Fraunhofer-Institut für Produktionsanlagen und Konstruktionstechnik IPK

Filters

2
2
Reset filters

Settings
Now showing 1 - 2 of 2
  • Publication
    Using codebooks of fragmented connected-component contours in forensic and historic writer identification
    ( 2007)
    Schomaker, L.
    ;
    Franke, K.
    ;
    Bulacu, M.
    Recent advances in "off-line" writer identification allow for new applications in handwritten text retrieval from archives of scanned historical documents. This paper describes new algorithms for forensic or historical writer identification, using the contours of fragmented connected-components in free-style handwriting. The writer is considered to be characterized by a stochastic pattern generator, producing a family of character fragments (fraglets). Using a codebook of such fraglets from an independent training set, the probability distribution of fraglet contours was computed for an independent test set. Results revealed a high sensitivity of the fraglet histogram in identifying individual writers on the basis of a paragraph of text. Large-scale experiments on the optimal size of Kohonen maps of fraglet contours were performed, showing usable classification rates within a non-critical range of Kohonen map dimensions. The proposed automatic approach bridges the gap between image-statistics approaches and purely knowledge-based manual character-based methods.
  • Publication
    Tiny GAs for image processing applications
    ( 2006)
    Köppen, M.
    ;
    Franke, K.
    ;
    Vicente-Garcia, R.
    The expedience of today's image-processing applications is no longer based on the performance of a single algorithm alone. These systems appear to be complex frameworks with a lot of sub-tasks that are solved by specific algorithms, adaptation procedures, data handling, scheduling, and parameter choices. The venture of using computational intelligence (CI) in such a context, thus, is not a matter of a single approach. Among the great choice of techniques to inject CI in an image-processing framework, the primary focus of this presentation will be on the usage of so-called Tiny-GAs. This stands for an evolutionary procedure with low efforts, i.e. small population size (like 10 individuals), little number of generations, and a simple fitness. Obviously, this is not suitable for solving highly complete: optimization tasks, but the primary interest here is not the best individual's fitness, but the fortune of the algorithm and its population, which has just escaped the Monte-Carlo domain after random initialization. That this approach can work in practice will be demonstrated by means of selected image-processing applications, especially in the context of linear regression and line fitting; evolutionary post processing of various clustering results, in order to select a most suitable one by similarity; and classification by the fitness values obtained after a few generations.