Plenary Lecture

Image Representation with Inverse Difference Pyramid: Algorithms and Applications

Professor Roumen Kountchev
Faculty of Telecommunications
Technical University of Sofia
Bulgaria
E-mail: rkountch@tu-sofia.bg

Abstract: The speech is dedicated to one new approach for still image decomposition called Inverse Difference Pyramid, (IDP). Unlike the famous decompositions: Laplacian Pyramid, Enhanced Laplacian Pyramid, Reduced Differences Pyramid, Hierarchy Embedded Differential Image, Polynomial Approximation Pyramid, Morphological Pyramid, Discrete Wavelet Transform, etc., the new pyramid is build in the image spectrum domain, starting the calculations of the consecutive pyramid layers from it’s top. The essence of the IDP decomposition is presented as follows. First, the digital image is processed with some kind of orthogonal transform (DCT, WHT, KLT, etc.) using limited number of coefficients only. The values of the coefficients, calculated in result of the transform, constitute the lowest pyramid level. Then, using these values, the image is restored with Inverse Orthogonal Transform. In result is obtained the first (coarse) approximation of the original image, which is then subtracted pixel by pixel from the original one. The difference image, which is of same size as the original, is divided into 4 sub-images and each is processed with the orthogonal transform again. The values of the so calculated coefficients constitute the second pyramid level. The processing continues in similar way with the next pyramid layers. The set of the orthogonal transform coefficients, chosen for every pyramid layer, can be different and defines the restored image quality. The image decomposition is stopped when the required quality of the approximating image is obtained – usually earlier than the last possible pyramid layer.
A variety of the IDP is the pyramid decomposition with error Back Propagation Neural Networks (BPNN), called Learning IDP (LIDP). Instead of the spectrum coefficients, in this pyramid are used a small number of nodes in the hidden layer of the participating neural networks. This approach ensures higher efficiency of the image representation compactness.
The basic features of the IDP image representation are analyzed in respect of computational complexity, compactness of the image description, ability for recursive implementation, etc., which define the advantages of the IDP decomposition in various application areas.
The experimental results obtained for the IDP and LIDP algorithms for lossy and lossless compression for large number of test images are compared with the results for the standards JPEG and JPEG 2000. The advantages of the developed algorithms are shown for such applications as multi-layer progressive image transfer, content-based image data mining in large data bases, hierarchical match evaluation for image fusion, multi-layer watermarking, etc.

Brief Biography of the Speaker:
Roumen Kountchev, Ph.D., D. Sc. is a professor at the Faculty of Telecommunications at the Technical University of Sofia, Bulgaria and the head of the Image Processing Laboratory.
His main areas of interest are: Digital image processing, Image compression, Multimedia watermarking, Video communications via Internet, Pattern recognition and neural networks. He has 259 papers published in magazines and proceedings of conferences; 12 books and books chapters, 20 patents, and participated in 46 scientific research projects (in 38 projects he was the principal investigator).
He is the President of the Bulgarian Association for Pattern Recognition (BAPR), member of International Association for Pattern Recognition (IAPR), member of editorial board of “International Journal of Reasoning-based Intelligent Systems” (IJRIS), member of the Scientific Expert Commission of Bulgarian Ministry of Education and Science; President of the Technological Council of Bulgarian National Radio, member of the Higher Attestation Commission of the Council of Ministers of Bulgaria.