The Study On Algorithms For Chain Code Generation, Region Filling And Unevenly Lighted Document Segmentation

As an important tool in digital image processing, chain code transfers the two-dimensional image to one-dimensional codes, which greatly reduces storage space for the image, and holds certain advantages in image processing and analysis. This paper introduces common types of chain code and the contour tracing and chain code generation algorithm by using automaton.Characters in document image are normally composed by nested chain code, while only the most outer layer is useful in document analysis. This paper proposes a new code pattern for multi-connected regions by combining Freeman chain code and run-length code. The new coding handles nested chain code as a unit, which provides convenience for image analysis and character recognition.Region filling is one of the common problems in graphics and image processing. This paper presents a new region filling algorithm based on the pair of boundary nodes, which limits the scan within the filling regions, and requires no additional working memory or assistant colors. Repeated experiments and theory analysis have proved that the new algorithm can correctly fill any complex inner and outer contours and the execution speed is higher than REN algorithm which is the fastest one so far.As the popularization of photo-taking devices, it is convenient to aquire unevenly lighted document images. However, there does not exist a very efficient segmentation algorithm for these images. This paper presents a new method for binarizing unevenly lighted document images. An illumination compensation method is proposed to convert unevenly lighted document to evenly lighted document. The compensated image is then binarized through an improved locally adaptive approach. The new method effectively overcomes the difficulty of ha(?)dling uneven lighting and remarkably enhances segmentation quality. Experimental results show that the proposed method is more efficient to process unevenly lighted document than traditional binarization methods.