Scalable Vector Graphics

"SVG is a language for describing two-dimensional graphics and graphical applications in XML" (from W3 Consortium).


The aim of this research is to perform automatic translation between scenes/sequences acquired by image capturing devices in classical raster format and a vectorial representation using the SVG (Scalable Vector Graphics) standard.

To obtain SVG images with high perceived quality we plan to investigate/use classical image processing techniques such as: image classification, edge detection, image segmentation, etc.. In order to cope the gap between the two differents worlds, expecially for "natural" scenes, initially the research will be restricted to obtain SVG rendering of specific classes of images, properly classified (e.g. face, text images).

Also the possiblity to work directly on the actual output of the camera sensor, the socalled "Bayer Pattern" will be investigated. The main goal of the research will be related to the final rendering on specific viewing condition (e.g. LCD with limited bit-depth and size resolution) using the SVG scalability. Also the SVG Mobile Profile will be considered, providing a complete documentation about characteristics and differences in term of related performances with the full standard profile.

Main research topics involved will regard "smart" techniques (Fuzzy B-splines, subpixel offset interpolation, anisotropic diffusion, multiscale filtering, ? ) able to obtain an effective SVG image from the rgb world maintaining acceptable quality level. As first step the Data Dependent Triangulation technique will be evaluated as possible candidates for continuous representation of raster images. Also the possibility to use some kind of pre-processing in order to discard redundant information and/or extracting key-features will be investigated.

We plan to exploit the pre-classification of the input images to define heuristics for each class under investigation and to use this source of additional information to compensate for the unavailable user input. Here follows the schedule for the proposed research:

  1. Bibliographical research about image vectorialization.
  2. Definition of heuristics for each class of images to enhance, integrating sensor information avoiding user intervention.
  3. Definition of architectures for edge-map estimation based on the above mentioned heuristics.
  4. Software implementation of the architectures and test on significant images for each class.
  5. Analysis of the test results and refinement of the implemented techniques.
  6. New software implementation, test, and validation steps.
  7. Software prototype release.

In what follow is described a possible prosecution of the research activity, after having attained the main goal:

  1. Definition of a technique to perform automatic classification of input images.
  2. Software implementation of the technique and test on significant images for each class.
  3. Analysis of the test results and refinement of the implemented techniques.
  4. New software implementation, test, and validation steps.
  5. Software prototype release.

The whole activity will involve the Image Processing research group of the DMI(Dipartimento di Matematica ed Informatica) of the University of Catania, and the Imaging and Multimedia Mobile research group of the AST-Catania Laboratory of the STMicroelectronics.