Research On High Resolution Remote Sensing Image Segmentation Based On Mean Shift

The rapid development of remote sensing technology led to the enhancement of remotely sensed data acquisition technology, allowing higher accuracy and larger-scale remotely sensed data to become the main object in the field of remote sensing. Therefore, the more targeted technique needs to be provided because the traditional remotely sensed image techniques are not apply to the current high resolution and large-scale image. Image segmentation technology is a key step in the process of remote sensing image processing and provides convenience for the operation of image classification, feature identification and so on, so the accuracy of the result is especially important on the final image quality.In this paper, the study on high resolution remote sensing image, which is based on Mean Shift algorithm combined with the features of remote sensed image, has been processing, and the main work and achievements are as followed:1. Introduce the main principle of Mean Shift algorithm, including the selection for critical part and its features. And introduce the basic process of image segmentation, which elaborates the process and effect of key process on Mean Shift.2. The optimized improvement has been carried out on segmentation efficiency and segmentation accuracy of high resolution remote sensing image segmentation based on Mean Shift algorithm. At first, an optimized algorithm for parallel segmentation of large-scale image on Mean Shift has been proposed to deal with some problems of traditional segmentation technique, such as the low efficiency and low memory, or even impossible to separate. At the same time, it can eliminate the “block line” existed in block parallel process to some extent, which improves the segmentation accuracy. And the results is compared with eCognition software to prove the effectiveness of this algorithm. Secondly, make the study for distance metric involved in merge process of remote sensing image segmentation, which replace the tradition Euclidean distance with the similarity metric in line with the feature of remotely sensed data. In the application of high resolution remote sensing image segmentation on Mean Shift, the relevant experiment data has proved the replaced similarity metric has better effect on the process of image segmentation.3. Make the study for related bandwidth and scale parameters on Mean Shift segmentation of remote sensing image. Through the experimental analysis of the different combination of parameters and scale experiments, it has been basis on determining the suitable parameters of image segmentation for better performance.