systems remove spatio-temporal redundancies to minimize the bandwidth required. However, this increases the exposure of the system to transmission errors, thus presenting a significant degradation in perceptual quality of the reconstructed video sequences. A number of mechanisms were investigated in the past to make these codecs more robust against transmission errors. Nevertheless, these techniques achieved little success, forcing the transmission to be held at lower bit-error rates (BERs) to guarantee acceptable quality. This paper presents a novel solution to this problem based on the error detection capabilities of the transport protocols to identify potentially corrupted group-of-blocks (GOBs). The algorithm uses a support vector machine (SVM) at its core to localize visually impaired macroblocks (MBs) that require concealment within these GOBs. Hence, this method drastically reduces the region to be concealed compared to state-of-the-art error resilient strategies which assume a packet loss scenario. Testing on a standard H.263++ codec confirms that a significant gain in quality is achieved with error detection rates of 97.8% and peak signal-to-noise ratio (PSNR) gains of up to 5.33 dB. Moreover, most of the undetected errors provide minimal visual artifacts and are thus of little influence to the perceived quality of the reconstructed sequences.

The transmission of H.264/AVC encoded sequences over noisy wireless channels generally adopt the error detection capabilities of the transport protocol to identify and discard corrupted slices. All the macroblocks (MBs) within each corrupted slice are then concealed. This paper presents an algorithm that does not discard the corrupted slices but tries to detect those MBs which provide major visual artefacts and then conceal only these MBs. Results show that the proposed solution, based on a set of image-level features and two support vector machines (SVMs), manages to detect 94.6% of those artefacts. Gains in peak signal-to-noise ratios (PSNR) of up to 5.74 dB have been obtained when compared to the standard H.264/AVC decoder.

Current trends in wireless communications provide fast and location-independent access to multimedia services. Due to its high compression efficiency, H.264/AVC is expected to become the dominant underlying technology in the delivery of future wireless video applications. The error resilient mechanisms adopted by this standard alleviate the problem of spatio-temporal propagation of visual artifacts caused by transmission errors by dropping and concealing all macroblocks (MBs) contained within corrupted segments, including uncorrupted MBs. Concealing these uncorrupted MBs generally causes a reduction in quality of the reconstructed video sequence.

Forward error correction (FEC schemes are generally used in wireless communication systems to maintain an acceptable quality of service. Various models have been proposed in literature to predict the end-to-end quality of wireless video systems. However, most of these models utilize simplistic error generators which do not accurately represent any practical wireless channel. A more accurate way is to evaluate the quality of a video system using Monte Carlo techniques. However these necessitate huge computational times, making these methods unpractical. This paper proposes an alternative method that can be used in modeling of complex communications systems with minimal computational time. The proposed three random variable method was used to model two FEC schemes adopted by the digital video broadcasting (DVB) standard. Simulation results confirm that this method closely matches the performance of the considered communication systems in both bit error rate (BER) and peak signal-to-noise ratio (PSNR).