In this section, we will introduce our feature based method for estimating the coordinate transformation parameter. When we find some matched pairs then we try to find transformation parameter by using the coordinate of the matched image points.
Consider two camera view, one denoted as the reference camera and other
is inspected camera. A 3D P point in the reference camera coordinate
system gets mapped to the 3D point 
in the inspection camera
coordinate system by a rigid body transformation:
The mapping can be represented by a rotation followed by a
translation. Two aligned reference image and inspected image, we try to
motion between reference and inspected images.
The 2D motion filed of a 3D planar surface can be represented by using
8-parameter pseudo perspective model.
Note for discrete views with significant camera rotation, this 8 parameter projective transformation is required to align a planar surface.
We use the Normalized Sum of Square Difference (NSSD) measure to measure how good our parameters is to map the reference image to inspected image.
where 
denotes the spatial image position of a point in images.
I and I' is the color information of a pixel. 
is the
motion field between these images.
The E term is normalize by using number of pixel that are in the
overlapped regions.
We use least-square approach to find the transformation parameters where normalized error is minimum. Also we tried another approach to find the good transformation parameters. In that method, we have n number of matched pairs between two images. And 4 pair is enough to find a solutions. Therefore, we choose some m number of pair set which has 4 different matched pair, then we use this m solution to find the minimum error. Also this method gives as good results as least square approach.