A Novel Rendering Strategy for SVG Vectorization

Vector representation of digital images offers a number of advantages over the more common raster representation, such as scalability and resolution independence. These features make it amenable for portable applications since it can accomodate for a wide range different displaying conditions, varying in resolution, quality, and level of detail. Many efforts have been made to deploy scalable raster standards for photographic imagery addressed to portable applications, such as JPEG2K [JPE]. Anyway, since they have been focused on raster images, they lack the flessibility and simplicity of vector representation. On the other hand, while many applications exist to enable artists to build vector images from scratch, converting photographic imagery from raster to vector formats is a relatively new topic. Recently, SVG (Scalable Vector Graphics), a new vector format for web deployment, has been released ([DHH02], [Qui03]). A number of applications have appeared to convert raster images to vector graphics in the SVG format (Vector Eye [VLDP03], Autotrace [Web02], Kvec [Kuh03], VISTA [PS05]). Anyway, most of these methods are devoted to synthetic images with a small colour palette and strong neat borders between image regions. They often fail to vectorialize photographic images, because they have blurred and fuzzy edges and huge colour palettes. As we already shown in ([BFP06b], [BFP06a]), our technique SVGStat, outperforms other methods both in terms of rendering quality and overall compression rate. Moreover it is based on a single input parameter that makes easy to find the correct trade-off between final perceived quality, scalability and corresponding file size.

SVGStat is a raster to vector technique that consists of three main steps:

In this paper we have introduced a boundaries simplification step in the algorithm that permits us to obtain smaller file size without losing in perceived and measured quality. Previous SVGStat algorithm codifies every region separately. This representation is not optimal in terms of coding efficiency; in fact each border is considered twice. In order to reduce such redundancy the rendering order is used to simplify some boundaries. Just for example considering two nearby regions Ri and Rj, some borders of Ri can be extended under the image region covered by Rj with ordering such that i<j. Such simplification strategy is useful to reduce the overall number of points and primitives required by the classical boundary description.

[BFP06a] BATTIATO S., FARINELLA G. M., PUGLISI G.:Statistical Based Vectorization for Standard Vector Graphics.In Fifth Int.Workshop on Computer Graphics and Geometric Modeling.(to appear) LNCS (2006), vol. 5670.3.
[BFP06b] BATTIATO S., FARINELLA G. M., PUGLISI G.: SVG Vectorization by Statistical RegionMerging. In Proocedings of Fouth Conference Eurographics Italian Chapter (2006), Catania (Italy).
[DHH02] DUCE D., HERMAN I., HOPGOOD B.: Web 2D Graphics File Format. Computer Graphics forum 21, 1(2002), 43–64.
[JPE] : ISO/IEC JTCI/SC29/WG! N1646: JPEG2000 final committee draft v1.0. http://www.jpeg.org/jpeg2000/.
[Kuh03] KUHL K.: Kvec 2.99, 2003. Copyright KKSoftware,http://www.kvec.de.
[NN04] NOCK R., NIELSEN F.: Statistical Region Merging. IEEE Transaction on Pattern Analysis and Machine Intelligence 26, 11 (NOVEMBER 2004), 1452–1458.
[PS05] PRASAD L., SKOURIKHINE A.: Raster to Vector Conversion of Images for Efficient SVG Representation. In Proceedings of SVGOpen’05 (August 2005), NL.
[Qui03] QUINT A.: Scalable Vector Graphics. IEEE Multimedia 3 (2003), 99–101.
[VLDP03] VANTIGHEM C., LAURENT N., DECKEUR D., PLANTINET V.: Vector eye 1.0.7.6, 2003. Copyright SIAME e Celinea, http://www.siame.com, http:// www.celinea.com.
[Web02] WEBER M.: Autotrace 0.31, 2002. GNU General Public License, http://www.autotrace.sourceforge.net.