Automatic road extraction from satellite images by using chaincode method

It is one of the most important research subjects of recent years to have current road maps in terms of both navigation and Geographic Information Systems. Because of data size it takes a long time to process and it makes automating this process is very important. Therefore, in this study, a new semi-automatic method has been tried to be developed for the detection of land roads in satellite images. The system consists of five main stages. The first stage is the edge detection in the satellite image. The second step is to obtain the chaincode for each edge by applying chaincode method. The third stage is the automatic detection of similar edges by comparing their chaincodes by longest common subsequence algorithm. At the end of this step the areas between the similar edges will be marked as possible road pixels. The fourth stage is to calculate the average gray color values of the pixels specified as the road in the previous stage and eliminate the pixels smaller than this value. The last step is to determine the actual road pixels by giving the pixels as the feed value to the region growing algorithm, which is likely to be the road in the previous step. The developed method was applied on four different satellite images those includes roads in the non-residential areas. Their average performance values were 88%, specifity 96%, accuracy 94%, F1 score 86%, respectively. The results were compared with another method in the literature using the same satellite images, and the proposed system was able to achieve approximately the same results with fewer steps.