An Improvement of Geometric Correlation of Satellite Image
Sompong Wisetphanichkij, Kobchai Dejhan, Fusak Cheevasuvit,
Somsak Mitatha and Somkid Hanpipatpongsa
Faculty of Engineering and Research Center for Communication and Information Technique,
King Mongkut's Institute of Technology Ladkrabang, Ladkrabang,
Bangkok 10520, Thailand
Tel-: 66-2-3269967, 66-2-3269081, Fax : 66-2-3269086.
E-mail: kobchai@telean.telecom.eng.kmitl.ac.th
Chanchai Pienvijarnpong
National Research Council of Thailand,
MOS-I Receiving Ground Station,
Ladkrabang, Bangkok 10520, Thailand
Chatcharin Soonyeekan
Faculty of Engineering, Kasem Bundit University
Pattankaran Road, Klongton, Bangkok 10250, Thailand
Abstract
The geometric distortion of the satellite image can be occurred according to many reasons. It can be divided into systematic and non-systematic (or random) distortion. It is necessary to correct before using. The precision of geometric correction depends on the ground control points (GCPs). The noise is the satellite image may be generated from the interface or the blur according to the cloud coverage, thus it is difficult to assign the positions of GCPs. This paper presents a method to assign automatically the GCPs by using image registration method based on a sequential similarly detection algorithms (SSDAs) technique (statistic correlation measurement). This method used the sub-image around GCP from the improved image acting as prototype image to assign the positions of GCPs for the improving satellite image, the correlation will also be used in order to obtain the precision and accuracy.
1. Introduction
The geometric distortion in remotely sensed images come from two sources: the internal distortion caused by the sensor and the external distortion caused by the platform and the subjects
- Internal distortion (Systematic distortion)
Variations of beam width and sampling in the sensor (or radar).
- External distortion (Non systematic distortion)
Variations of the location, altitude, attitude and speed of the platform,
the relief and curvature of the ground surface and the earth's revolution.
The geometric correction process geometrically converts the image coordinates from (x,y) into (x,y), where corrected coordinates without geometrical distortion are expressed by (x,y) and input image coordinates given by (x,y). Two types of distortion, geometric distortion correction have divided with into systematic and precision geometric correction to correct the internal and external distortion, respectively. Systematic correction is a radiometrically and geometrically corrected without using ground truth and usually performed before distributing.
- Image to map (grid) registration is used the topographic map as a reference
- Image to image registration that one image referred to as the reference image and assumed to be good quality, i.e., no cloud, contrast is good and geometrical distortion are negligible. The search (input) image, on the other hand, is relatively unknown quality. Some coverage fog/cloud may be present along the geometrical distortion.
The importance issue to achieve such process is the way to choose the ground control point (GCP), which should be evidence on both images or topographic map, time invariance and spread through the image.
Image to image registration is a procedure to determine the spatial best fit between two images that overlap the same scene. In the sake of two images for the same scene cannot be meaningfully compared. Several digital techniques have been used for registering imagery Principal among these are cross-correlation, normalized cross-correlation (Eq.1) and minimum distance criteria. Fast algorithms for determining translational difference between images have been developed in recent years for all techniques [1].
Where (j,k) are indices in a JxK point window area (W) that is located within an MxN point of search area (S)
Fig. 1 Relationship between search and window areas
In this paper, an applied technique for registering a pair of function is a form of the statistical correlation measurement [2] as in equation (2).
Where the g
i(j,k) are obtained by spatially convolving the sampled images f(j,k) with spatial filter function Di(j,k). Thus
gi(j,k) = fi(j,k) Di (j,k)
The spatial filter function is chosen to maximize the correlation peak ratio and the registration filter, D
i(j,k)[3].
