SIGNAL AND DATA PROCESSING
Year:2013 | Volume:- | Issue:1 (SERIAL 19)|Papers Count:6

Transmission of compressed video over error prone channels may result in packet losses, which can degrade the image quality. Error concealment (EC) is an effective approach to reduce the degradation caused by the missed information. The conventional temporal EC techniques are always inefficient when the motions of the video object are irregular. In this paper, in order to overcome this problem, an efficient temporal EC approach to conceal the macroblock error for video coding systems is proposed. The proposed EC method employs a RBF neural network to estimate the motion vectors of the damaged macroblocks. RBF estimator is used only for areas of the fast motions, which reduces computation complexity. Because the neural networks have a great capacity for visualizing and interpreting high-dimensional data sets, the estimation model proposed herein can exploit the nonlinearity property of the neural networks to estimate lost motion vectors more accurately. Simulation results show that the proposed technique enhances both subjective and objective quality of reconstructed frames, such as the average PSNR increases about 1.5 dB compared to the BMA method for the test video sequences in some frames.

In this paper, we propose efficient method for pre-training of deep bottleneck neural network (DBNN). Pre-training is used for initial value of network weights; convergence of DBNN is difficult because of different local minimums. While with efficient initial value for network weights can avoided some local minimums. This method divides DBNN to multi single hidden layer and adjusts them, then weighs of these networks is used for initial value of DBNN weights and then train network. Proposed network is used for extraction of face component. This Method is implemented on Bosphorus database. Comparing results shows that new method has more convergence speed and generalization than random initial value. By means of this new training method and with same training error rate pixel reconstruction error is decreased 13.69% and recognition rate is increased 10%. Besides, it has been shown that this method bears higher efficiency and convergence speed in comparison with some of the previous pre-training methods.

Design of new feature extraction methods out of the speech signal and combination of their obtained information are the most effective approaches to improve the performance of automatic speech recognition (ASR) system. Recent researches have been shown that the speech signal contains nonlinear and chaotic properties, but the effects of these properties were not used in the continuous ASR systems. Reconstructed phase space (RPS) is an appropriate domain to exhibit nonlinear properties of a chaotic signal. Therefore, in this paper a new method is proposed to utilize the RPS-based features (LLRPS). These features will be computed using similarity scores between the embedded speech signal in the RPS and a set of predefined phoneme manifolds. Then, TMLP-based neural network estimates phoneme posterior probability over the LLRPS features. This network includes some useful properties such as extracting dynamic information and output combination methods. Experimental results using Farsdat speech database show that nonlinear combination of the speech recognition outputs including traditional MFCC features and LLRPS features, leading to improvement of 3.94% and 4.02% in the accuracy of frame and phoneme recognition, respectively.

Watermarking systems have specific feathers in accordance with their applications. In many applications transparency and robustness are needed which are the most important features. These two features are in contrast to each other and are controlled by a parameter named watermark strength. With decreasing the watermark strength, transparency of the watermarking system increases while the robustness of the watermarking system decreases and vice versa. Having these two features is not possible at the same time in a watermarking system and there should be tradeoff between transparency and robustness choosing correct watermark strength. In this paper, the imperialist competitive algorithm is used for determining the watermark strength for having transparency and robustness at the same time. The simulation results show that the imperialist competitive algorithm can propose proper watermark strength for a watermarking system with less computational complexity.

In this paper, we present a Compressive Sampling (CS)-based feature extraction method for audio signals. In the proposed approach, the audio signal is firstly segmented by hamming windows and the Discrete Fourier Transform (DFT) of the samples is calculated within each frame. Then, the normalized values of the DFT coefficients of each frame are accumulated. At the next step, the second DFT is applied on the vector formed from the accumulated sum in consecutive frames. Finally, considering the sparseness of the resulted vector, our proposed CS-2FFT feature vector is achieved by a random sampling. In this research, the performance of CS-2FFT feature vector has been examined in the applications of audio classification and audio source localization. The simulation show that the proposed feature vector results in a classifier which is more accurate and less computationally complex compared to the classical classifiers. Also, it is shown that the employing CS-2FFT feature vector, the localization error will be less than 2%.

Obtaining an image with high spectral and spatial resolution is the goal of image fusion. PCA is a well-known pan-sharpening approach widely used for its efficiency and high spatial resolution. However, it can distort the spectral characteristics of the multispectral images. To avoid the weak points of the standard PCA technique, spatial PCA transform has been proposed and reasons of superiority of this method in maintaining the spectral information are discussed in this paper. Also, a new assessment criterion is proposed and the advantage of this criterion relative to the conventional mutual information criterion is argued. The proposed assessment metric and other popular metrics such as: ERGAS, SAM, correlation coefficient UIQI, and mutual information are used to analyze the fusion result. These assessments show that the proposed method has the least color distortions and contains more spatial information.