In this paper a field programmable gate array (FPGA) is considered as a digital signal processing platform for the implementation of an exponential sine sweep measurement algorithm. Aiming at minimizing the required computational resources, two strategies are proposed. Firstly, an oscillator implemented with the coordinate rotation digital computer (CORDIC) algorithm is used to generate the exponential sine sweep. Secondly, only the calculations are performed that lead to the linear impulse response of the system for a desired length. Furthermore, aiming at minimizing the required memory resources, the measured impulse response is stored in the memory previously allocated to the recorded signal. In order to validate the proposed implementation, measurements of an acoustical system are performed using a platform that is equipped with an FPGA and a processor. In this way, the results achieved by the FPGA fixed-point implementation can be compared to reference results achieved using a floating-point MATLAB implementation running on the processor. This comparison corroborates the validity of the proposed implementation.

We develop a new contour tracing algorithm to enhance the results of the latest object contour detectors. The goal is to achieve a perfectly closed, single-pixel wide and detailed object contour, since this type of contour could be analyzed using methods such as Fourier descriptors. Convolutional Neural Networks (CNNs) are rarely used for contour tracing, and we see great potential in using their capabilities for this task. Therefore we present the Walk the Lines (WtL) algorithm: A standard regression CNN trained to follow object contours. As initial step, we train the CNN only on ship contours, but the principle is applicable to other objects. Input data are the image and the associated object contour prediction of the recently published RefineContourNet (RCN). The WtL gets the center pixel coordinates, which defines an input section, plus an angle for rotating this section. Ideally, the center pixel moves on the contour, while the angle describes upcoming directional contour changes. The WtL predicts its steps pixelwise in a selfrouting way. To obtain a complete object contour the WtL runs in parallel at different image locations and the traces of its individual paths are summed. In contrast to the comparable Non-Maximum Suppression (NMS) method, our approach produces connected contours with finer details. Finally, the object contour is binarized under the condition of being closed. In case all procedures work as desired, excellent ship segmentations with high IoUs are produced, showing details such as antennas and ship superstructures that are easily omitted by other segmentation methods.

A convolutional neural network is proposed for the reduction of artifacts introduced by JPEG compression. In this work the existing DnCNN architecture has been considered as the basis network and it is extended by introducing inception blocks on top of a normal convolution block consisting of convolution, rectified linear unit and batch-normalization layers. Each inception block contains parallel convolution blocks with various dilation factors, essentially performing Atrous convolution. Compression artifacts result in multiple image neighborhoods or blocks having similar or redundant information. In spatial domain the Atrous convolution introduces a larger receptive field while ignoring redundant neighboring pixels due to the dilation, thus helping in a better reconstruction. The proposed network is trained on a small dataset as used by the basis network. Evaluation of this network on the standard test datasets shows a substantial quantitative and qualitative improvement, particularly for higher compression rates.

In this work a convolutional neural network with two input channels is proposed for image super-resolution. Initially, the original image is downsampled with bicubic and nearest neighbour interpolation methods and the low-resolution image pair is used as the input to our network. Additionally, the input channels are randomly swapped as an augmentation method during training. The proposed network is a combination of a feedforward architecture and a residual network architecture, to provide a smooth image estimate. We show that the additional input image obtained with nearest neighbour interpolation improves image super-resolution performance with our network, since it acts as an image specific prior information and constrains the solution space for the model to learn. We train the network end-to-end for multiple upscale factors, evaluate the method on standard test datasets, and show that the proposed approach can produce state-of-the-art results.

A novel digital radio-frequency (RF) condenser microphone system is proposed. Condenser microphones using the audio-frequency (AF) implementation method basically suffer from their high impedance sensor circuitry. The RF implementation method solves this problem. In RF condenser microphones, the inherent impedance can be lowered leading to superior noise performances. Since state of the art RF microphones are still entirely analog, a digital approach is presented which can further increase the demodulation linearity and can circumvent the 1/f noise as well as additional analog disadvantages.

Nowadays, many different image processing applications are of high interest to maritime authorities because of security reasons. Depending on the application, different kinds of images are employed. The extraction of ship silhouettes requires high resolution images in order to obtain accurate results. However, when the characteristics of the naval environment are visible the background complexity increases greatly and automatic approaches fail. In order to overcome these difficulties we propose an automatic initialization for graph segmentation techniques. A comparative study of earlier suggested initializations for different graph segmentation techniques is also presented. It shows that, under such unfavorable image conditions, finding the proper initialization in an automatic way is not trivial. Yet, the precision and recall achieved by our initialization are considerable higher regardless the graph segmentation. Furthermore, the performance is highly increased since the best results are obtained after only the first iteration.

This work comprises an extension of a backward adaptive quantizer which is employed together with a robust lattice predictor in an ADPCM coding scheme. Predictors of the ADPCM audio coding schemes are often considered as the part most sensitive to transmission errors. Nevertheless, a single transmission error causes a short destabilization of the adaptive quantizer at the decoder side. Therefore, this destabilization boosts the deviation of the prediction filter at the decoder from the encoder side. Desired damping of the quantizer shortens synchronization periods. However, damping leads to degradation of the quantizer's adaptation properties and consequently to a decrease in audio quality. We show that the transmission of the quantizer's envelope to the decoder in short intervals and in combination with envelope error detector reduces the impairment of the reconstructed audio signal in noisy transmissions. An objective audio quality evaluation confirms significant quality improvement at BER higher than 10-4and no quality degradation if an ideal channel is employed.

A visual saliency based approach for macroblock (MB) mode selection in video compression is presented. It is based on the Laplace distribution of transformed residuals. Visual saliency of a MB is used to make the encoder vote for a mode which requires less bits in a region which is visually less important. It shows a gain of up to 0.6 dB in terms of PSNR compared to conventional rate-distortion optimization (RDO) procedure in H.264/AVC. Moreover, it suggests to be a competitive alternative to a well-established perceptually based video coding method. At the same time it has no negative impact on the PSNR.

