Szendrei, Rudolf  Elek, István  Fekete, István  Márton, Mátyás
2010 Conference of PHD Students in Computer Science (CS^{2}), Szeged, Hungary, June 29  July 2, 2010. Organizing committee: Kálmán Palágyi, Balázs Bánhelyi, Tamás Gergely, Zoltán Kincses Abstract
In the IRIS project the authors have elaborated the theoretical background of a rastervector conversion system, and they have developed the prototype of some components of the system. The aim of the development is to automatize the rastervector conversion as much as possible. This goal puts an emphasis on the knowledge based approach. This article will focus on the automatic recognition and conversion of the three main types of map symbols, to improve the efficiency of the recognition system. Pointlike symbols are small icons each representing a real object (e.g. a monument). The recognition algorithm tries to identify these symbols based on given symbol patterns. Each connected pixel set under a given size limit will be matched against the data base of patterns. Surfacelike symbols cover a region with a solid color, or with a pattern (e.g. lake or scrub). The procedure first determines the smallest repetitive part (kernel) of the texture which can be identified by the algorithm used for pointlike symbols. In order to identify linear symbols (e.g. roads, railroad) both line style and topology must be recognized. To determine topology a graph is created using the end and forkpoints of the roadlike graphics. Currently, the automatic recognition of some kinds of map symbols (e.g. texts) is beyond the scope. Thus, the vectorized coverage generated automatically does not contain all of the elements occurring on the original raster map. Furthermore, the algorithms used for recognition provide the possibility for human expert's intervention in the case of false detection. An important point in the expertise of human interpreters is, for example, the knowledge of the order of map layers they have been printed in. The inclusion of this knowledge would make the conversion much more intelligent.


Szendrei, Rudolf
2010 International Conference on Artificial Intelligence and Pattern Recognition (AIPR10), Orlando, Florida, USA, July 1214, 2010. ISBN: 9781606510155, Publisher: ISRST Editors: Zoran Majkic, Dan Tamir, Guoyin Wang Abstract
We will show a method that is able to automatically parallelize a sequential image filter. Furthermore, we will show another method that decreases the memory consumption of raster image filters. The later one uses a virtual image, so the input image can be safely overwritten with the result image. By using these methods, beginner programmers could even use the benefits of the multicore processor based computers, turning their filters into faster and multithreadsafe filters. We have carried out a basic testing of kernel based raster image filters, and we have experienced a linear, n times speedup factor, where n is the number of CPU cores.


Szendrei, Rudolf  Elek, Istvan  Fekete, Istvan
Riga Series 11, Vol. 6, 2009 Abstract


Szendrei, Rudolf  Elek, Istvan  Fekete, Istvan
2009 International Conference on Artificial Intelligence and Pattern Recognition (AIPR09), Orlando, Florida, USA, July 1316, 2009. ISBN: 9781606510070, Publisher: ISRST Editors: Dimitrios A. Karras, Zoran Majkic, Etienne E. Kerre, Chunping Li Abstract
This procedure is presented with optimized pattern matching on the raster image source of the map, where the symbols are handled as special textures. This method will be improved by using a raw vector model and the kernel symbols.

Iványi, Antal  Szendrei, Rudolf
Abstract
A binary matrix A = [A_{ij}]_{mxn} is called good,
if it contains at most one 1's in each column; the matrix is called schedulable,
[5,6,8] if it can be transformed into a good matrix repeating the following operation:
we remove any zero element a_{ij} = 0, shift the elements
a_{i,j+1}...a_{i,n} to left and add a new element a_{i,n}
= 0 to the ith row of the matrix; the matrix is called safe, if for any k
(k = 1,...,n) holds that the first k column of the matrix contain at most
k 1's.
Let Z = [z_{ij}]_{mxn} be a matrix containing independent random variables having the join distribution P(z_{i,j} = 1) = p and P(z_{i,j} = 0) = 1  p. Investigating the features of the good and safe matrices in the case m >= 1 we give lower and upper bound of the asymptotic probability [1,2,3,4,7] of the event that a concrete realisation of the matrix Z is schedulable. 

Szendrei, Rudolf
Abstract
A binary matrix A = [A_{ij}]_{mxn} is called good,
if it contains at most one 1's in each column; the matrix is called schedulable,
[5,6,8] if it can be transformed into a good matrix repeating the following operation:
we remove any zero element a_{ij} = 0, shift the elements
a_{i,j+1}...a_{i,n} to left and add a new element a_{i,n}
= 0 to the ith row of the matrix; the matrix is called safe, if for any k
(k = 1,...,n) holds that the first k column of the matrix contain at most
k 1's.
Let Z = [z_{ij}]_{mxn} be a matrix containing independent random variables having the join distribution P(z_{i,j} = 1) = p and P(z_{i,j} = 0) = 1  p. Investigating the features of the good and safe matrices in the case m >= 1 we give lower and upper bound of the asymptotic probability [1,2,3,4,7] of the event that a concrete realisation of the matrix Z is schedulable. Full Text (PDF) , PowerPoint Presentation (PPT) 