In this paper

firstly the distribution characteristic of accumulate granule on the image gray space is analyzed under the different angles and distributions of light ray

the foundation is provided for abstracting granule contour information. A non-supervisory roof edge detection method is proposed based on a logic judgement

which includes four steps:(1) The analysis window consisting of 3×3 sub-windows is assumed

where each sub-window consist ofl×lpixels

the size of sub-window is related to the size of objects. (2) The gray average value of 3×3 sub-windows is calculated. (3) The gray average values of 3×3 sub-windows is compared respectively along four directions

if the gray average value of center sub-window is smaller than one of other two sub-windows in some direction

the center or some near center pixel in the sub-window is labeled as the candidate of edge pixel. (4) Scanning original image get the edge bit-map by the analysis window. (5) The false edge pixels in the bit-map can be removed by the proposed false edge filtering method. This method need not to the conventional procedure of abstracting edge pixels through calculating edge intensity and then thresolding edge intensity

is suited to the roof edge detection. The experiment results of the granule contour abstraction in accumulate global granule image proved that the method is effective.