Due to the ever-growing importance of video content delivered through today's Internet, not only the availability of high-definition video material but also its high quality must be guaranteed. The latter can be achieved by means of objective video quality metrics. Recently, insufficient quality prediction accuracy for well-known full-reference objective quality metrics on high-definition video content has been reported. With this paper we aim at identifying possible reasons for the observed imperfection.

The timber of pitched musical instruments is analyzed through the excitation signature by means of Higher Order Statistics (HOS) and subspace analysis. To describe the features of instrument sounding mechanism, the excitation signature is proposed, which decisively characterizes the musical instrument category rather than the difference within one kind of instrument family. Subspace analysis is applied to get more efficient timbre representations for musical instrument classification. Experimental results show that HOS based features provide more significant timbre patterns in both time and frequency domain in comparison with the 2ndorder statistics features. Dimensional reduction of excitation signature is also considered for the efficiency of musical instrument classification.

The 16 papers in this special issue focus on virtual audio effects and musical instruments.

In audio codec design, often various parameters have to be fixed which may have a dramatic impact on codec performance. In this paper, we report on successful optimization of a codec based on perceptual criteria. Specifically, the PEAQ measure is used to determine the audio quality over a set of test items and search algorithms are used for optimization. First, simulated annealing is used for global search, then Rosenbrock's method is used to further refine the result. As shown by an example, the improvements gained by optimization compared to an educated guess are substantial.

A delay-free audio coding scheme based on ADPCM with adaptive pre- and post-filtering is presented. The pre-/post- filters are realized as a cascade of shelving filters, designed to match the characteristics of human perception. The pre- and post-filters are adapted by dynamic compression of the respective sub-bands. The adaption is backward-adaptive, i.e. is fed by the reconstructed signal, which eliminates the need to transmit the filter coefficients and allows delay-free operation. This pre- and post-filtering significantly improves the audio quality compared to a plain ADPCM codec, as underlined by objective measurements. Since the base ADPCM used is also delay-free, the resulting coding system works without any algorithmic delay.

With no calibrated camera setup at hand, careful inspection of the imagery is needed to guarantee a feasible 3D reconstruction result based upon the images. We propose a new approach for image registration based on reconstructed 3D octrees by voxel carving. Correlation of these models gives rise to a translation offset for a maximum intersection between different models from different images. Projecting the resulting three-dimensional translation offsets back into the image plane results in two two-dimensional image offsets that are used for the image registration.

We propose an extension of ADPCM that includes adaptive pre- and post-filtering to achieve spectral shaping of the coding noise. The advantage of this coding scheme is that it allows a realization without algorithmic delay by making the filters backwards-adaptive. The measurements we present indicate that the addition of adaptive pre- and post-filtering to ADPCM results in a significant improvement in perceived audio quality. We therefore believe that the proposed system is a viable way to near-transparent lossy audio coding without algorithmic delay.

The semi-blind equalization for a wireless multiple-input multiple-output (MIMO) system with frequency selective Ricean channels is addressed. A reformulation of the problem accepting higher complexity allows to apply ordinary complex Independent Component Analysis (ICA). An algorithm presented here resolves the increased number of permutations due to the reformulation of the convolutive blind signal separation problem. The remaining ambiguities are as many as in the non-frequency selective case and solved by a short preamble. The efficiency of the proposed method is illustrated by numerical simulations. According to the Bit Error Rate the semi-blind equalization shows a good performance in comparison to training based channel estimation & equalization.

After covering the basics of sound perception and giving an overview of commonly used audio effects (using a perceptual categorization), we propose a new concept called adaptive digital audio effects (A-DAFx). This consists of combining a sound transformation with an adaptive control. To create A-DAFx, low-level and perceptual features are extracted from the input signal, in order to derive the control values according to specific mapping functions. We detail the implementation of various new adaptive effects and give examples of their musical use.

Estimating depth information from two-dimensional stereoscopic image pairs is an important task in image processing. The disparity, i.e. the displacement of corresponding pixels in the image pair, can be estimated directly from local image phase in the Fourier domain. In this paper a combined FFT-based approach for disparity estimation is proposed, where additional magnitude information is included in the disparity estimation process. Confidence criteria derived from the magnitude spectra are used to eliminate errors which makes this approach both easier and more robust compared to single phase-based algorithms. Results are presented for an artificial image pair. A stereo image coding approach is applied to compare the results in an objective manner.

In this paper, we develop a parametric recursive second-order lowpass/highpass shelving filters and fourth-order bandpass shelving filters. These filters allow a better frequency selectivity for audio equalization as compared to the well-known first-order lowpass/highpass shelving filters and second-order peak filters. In the parametric structures, the cut-off frequencies and the gain factors can be set independently. We derive the filter designs for the discrete-time implementation. Furthermore, we apply a technique for the implementation of delay-free loops which is necessary to yield the exact inverse filter to realize symmetric boost and cut of a frequency band.

In stereovision systems, the depth information of objects in the scene can be obtained by estimating the disparity, i.e. the displacement of corresponding pixels in the image pair. In this paper a new FFT-based algorithm for disparity estimation is proposed. The phase difference between the two images is calculated at several frequencies for each pixel. These phase differences have a linear coherence which can be used to make the approach robust and to avoid local errors. The resulting disparity can then be estimated from the gradient of these phase differences. In the second part, these disparity maps are used for an efficient stereo image coding approach. Here, the transmission of both left and right image is not necessary. At the receiver, e.g. the right image can be reconstructed having only the left image and the disparity map.