System and method for global motion estimation using profile matching
View Patent ↗In a global motion estimation, profiles (Prof 1 , Prof 2 , ProO, Prof 4 ) are determined of frames (FR 1 , FR 2 , FR 3 , FR 4 ) of an input video signal. Each profile comprises a value, for each pixel in a first direction of the frame, which is obtained by combining pixels in a second direction of the frame. The profiles are updated (Upd Prof) with a first shift (Δx′ 12 , Δx′ 2 , Δx′ 34 ) to obtain updated profiles (Prof 1 ′, Prof 2 ′, Pro 3 ′), and the updated profiles are matched (S/M) to obtain a secondary shift (Δx′ 12 , Δx′ 23 ). Preferably, the first shift is a zero shift. The updating may be effected by shifting one of the profiles over the first shift to obtain a shifted profile, and subtracting another one of the profiles from the shifted profile. If the second direction is the vertical direction, the profile may comprise a vertically combined (e.g. averaged) pixel value for each pixel in the horizontal direction.
1. A method of global motion estimation, comprising the steps of:
determining profiles of frames of an input video signal, wherein each profile comprises a value, for each pixel in a first direction of the frame, which is obtained by combining pixels in a second direction of the frame;
comparing profiles of frames by shifting a profile of one frame with respect to a profile of another frame to determine at least one first shift value where a minimum in a difference between profiles occurs;
updating the profiles by shifting one of the profiles by using one of the determined first shift values representing a pixel shift in the first direction and subtracting another one of the profiles from the shifted profile to obtain updated profiles; and
comparing updated profiles to obtain at least one second shift value where a minimum in a difference between updated profiles occurs,
wherein a plurality of possible first shift values are determined and a confidence metric is used for each of the possible first shift values to select the first shift value that is used in the step of updating the profiles.
2. A method according to claim 1 further comprising calculating a confidence metric for a shift value obtained when comparing profiles of frames.
3. A method according to claim 2 wherein the confidence metric is based on at least one of:
the sharpness of the response at the position of the shift value;
other minimum values.
4. A method according to claim 2 further comprising using the confidence metric to quantify the quality of the shift value.
5. A method as claimed in claim 1 , further comprising the steps of:
updating the profiles with the second shift value to obtain second updated profiles, and
comparing the second updated profiles to obtain an improved estimate of the first shift value.
6. A method as claimed in claim 1 , further comprising iteratively performing the comparing and updating steps to identify other shift values.
7. A method as claimed in claim 1 , wherein the first shift value or the second shift value is a zero shift.
8. A method as claimed in claim 1 wherein the second direction is the vertical direction, and the profile comprises a vertically combined pixel value for each pixel in the horizontal direction.
9. A method as claimed in claim 1 , further comprising repeating the method by determining a second set of profiles of frames of an input video signal wherein each profile comprises a value, for each pixel in the second direction of the frame, which is obtained by combining pixels in the first direction of the frame and performing the steps of comparing the profiles and updating the profiles using the second set of profiles.