Papers by Keyword: Sparse Representation

Abstract: Ultrasonic signal reconstruction for Structural Health Monitoring is a topic that has been discussed extensively. In this paper, we will apply the techniques of compressed sensing to reconstruct ultrasonic signals that are seriously damaged. To reconstruct the data, the application of conventional interpolation techniques is restricted under the criteria of Nyquist sampling theorem. The newly developed technique - compressed sensing breaks the limitations of Nyquist rate and provides effective results based upon sparse signal reconstruction. Sparse representation is constructed using Fourier transform basis. An l₁-norm optimization is then applied for reconstruction. Signals with temperature characteristics were synthetically created. We seriously corrupted these signals and tested the efficacy of our approach under two different scenarios. Firstly, the signal is randomly sampled at very low rates. Secondly, selected intervals were completely blank out. Simulation results show that the signals are effectively reconstructed. It outperforms conventional Spline interpolation in signal-to-noise ratio (SNR) with low variation, especially under very low data rates. This research demonstrates very promising results of using compressed sensing for ultrasonic signal reconstruction.

Abstract: This paper proposed a new face recognition algorithm based on Haar-Like features and Gentle Adaboost feature selection via sparse representation. Firstly, All the images including face images and non face images are normalized to size and then Haar-Like features are extracted . The number of Haar-Like features can be as large as 12,519. In order to reduce the feature dimension and retain the most effective features for face recognition, Gentle Adaboost algorithm is used for feature selection. Selected features are used for face recognition via sparse representation classification (SRC) algorithm. Testing experiments were carried out on the AR database to test the performance of the new proposed algorithm. Compared with traditional algorithms like NS, NN, SRC, and SVM, the new algorithm achieved a better recognition rate. The effect of face recognition rate changing with feature dimension showed that the new proposed algorithm performed a higher recognition rate than SRC algorithm all the time with the increasing of feature dimension, which fully proved the effectiveness and superiority of the new proposed algorithm.

Abstract: De-blurring the high resolution remote sensing images is an important issue in the relative research field of remote sensing. In this paper a novel algorithm of de-blurring the high resolution remote sensing images is proposed based on sparse representation. The high spatial resolution remote sensing images can be de-blurred by gradient projection algorithm, and keep the useful information of the image. The experimental results of the remote sensing images obtained by “the first satellite of high resolution” show that the algorithm can de-blur the image more effectively and improve the PSNR, this method has better performance than other dictionary learning algorithm.

Abstract: This paper, both theoretically and numerically, investigates an effective reconstruction of EEG signal. An optimization model is presented, which unifies different sparse signals. The model is solved by employing the proximal algorithm. Based on the theoretical analysis, the simulation of EEG signal is performed. Sparse representation of EEG signal is got by the technique of wavelet transform and the signal denoising is also obtained. Then, by using compressed sensing, the EEG signal is reconstructed. Our results show that the reconstructed signal is in good agreement with the original signal and retains the leading characteristic.

Abstract: Recent years, the image sparse representation has been the popular method in the study of image representation, which has put forward a new idea in the image denoising. Its basic principle is that the original image has the sparse representation under the proper over-complete dictionary. Filter out the noise, we should find out the sparse representation of the image through the design of the dictionary. Its mechanism is that one hand the useful information of the image would be effectively expressed because of the sparse decomposition algorithm based on the redundant dictionary. The other the noise would not be expressed through the dictionary atoms. We do the image denoising according to the image sparse representation. Because of the superiority of the adaptive dictionary algorithm in the image, in this paper, we discuss the over-complete dictionary training algorithm. And we prove the effectiveness through the MATLAB.

Abstract: In signal analysis, as a new representation, the sparse representation caused widespread concern of scholars at home and abroad, and signal processing and analysis produced a very significant impact. Feather and Down is closely related to people's lives, different kinds of down, the price difference is bigger, and thermal properties are different, and therefore, feather species identification has always been an important issue. This paper studies the sparse representation in image processing, while the types of detection of Down key technologies studied, proposed a new algorithm to detect the type of feather sparse representation. In this algorithm, an improved sparse we will de-noising method based on the introduction of a global atomic library come down on the previous species detection algorithm has been improved. The algorithm is applied to obtain a certain effect, making feather and down recognition rate has improved to some extent, the effect is significant.

Abstract: Image denoising method based on K-SVD self-learning dictionary can effectively filter Gaussian white noise in an image and retain image details and texture information. This paper proposed an improved denoising algorithm which was based on K-SVD algorithm, but with adaptive dictionary size. Depending on the complexity of image content and the noise level, our algorithm determines the size of the dictionary adaptively. Experimental results show that proposed algorithm can reduce the number of entries of the dictionary significantly for the simple images, and increase the number of entries for the complex images. Both the efficiency and the denoising performance are improved compared with original K-SVD algorithm.

Abstract: In this paper, we propose a novel speech coding scheme based on compressed sensing and sparse representation. Compressed sensing (CS) attracts great interest for its ability to utilize a few measurements to recover original signals. Measurements preserve part of speech features while projected by row echelon matrix. A dictionary is learned in order to contain redundant information about speech measurements. The synthesized speech is recovered from a sparse approximation of the corresponding measurement. A rear low-pass filter is adopted to improve the subject quality of synthesized speech. Results show that the proposed coding scheme has achieved average Mean Opinion Score (MOS) of the synthesized speech 3.083 in an appropriate bit rate (4.2 Kbps), which outperforms the quality of Code excited linear prediction (CELP).

Abstract: In order to suppress the noise, improve equipment's ability to further process information and improve the quality of voice, speech enhancement is often an important part of the speech signal preprocess. Contrastively analyze the characteristic that the clean speech signal coefficients in over-complete discrete cosine dictionary are much sparser than the traditional discrete cosine transform coefficients. Under noisy conditions, by setting the iterative threshold of orthogonal matching pursuit (OMP) algorithm, clean speech can be gotten, thus realize the speech enhancement. Simulation results of the signal waveform and spectrogram enhanced by the proposed algorithm are very similar to the original signal,comparative experiments also indicate that the signal to noise ratio (SNR) and the perceptual evaluation of speech quality (PESQ) score of the processed signal are superior to traditional discrete cosine transform (DCT).

Abstract: In recent years, sparse representation has become a very popular method for pattern recognition which could outperform the traditional methods. This paper presents a novel combination of sparse representation and traditional Gaussian mixture models. Each person’s dictionary or termed as subspace in this paper are learned using K-SVD algorithm while the entries are GMM mean matrixes union for each speaker. Then project the test utterance into each dictionary and finally make decision depending on the reconstruction errors. The experiments are conducted on the database collected in our anechoic chamber. The proposed approach results in different accuracy for different sparsity and dictionary size. In appropriate parameters, the accuracy can reach 98.5% which is fairly good.