Resolution Improvement of Frequently Observed Satellite Imagery for Urban Characterization

ABSTRACT: This paper proposes the method to develop a high-resolution image of urban properties from frequently observed low-resolution images based on the statistic process assuming urban properties have little temporal change. Because disaster responders sometimes have difficulty to obtain the map data in a digital form especially in developing countries, it is well known that satellite imagery and aerial photographs can support disaster response and rescue activities. But high-resolution images that can provide roads and buildings information expense considerably high when acquiring wide range data by means of aerial photographs or imagery of recently launched satellites with high-resolution sensors. On the other hand, satellite images of spatial resolution of 10 to 100 meter order cost comparatively low. The paper discusses the number of low-resolution images to develop a high-resolution image. First, a one-meter-resolution image derived from an aerial photograph is used as a truth image. Three-meter-resolution image is also evaluated as a reference high-resolution image. Secondly, a nine-meter-resolution image, which simulates an image scanned by a low-resolution sensor of satellites, is evaluated with a randomly selected offset of scan line. The same cell size as a reference image, i.e. three meter is employed for the low-resolution images. Thirdly, a resolution-improved image is synthesized from the average and standard deviation based on each pixel calculation of the simulated images. Then the difference between the reference image and synthesized image is evaluated. Finally, the relationship between the number of low-resolution images and the standard deviation of differences of all pixels in the average image is analyzed. The result quantitatively clarifies that a high-resolution image of urban properties with little temporal change can be developed with a sufficient number of low-resolution images when the precise geolocation is guaranteed.