vgreg − register a sequence of images based on the entire image

vgreg if=infile of=outfile [n=numpix] [a=alpha] [s=] [-sqdif | -sumdif | -norm | -cross] [-f] [-p] [m=] [-v] [h=]

Vgreg Vgreg is a image registration program. It takes in a sequence of images. It then determines how to best move fit the next frame on a history frame. This is done by trying to move the current frame by [n=] in all directions. Default n is 20. As it is moving the frame it finds the best correlation by using [-sqdif | -sumdif | -norm | -cross]. Default is -sqdif. When the best correlation is at +/- [n=] in any direction it searches [s=]*[n=] in that direction. Default is is 2. A history file [h=] can be specified to see the movements of the image. The history frame in vgreg can be specified two different ways. [-p] uses the previous frame as the history frame. The current frame is then only checked against the pixels that were valid in the previous frame. [-f] uses the first frame as the history frame. This history frame never changes and t here is no mask. An alpha value may also be specified [a=]. This value averages the history frame with the valid pixels in the last frame. a=0 is the same as the [-f], but a=0 is not the same as [-p] since the border pixels are kept from the history frame. [h=] may be specified to output the sequence of history frames, but all history frames will be of type float.

The input can be of type unsigned char, float char, double, char, short, or int. Note double may lose precision. Cannot currently handle multi channel images.

if=

Specifies the input file which must be a VisionX image set

of=

Specifies the name of the annotated image set output file.

n=

Number of pixels to check in all directions (up, down, left, right) to find the best correlation. (n=18 is default)

a=

alpha: Use a weighted average find the next history frame. It is weighted by HF = a(CF) + (1-a)HF where CF is the current frame and HF is the last history frame

s=

How much farther to search if the best fit for a frame is at a search boundry (+/- n in any direction). It will search s*n farther in that direction. (s=2 is default)

The correlation matching algorithm is specified by one of the following :

-sqdif

sum of squared differences (equivalent to mse) (default)

-sumdif

sum of absolute differences

-norm

normalized correlation

-cross

cross correlation

-f

history frame is always the first frame (a=0) (default)

-p

history frame is always the previous frame with a mask using only the the valid pixels.

m=

specifies a file to show the vector movements of the registered image

h=

specifies a history output file to output every history frame to. This image is of type float.

A.P. Reeves, M. Udoff, J. Lee