SVG Vectorization by Statistical Region Merging

Abstract

In this paper a novel algorithm for raster to vector conversion is presented. The technique is mainly devoted to vectorize digital picture maintaining an high degree of photorealistic appearance specifically addressed to the human visual system. The algorithm makes use of an advanced segmentation strategy based on statistical region analysis together with a few ad-hoc heuristics devoted to track boundaries of segmented regions. The final output is rendered by Standard Vector Graphics. Experimental results confirm the effectiveness of the proposed approach both in terms of perceived and measured quality. Moreover, the final overall size of the vectorized images outperforms existing methods.