ALBERT

Description: A method of stabilizing and registering a video image in multiple video fields of a video sequence provides accurate determination of the image change in magnification, rotation and translation between video fields, so that the video fields may be accurately corrected for these changes in the image in the video sequence. In a described embodiment, a key area of a key video field is selected which contains an image which it is desired to stabilize in a video sequence. The key area is subdivided into nested pixel blocks and the translation of each of the pixel blocks from the key video field to a new video field is determined as a precursor to determining change in magnification, rotation and translation of the image from the key video field to the new video field.

Description: A method of stabilizing and registering a video image in multiple video fields of a video sequence provides accurate determination of the image change in magnification, rotation and translation between video fields, so that the video fields may be accurately corrected for these changes in the image in the video sequence. In a described embodiment, a key area of a key video field is selected which contains an image which it is desired to stabilize in a video sequence. The key area is subdivided into nested pixel blocks and the translation of each of the pixel blocks from the key video field to a new video field is determined as a precursor to determining change in magnification, rotation and translation of the image from the key video field to the new video field.

Description: The Video Image Stabilization And Registration (VISAR) process is an award winning video image processing software developed at NASA's Marshall Space Flight Center. VISAR has a wide variety of application areas where the refinement of digital video is needed. It is used to correct jitter, rotation, and zoom effects by registering and processing on individual image captures that are a part of normal video capturing. Its most prominent uses were the 1996 Olympic Bombing case and in identifying Saddam Hussein during the Iraq war. Based on first-hand knowledge, this paper describes the VISAR process, which consists of several steps designed to refine digital video using VISAR software. The process determines the differences between two video images so that one, or both, of the images can be changed in ways that make them match as well as possible. Corrections include changes in position (horizontal and vertical image shifts), changes in orientation (image rotation), and changes in magnification (image zoom). While much of the VISAR process is automated, in its current embodiment it requires the user to initially identify the area of interest and to reset a threshold parameter if the default gives unacceptable results. The basic process that is used is an old tried and true method that determines how well the two images match. This process is called cross-correlation. It gives a single number, the correlation coefficient, that is equal to 1.0 if the images are perfectly matched, is equal to 0.0 if the images have nothing in common, and is equal to -1.0 if one image is the negative of the other. This basic process is used by many image stabilization methods. With VISAR we use it in a manner that provides statistical information needed to best determine orientation and magnification.