Aiming at the problems of serious feature loss and more holes in the traditional point cloud simplification algorithm, a point cloud simplification algorithm considering the features and integrity of point clouds was proposed. Firstly, the whole feature points of the model are extracted by using the neighborhood normal Angle of the point cloud. Then the fuzzy C-means clustering algorithm is used to extract the local feature points according to the curvature of the point cloud and the fast point feature histogram. Then the non-local feature points are subsampled using the improved voxel reduction method to obtain the non-feature points. Finally, the point clouds obtained from each step are fused to obtain the final reduced point cloud. The proposed algorithm is compared with the traditional methods and method of other literature, and the Hausdorff distance, a quantitative index describing the error between datasets, is used as the evaluation index of the simplification accuracy. The experimental results show that the Hausdorff distance of the proposed algorithm on Bunny dataset and Skull dataset is about 25% and 39% lower than that of the random reduction method, about 86% and 95% lower than that of the curvature reduction method, and about 86% and 81% lower than method of other literature. It can be seen that the simplification algorithm in this paper has high simplification accuracy.