Traditional landscape painting has many elements, different painting methods, no uniform rules of object shape, and the classification of some landscape paintings is based on the deep meaning of painting itself, which increases the difficulty of classification and retrieval. In order to improve the feature extraction effect of landscape painting, this paper proposes a reliable image segmentation algorithm combined with machine learning technology, which implements 3D modeling of landscape painting in 3DsMax scene by collecting basic data information of landscape painting in real time. Moreover, this paper reconstructs three-dimensional model based on panel multi-view stereo (PMVS) algorithm, and extracts mountain contour curve according to the characteristics of natural landscape pictures, and determines the artistic conception type of natural landscape pictures by analyzing mountain contour. In order to verify the feature extraction effect of landscape painting, the algorithm proposed in this paper is compared with ELM (Extreme Learning Machine), SVM (Support Vector Machine, SVM) and BPNN (Back Propagation Neural Network). Finally, this paper verifies that the algorithm model proposed in this paper has certain advantages through experiments, which can promote the progress and development of intelligent technology of landscape painting recognition.