Seminario "Reconstruction and Visualization of 2D and 3D Images Using Wavelets"
Firstly, this lecture presents short review of novel class of functions: Atomic Functions (AF) and following from them novel families of wavelets: Wavelet Atomic Function (WAF). Second part of this lecture exposes several examples of successful applications of WAF in different problems explained below.
a) 3D Video Visualization Employing Wavelet Multilevel Decomposition: Novel approach based on classic wavelets and WAF has been efficiently employed in 3D video sequence visualization. The procedure consists of multilevel decomposition and 3D visualization applying color anaglyphs synthesis. Simulations on synthetic images and video sequences, also on real-life video sequences have confirmed sufficiently better proposal performance in depth and spatial perception in comparison with existing methods. We present the real-life 3D videos reconstructed from 2D video sequences on DSP.
b) Super-resolution applying classic wavelets and WAF: Novel method in high-resolution uses WAF in the reconstruction of color and greyscale video sequences of different types. The approach is theoretically justified by analysis based on key wavelet properties (cosine projection, Reisz values, etc.). Statistical simulation results have shown that method based on WAFs performs better at improving resolution than do existing frameworks, both in terms of objective criteria and based on the more subjective measure of human vision. Implementations on DSP have demonstrated the possibility of real-time processing.
c) Mammography features classification in wavelet transform space: The principal mammography (MG) signs of breast cancer are clustered in the microcalcifications (MCs) and masses. Because of the nonspecific features of malignant lesions, MG interpretation is a very challenging task for radiologists; therefore, there existed several methods of automatic classifications. Novel method for masses and MCs classification in the MG employs Wavelet Transform (WT) based on classical and WAF, decomposing MG for reducing data volume in the classification stage that is performed via multilayer artificial neural network (ANN) type classifier. The experimental results have shown sufficiently good performance of proposal on real data, showing better rate of recognition for benign, malign and normal MGs. The implemented scheme permits to reduce the iterations number during the training of the ANN MLP applying WT. Daubechies wavelets and WAF have been adapted in mentioned network in classification of MG principal features: MCs, spiculated and circumscribed masses, presenting sufficiently better classification results than existed methods can do.
Dr. Volodymyr Ponomaryov. National Polytechnic Institute of Mexico, ESIME-Culhuacan, Mexico-city.