Accepted Papers

  • Objective Evaluation of A Deep Neural Network Approach for Single-Channel Speech Intelligibility Enhancement
    Dongfu Li and Martin Bouchard, University of Ottawa, Canada
    Single-channel speech intelligibility enhancement is much more difficult than multi-channel intelligibility enhancement. It has recently been reported that machine learning training-based single-channel speech intelligibility enhancement algorithms perform better than traditional algorithms. In this paper, the performance of a deep neural network method using a multi-resolution cochlea-gram feature set recently proposed to perform single-channel speech intelligibility enhancement processing is evaluated. Various conditions such as different speakers for training and testing as well as different noise conditions are tested.Simulations and objective test results show that themethod performs better than another deep neural networkssetup recently proposed for the same task, and leads to a more robust convergence compared to a recently proposed Gaussian mixture model approach.
  • Automated Drone for Road Anomaly Detection (ADRAD)
    Kausic Gunasekar,Alex Noel Joseph Raj and Vijaylakshmi GV, VIT University, India
    The aim of this project is to build an automated system using drones (quad copters in this case) to identify abnormalities on the road such as speed-breakers and potholes which could be dangerous when they are not clearly visible during bad weather conditions or in the absence of proper lighting. The anomalies are identified and detected using a computer vision algorithm and after detection the anomalies are geo-tagged and the information is updated to a server for further use and display.
  • Fingerprint Identification by Wave atoms Transform and SVM
    Leila Boutella and Amina Serir,University in Bab Ezzouar, Algeria
    In Automatic Fingerprint Identification System (AFIS), the performance highly depends on features extraction. In this paper, a new minutiae detection method based on wave atoms transform is presented. The proposed approach is based on analyzing fingerprints by specific representation highlighting the oscillatory structure. Thereby, a fine analysis from local characteristics via the block wave atoms transform is introduced. We explain how minutiae are detected by the use of wave atoms transform associated to a support vector machine (SVM) classifier. In order to evaluate the algorithm, FVC 2002 fingerprint databases have been considered. The tests show that for fingerprints features extraction, the wave atoms transform offers a good performance.
  • Cosine Similarity Based Dictionary Learning And Source Recovery for Classification of Diverse Audio Sources
    K V Vijay Girish, T V Ananthapadmanabha and A G Ramakrishnan, Indian Institute of Science, India
    A dictionary learning based audio source classification algorithm is proposed. Cosine similarity measure is used to select the atoms during dictionary learning. Based on three proposed objective measures, namely, signal to distortion ratio (SDR), the number of non-zero weights and the sum of weights, a frame-wise source classification accuracy of 98.2% is obtained for twelve different sources. 100% accuracy has been obtained using moving SDR accumulated over 14 successive frames. For ten of the audio sources tested, 100% accuracy requires accumulation of only 6 frames of a signal.
    Yuanyuan Zhang and Fuxiang Wang,Beihang University, Beijing, China
    Recently, visual tracking based on sparse principle component analysis has drawn much research attention. As we all know, principle component analysis (PCA) is widely used in data processing and dimensionality reduction. But PCA is difficult to interpret in practical application and all those principal components are linear combinations of all variables. In our paper, a novel visual tracking method based on sparse principal component analysis and L1 tracking is introduced, which we named the method SPCA-L1 tracking. We firstly introduce trivial templates of L1 tracking method, which are used to describe noise, into PCA appearance model. Then we use lasso model to achieve sparse coefficients. Then we update the eigenbasis and mean incrementally to make the method robust when solving different kinds of changes of the target. Numerous experiments, where the targets undergo large changes in pose, scale and illumination, demonstrate the effectiveness and robustness of the proposed method.
  • Evaluation of Data Mining Algorithms In High Dimensional Datasets
    Silvio Normey1,2, Joaquim Assuncao 3and Walkiria Cordenonzi 1, 1Federal Institute of Education, Brazil, 2University of the Republic, Montevideo, Uruguay and 3Pontificia Catholic University of Rio Grande do Sul, Brazil
    The need for good mining algorithms grows along with the complexity and volume of databases. The difficulty of understanding systems and obtaining patterns makes the importance of data mining be relative to the volume, variability and speed that the data are generated. However, the dimensionality of the data implies the need for efficient algorithms. This study has its pillars on two algorithms commonly used to analyze high dimensional data: Random Forests (RF) and Sequential Minimal Optimization (SMO). Widely known and awarded in the available literature, RF has proven to be an efficient algorithm. On the other hand, SMO is relatively unknown. This study reports a series of experiments involving these algorithms on large and high dimensional datasets, with meaningful results.
Copyright ® SIPR 2016