Multiscale Segmentation for Mixed Raster Content Applicable to Document Coding

Compound document images contain graphic or textual content along with pictures. They are found in magazines, brochures, web-sites, etc in a document format. The goal is to compress an image containing the mixed raster content (MRC) using multi-layer approach. The proposed methodology segments the image into regions such as text, pictures and background. The key to MRC compression is the separation of the document into foreground and background layers, represented as a binary mask. The compression quality depends on the segmentation algorithm used to compute the binary mask.
The proposed multi-scale segmentation algorithm models the complex aspects of both local and global contextual behavior. The proposed algorithm finds the block-wise segmentation of the raster image in a global cost optimization framework. Then the initial segmentation is refined by classifying feature vectors of connected components using a Markov random field (MRF) model. Then hybrid procedures of the previous steps are then incorporated into a multi-scale framework in order to improve the segmentation accuracy of text with varying size. It is shown that the proposed methodology achieves greater accuracy of text detection but with a lower false detection rate of non-text features. This segmentation algorithm can improve the quality of decoded documents while simultaneously lowering the bit rate. It is also shown that execution time can be greatly reduced by the use of features that are not computationally intensive.