Laurent Wendling

Fast polygonal approximation of digital curves

In this paper, we have extended the approach defined in to a multiorder analysis. The approach is based on the arithmetical definition of discrete lines with variable thickness. We provide a framework to analyse a digital curve at different levels of thickness. The extremities points of a segment provided at a high resolution are tracked at lower resolution in order to refine their locations. The high resolution level is automatically defined from the stability of the number of segments between two consecutive levels. The method is threshold-free and automatically provides a partitioning of a digital curve into its meaningful parts.

Recognition of symbols in grey level line-drawings from an adaptation of the Radon transform

An adaptation of the Radon transform is proposed to recognize grey level symbols. We directly process on the grey level document in order to improve the recognition step. A tridimensional signature is defined which takes into account the variations of photometry and the shape of a symbol. The proposed signature also allows to keep fundamental geometric transformations (scale, translation and rotation). Experimental results on technical documents show the promising aspect of our approach.

Retrieving images by content from strong relational graph matching

A new method is proposed to retrieve images by content. The approach is based on the matching of strong graphs defined by relational signatures computed between clusters of a fuzzy partition. Such a representation combines color and spatial information between the regions and have nice geometric properties w.r.t scale factor, rotation and translation.

Binarization of color images from an adaptation of possibilistic c-means algorithm

A color image binarization is presented in this paper. The iterative possibilistic c-means algorithm is adapted by adding a fuzzy entropy criterion to split the membership function into two clusters (background and object). Such an improvement allows to perform a threshold free color binarization. Experimental results show the promising aspect of our approach.