Compound Image Segmentation for Multiscale Recurrent Pattern Coding
Rodrigues, Nuno M. M.
Compound Image Segmentation for Multiscale Recurrent Pattern Coding, Proc Conf. on Telecommunications - ConfTele, Santa Maria da Feira, Portugal, Vol. 1, pp. 267 - 270, May, 2009.
Digital Object Identifier:
In this paper we propose a new compound image segmentation algorithm, specially developed for a digital document encoding scheme based on theMultidimensionalMultiscale Parser (MMP) encoding algorithm. In order to increase theMMP algorithm performance, a block based segmentation defines a mask that divides the input image into two types of blocks: text-like blocks and non-text, or smooth image-like blocks. Different versions of the MMP algorithm are then applied on each image layer. The segmentation algorithm presented in this paper is based in two steps. The first step relies on morphological filtering and image gradient analysis. In the second step, false text blocks are eliminated using a morphological based algorithm. Experimental results show that this algorithm is not only efficient for our encoder, but also improves the segmentation accuracy in relation to other methods proposed in the literature.