Research On Image Characteristic Scale And Its Application In Scene Matching

Scale is an intrinsic property of images. The study of images and its structures needs to be in a particular scale. It is very important for the study of scale. The better understanding of scale can contribute to a better analysis of images.In this study, satellite images with different resolutions in different areas are used in experiments to analyze the characteristic scale of images. This study suggests that there is relationship between images' texture structure and its scale. When the texture structure is in a scale, in which it has a maximum response to some function, then the scale is defined as the characteristic scale. Accordingly, we establish the scale space representation of original image, then use differential function to get the response in pixel neighborhood along the direction of scale, record the maximum value of the response curve for the pixel and characteristic scale is the scale with which the maximum value is related.On this basis, our study counts the number of pixels corresponding to each characteristic scale in one image area to establish the histogram of characteristic scale. We define the characteristic scale as the scale which has maximum value in the histogram of characteristic scale. Finally, the study shows the application of characteristic scale in scene matching.One of the applications is the characteristic scale of the image area can select a template size for scene matching quantitatively. The study gives a relationship between characteristic scale of the image area and the size of the template for scene matching. By using this quantitative standard, we can ensure the matching performance, while reducing the size of the image template, which can improve the real-time performance of scene matching.Another application of the image area is to analyze the degree of similarity between different images. We believe that the shape of the histogram of characteristic scale contains the important information about the structure and texture of the images. The shape of histogram of characteristic scale can be obtained by polynomial fitting. After polynomial fitting, we get a curve which we define as curve of characteristic scale. The curve can be used to describe the degree of similarity and matching performance between images.This paper studied the characteristic scale, which is the basic property of the image. The normalized characteristic scale is independent of the image resolution, reflecting the structural properties of the image. Our study has a theoretical and practical value indeed.