Generalization of Multi-linear Feature Based on Ordinary Least Squares Regression

For the preparation of small-scale maps from large-scale maps, generalization of vector features is important. This method increases the quality of maps published, enables the analysis of data at various levels of detail, and reduces the volume needed to store them. The methods of linear and polygonal features generalization are performed with the aim of preserving their geometry and area while reducing their details. Various models have been used and evaluated by researchers in this field; However, most of them summarize the features with the aim of selecting a few points from them and deleting other points. Even so, the deleted points may contain valuable information for this complication and their removal will lead to a defect in its geometry and area. In this study, the generalization of multi-linear features was performed using minimizing the vertical distance from the main line. In order to study the proposed model, after its implementation on different shapes, the multi-lines of Lake Urmia and its islands were generalized and the results of the proposed model were compared with the common Douglas-Poker and Viswalingam methods. The results were then evaluated using the indices of area differences, the similarity of the mean curvature, the similarity of the amount of angle changes and the modified average Hausdorff distance. The results showed an average superiority of 99.91, 66.29 and 60.99% compared to conventional simplification approaches with proposed model (in the first three indicators). The proposed model has a advantage over the Douglas-Poker and Viswalingam methods 0.16 and 0.2 percent based on the area difference index, 7 and 5 percent based on the average curvature index, and 6 and 2 percent based on the  sharp change index. but In the the modified average Hausdorff distance index,  it was about 2 meters worse than the aforementioned methods, due to the lack of reliance on the initial points of the complication.