Research On The High Spatial Resolution Remote Sensing Image Segmentation Integrating Edge Confidences

The application of remote sensing data lags far behind the development of space technology. The appearance of high spatial resolution remote sensing images has brought new opportunities for the development of remote sensing technology and increased the depth of traditional remote sensing applications. But the processing of these data is far from satisfying the needs of practical applications. The development of intelligent remote sensing image data analysis and understanding techniques, the simulation of remote sensing image interpretation of the human brain’s cognitive processes by computer, getting the semantic information from high-resolution images based on the purpose of practical applications are important tasks of image interpretation. At present, the object-based remote sensing image analysis technique has become a high resolution image processing of the basic method, has received the widespread concern. The object-based image analysis is that homogeneous pixel set (image object) as the basic processing units, through the object or objects in between spectrum, and texture, topological characteristics of the extraction, the classifier for the object image semantic information. Generally speaking, object-based image processing technology includes images of regional segmentation and remote sensing image classification two core steps. Regional segmentation is a remote sensing image of object-based analysis foundation. Regional segmentation used the local spectrum, and texture image characteristics of the homogeneity, segment image into each other is not part of the overlap of the process. Remote sensing image segmentation precision and efficiency of directly affect the accuracy of the subsequent interpretation. By images of semantic objects under segmentation causes the mixed problem, not through the subsequent classification method to improve. On the other hand, compared with the natural scene image, remote sensing data with mass sex, including more complex space structure information and multi-level features semantic information. This led to the existing originated in industrial image processing of computer vision theory and method and can’t completely suitable for remote sensing image processing, therefore, the suitable for high resolution of remote sensing image, high accuracy and efficiency of the regional segmentation method for high resolution of remote sensing image of interpretation is of great significance.The mean shift (MS) as a kind of drift algorithm parameters of the probability density gradient estimation algorithms, have a good theoretical basis, and used in the feature space analysis of the occasion. The algorithm is the characteristics of concurrent operation, which is the sea of remote sensing data processing. When the space location and spectral features constitute the united in the feature space, the algorithm is equivalent to a keep the edge of the image filtering process. Therefore, this paper for the mean shift algorithm tool, high spatial resolution of the image segmentation developed. The work can be summarized as the following:1、Induction and choose the three typical edge information detection operator. The image edge information consists of Oriented Energy, Brightness Gradients, Color Gradients and Texture Gradients. The programming three operators on high resolution of remote sensing image edge detection.2、Using manual criticize high resolution of remote sensing image object boundary the basis for training, we introduce the linear weighted model, Logistic regression model and support vector regression model established three edge cues and the return of real semantic objects border relations.3、Using three regression models for prediction of the result, by the weights of the form to mean shift integration iterative process, get better segmentation results. And the two groups of experimental image, through the visual interpreting and four quantitative indicators to comprehensive evaluation result. Image segmentation results show that:this paper proposed more than the edge of character information through the mathematical modeling of integration regression analysis, can effectively restrain the remote sensing image segmentation appears in the over-segmentation and the weak boundary or texture of the boundary of the object features under segmentation phenomenon also have improved greatly.