+ General Information
+ Members
+ Research
+ Intranet

PRAIg '16 / +pics


AERFAI Summer School on Deep Learning


Alicante July 5-6, 2017

10th international workshop on Machine Learning and Music


Barcelona: 16th October 2017



  1. Iñesta, J.M., Buendía, M., Sarti, M.A.
    "Reliable polygonal approximations of Imaged Real Objects through dominant point detection"
    Pattern Recognition, vol. 31, pp. 685-699 (1998)
    : bibtex

    The problem of data reduction in contours via dominant points is posed, taking into account what usually happens in practice. The algorithms found in the literature often prove their performance with laboratory contours (i.e., artificial curves designed by the authors in order to test their algorithms), but the shapes in real images are quite different: noise, quantization, and high inter and intra-shape variability are effects that should be taken into account. The presence of noise has already been faced by using a gaussian smoothing. To deal with variability only an algorithm working independently of input parameters, adjusting itself to each curve, could give satisfactory results if an efficient and reliable representation of the contours is desired. Here we focus in the problem of image quantization. The addition to the existent non parametric algorithms of a criterion for deleting collinear points is very useful, since it increments the compression rate, without raising the error committed by the polygonal approximation. Thus, the number of segments necessary to reliably represent the shape of the contours is reduced. For evaluating it, a new measurement of the error has been defined that assess the goodness of the approximation with a single value, taking into account both the compression rate and the reliability of the selected set of points. We will also focus on the conditions for an efficient (few points) and precise (low error) dominant point extraction that preserves the original shape. A measurement of the committed error (optimization error, E0) that takes into account both aspects is defined for studying this feature. Some classic algorithms are reviewed and compared to ours, showing that the latter fits well to data using few points (low E0), so accurate and efficient approximations are obtained.

@article {
 author = "Iñesta, J.M., Buendía, M., Sarti, M.A.",
 title  = "Reliable polygonal approximations of Imaged Real Objects through dominant point detection",
 journal = "Pattern Recognition",
 number = "6",
 pages = "685-699",
 volume = "31",
 year = "1998"
Valid XHTML 1.0!Valid CSS!