Method of maintaining eye contact in video conferencing using view morphing
View Patent ↗A view morphing algorithm is applied to synchronous collections of video images from at least two video imaging devices, and interpolating between the images, creates a composite image view of the local participant. This composite image approximates what might be seen from a point between the video imaging devices, presenting the image to other video session participants.
1. A method of providing eye contact with a first video conferencing participant comprising the steps of:
receiving at least two source images of said first participant;
generating a composite image of said first participant; and
providing said composite image to a second video conferencing participant;
wherein each of said source images is comprised of an array of source image pixels; and
wherein said step of generating said composite image comprises the steps of:
calculating a dense correspondence for at least one pair of images among said source images, said dense correspondence specifying a pixel displacement for each of said source image pixels within each image within said pair of images;
generating at least two interpolated images from said source images based upon said pixel displacement; and
combining at least two of said interpolated images using an averaging scheme to create said composite image.
2. The method claim 1 , said step of generating at least two interpolated images comprising displacing each of said source image pixels within at least two of said source images by a fraction of said pixel displacement.
3. The method of claim 2 , wherein said fraction is one half.
4. The method of claim 2 , wherein said source images are acquired from at least two imaging devices, and said fraction is based upon the positions of said image devices relative to a location of said second participant on a display within view of said first participant.
5. The method of claim 1 , wherein said composite image comprises an array of composite image pixels;
wherein each of said interpolated images comprises an array of interpolated image pixels; and
wherein said averaging scheme comprises calculating values for aid composite image pixels based on a weighted average of corresponding pixels among said interpolated image pixels.
6. The method of claim 5 , wherein said weighted average is based upon the location of said corresponding pixels within said interpolated images.
7. The method of claim 1 , said step of generating a composite image additionally comprising the step of, before said calculating a dense correspondence step:
applying a rectifying transformation to at least one of said source images.
8. The method of claim 1 , wherein said source images are acquired from a time series of source images, and said step of generating a composite image additionally comprising the step of, before said calculating a dense correspondence step: applying a warping transformation to at least one of said source images to displace each of said source image pixels by a fraction of a prior pixel displacement determined from a prior dense correspondence computed for source images prior to said source images within said time series of source images.
9. The method of claim 8 , wherein said fraction is less than one.
10. The method of claim 1 , wherein said pixels are characterized by color values within a perceptually uniform color space.
11. The method of claim 10 , wherein said color space is L*a*b* CIE.