Image analysis – Applications – 3-d or stereo imaging analysis
Reexamination Certificate
2001-02-12
2004-06-15
Bali, Vikkram (Department: 2623)
Image analysis
Applications
3-d or stereo imaging analysis
Reexamination Certificate
active
06751345
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to computer-vision technology. More specifically, this invention relates to improving the determination of object boundaries perceived within an image.
BACKGROUND OF THE INVENTION
Computer vision technology is undergoing considerable development to enable a computer to perceive the images of a scene. The ability of computers to recognize and perceive images of a scene or image is important so as to provide an additional means of inputting data into a computer. Hence, a computer may be presented a scene and the images or objects in the scene may be recorded for later enhancement, filtering, coloring, etc. One aspect of computer vision technology is the determination of depth of objects in an image. The determination of depth of objects aids in determining the boundaries of each object. One technique to achieve depth perception is using a stereoscopic technique, i.e., using two cameras to record the image. The concurrently viewed images are then processed to determine the relative positions of the objects in the image. Typically, the determination of matching objects between the images is performed by comparing a group of pixels of known size, e.g., a matrix 9×9 pixels, in one image with a similar group of pixels in a second image. In those cases, when image objects are sufficiently spaced apart, i.e., greater than the matching resolution criteria, the object boundaries are typically clearly denoted. As the objects are positioned closer together or a lack of textural difference between objects exists, the determination of the object boundaries becomes more difficult. Methods of improving the determination of object boundaries are known in the art. For example, methods using point-to-point matching, as disclosed in “
Depth Discontinuities by Pixel-to-Pixel Stereo
,” S. Birchfed, C. Tomasi, International Journal of Computer Vision 35(3), p. 1-25, 1999, referred to herein as the
Stanford algorithm
, produces sharp boundaries. However, these methods occasionally fail to determine the boundary position correctly and produce boundaries that are shifted significantly. Hence, there in a need to improve object boundary determination, without increasing the time of processing, in areas where a lack of texture causes blurring of object boundaries.
SUMMARY OF THE INVENTION
A method for refining object boundaries in stereoscopic images of a same scene by reprocessing determined pixel image data to refine boundary images. That is, using convenient processing techniques, two stereoscopic images are processed into a single resultant image, which includes matching corresponding pixels between the two images and determining a measure of disparity difference in position of the same object in the two images. The method then reprocesses the resultant image, by selecting each pixel in the resultant image, one pixel at a time, and determining a re-match such that a corresponding pixel in a first image is re-matched to an different pixel in the second image. A second dissimilarity measure for the re-matched pixel combination is then determined and the re-matched pixel combination is retained while the second dissimilarity measure is not substantially different from the first measure. A second dissimilarity measure is not substantially different from the first measure when the second measure is within known bounds about a determined value.
REFERENCES:
patent: 5202928 (1993-04-01), Tomita et al.
patent: 5530774 (1996-06-01), Fogel
patent: 5917937 (1999-06-01), Szeliski et al.
patent: 6046763 (2000-04-01), Roy
patent: 6075557 (2000-06-01), Holliman et al.
patent: 6701005 (2004-03-01), Nichani
patent: 195 09 373 (1995-03-01), None
Motion segmentation: A synergistic approach, by Fermuller et al. IEEE 1999, pp. 226-231.*
Birchfield et al.; Depth Discontinuities by Pixel-to-Pixel Stereo; Proceedings of the 1998 IEEE International Conference on Computer Vision; Bombay, India.
Bali Vikkram
Koninklijke Philips Electronics , N.V.
Thorne Gregory L.
LandOfFree
Method and apparatus for improving object boundaries... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for improving object boundaries..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for improving object boundaries... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3354003