email us
call us now


Keynote&Plenary Speakers

Prof. Dr. Smain Femmam, Strasbourg University of Haute Alsace, France

Smain FEMMAM is Director of research at the University of Haute-Alsace France and responsible of the Research team on Signals & Safety Systems of Polytechnic Engineers School Sceaux France. He received the MS and Ph.D. degrees in Signal Processing and Computers from Versailles University, France in 1997 and 1999 respectively. Since 2013 he is promoted to a rank of senior director of research (HDR). After that, he joined the CMU Carnegie Mellon University & West Virginia University as Postdoc Fellow and Distinguished Visiting Professor. His main research area is signal processing, safety systems, communication and embedded systems. He has a strong interest in perception and characterization of signals, optimal filtering, spectral analysis, wavelets and perception haptics. Dr. Femmam is a senior member of IEEE, SEE and GDR ISIS. Member of IEEE C.A. committee France section. Board of Director of the Institute for Engineering and Technology Innovations in the World. He has guided numerous thesis projects, including some doctoral theses. He is active reviewer for several scientific academic journals and committee member of international conferences. He is the (Academic, Chief) Editor-in-Chief, Editor, Editorial Board, Guest Editor & Advisory Board members of more than 20 International Journals. He has authored and co-authored more than 90 papers, four Chapters, one edited Book, seven Books published by Wiley & Elsevier.

Speech Title: Perception of Texture Classification Based on Multifractal Descriptor, Hybrid Time-Frequency and Detection of Abrupt Changes


The significance of the work:
We present a new approach for the texture classification based on multifractal descriptor. The presented approach tackles the problems of texture characterization and provides interesting results in order to identify defects in textures. The originality of the present work can be found in the proposed multiresolution quantities that improve the performance of multifractal spectrum based on wavelet leaders. These quantities will be called: Maximum coefficients of Discrete Wavelet Transform (DmaxWT) for 2D multifractal analysis. The performance of the proposed method is evaluated using both synthetic and experimental data. The experiments showed that our texture descriptor gives good performs in terms of classification accuracy on several texture datasets. Hybrid time-frequency is also used for perception of materials with detection of abrupt changes.

The problem that is being addressed:
We show explicitly the practical steps of numerical calculation of the multifractal spectrum. From fractal spectrum, selected discriminative features are applied to texture classification. The performance of the estimation procedure (bias and variance) has been analyzed by generating a large number of realizations of synthetic processes. The application of our method in texture classification context of early diagnosis of skin cancer shows that it is possible to discriminate between healthy tissue and pathologic tissue. Moreover, in the context of food quality, milk classification based on the multifractal descriptor indicates that our contribution enables the effective discrimination of different texture: the experiment consists of 30 classes of non-traditional textures, composed of 25 samples of high resolution images in each class and the same for the low resolution images. We show how the DmaxWT multifractal analysis is pertinent for texture classification. Indeed, we present a new analysis method for discriminating milk heat treatment effects. In texture classification context, multifractal features are estimated from each image and used in the classification for characterization within a standard k-NN (k-nearest neighbor) classification approach. We then conduct a comparative evaluation with state-of-the-art recognition methods including HLSR method, VG-fractal method and VZ-joint method. The experimental results on several texture datasets verified the accuracy and efficiency of the classification based on the multifractal descriptor. Finaly we state a new descriptor which uses a hybrid representation time-frequency and detection of abrupt changes.