With the significant improving for the spatial resolution of remote sensing imagery, the limitation of the traditional pixel-based methods for medium and low resolution remote sensing image have become obvious. In recent decades, the Object Based Image Analysis (OBIA) has become the most popular information extraction method for the high spatial resolution remote sensing imagery. The object is the based processing units in the OBIA method, so that the segmentation method obtaining the objects is a key step in the OBIA, because the classification accuracy is directly affected by the quality of segmentation results. However, the objects in high-resolution remote sensing images show multi-scale characteristics, and it is difficult to accurately obtain the optimal segmentation results using the single segmentation scale, so that he multi-scale segmentation method have become an inevitable choice. But the multi-scale segmentation algorithms proposed in previous literatures are difficult to achieve global and local optimization. In this paper, a new multiscale segmentation optimized algorithm was proposed. The algorithm mainly includes following steps: (1) the global optimal scale in multiscale segmentation was obtained for using the local variance criterion; (2) the over-segmentation and under-segmentation objects in global optimized scale were respectively optimized to obtain the local optimization results; (3) the global and local optimized results were fused to obtain the finishing optimized results. In this paper, the two high spatial resolution remote sensing images which respectively located in Dongguan, China and Florida, USA were used to verify the effectiveness of the proposed algorithm, and the experimental results were analyzed by the qualitative and quantitative evaluation .The results were shown following as: (1) From the perspective of visual effects, the more accurate segmentation boundary were obtained in the optimized results, and the objects (such as road, farmland, and water) in the large scale maintain better regional feature, and the objects (such as tree, house, shadow)in the small scale have more detail information. (2) From the perspective of quantitative analysis by the evaluation indicators(RR, RI and ARI), the presented algorithm increased the RR, RI and ARI by 2.1%, 2.4 and 30.2% in comparison with the global optimized segmentation scale, and by 8.3%，0.1% and 8.1% in comparison with the k-means optimized method, and by 0.7%，0.4% and 17.6% in comparison with the fused boundary optimized method in the test 1, and increased the RR,RI and ARI by 4.5%，2.7% and 29.3% in comparison with the global optimized segmentation scale, and by 17%，0.8% and 8.4% in comparison with the k-means optimized method, and by 1.7%，2.5% and 17.2% in comparison with the fused boundary optimized method in the test 2. In summary, compared with classical segmentation algorithms, the proposed algorithm obtained the best segmentation results by both local and global optimization, and reduced over-segmentation and under-segmentation objects in the segmentation results. Meanwhile, the heterogeneity of objects is different in the different types of scenes, for example, the objects in the city scenes are high heterogeneity, but the objects of the rural scenes are high homogeneity. So that, the optimal segmentation parameters in multi-scale segmentation is difficult to other scenes.