An Improvement Of A Region-based Graph Image Segmentation Method

In this paper, we mainly consider the problem of image segmentation andfocus ongrey-level images. First, a recently published segmentation method is introduced. Thisis a region-based segmentationmethod by graph representation. It defines a newmeasurement of the internal difference within and between areas. By comparing thismeasurement, one can decide whether to merge areas or do nothing. By analyze thismethod, we make some modification of the definition of the measurement, and establishour own measurement of the internal difference within and between each areas. Also,more measurement is established in order to get better results. Meanwhile,since theoriginal algorithm is not stable, pre-segmentation procedure is proposed to be donebefore segmentation. As a result,we can get more information to reduce the inevitablemistakes brought by the original model. Computational results show that our method isbetter for the gray-image for there is not too much elevation between the gray levelvalues of the pixels.