Detection of progressive pulldown in a video sequence
AssigneeSamsung Electronics Co., Ltd.
InventorsZhi Zhou, Yeong-Taeg Kim
Life of this patent
Estimating the term…
Abstract
System and methods for the detection of progressive pulldown in a video sequence are provided. The method comprises calculating a difference between temporally adjacent frames in the video sequence, generating a pattern based upon these differences, and comparing the generated pattern against patterns understood for known progressive film modes in order to determine if the video sequence comprises a known progressive film mode. In one embodiment, the pattern may be adjusted to account for uncertainty in assigning values to the pattern, reducing the likelihood of false detections. In another advantage, confidence metrics are built into the method in order to further reduce the incidence of false detections.
Claims20
A method of detecting a film mode of a video sequence, comprising:
- computing a difference, d(t), between a selected frame, F(t), and a temporally adjacent frame, F(t−1), where t is time, over a selected range of times;
- calculating a difference threshold, T(t), for each d(t) as the average of the maximum and minimum values of d(t−1), d(t), and d(t+1) over the selected range of times;
- generating a pattern by assigning a pattern value, e(t), to each d(t), wherein a first pattern value is generated when d(t) is greater than T(t), a second pattern value different than the first is generated when d(t) is less than T(t), and e(t) is adjusted to a third pattern value if either:
- the frame F(t−1) has been determined to be part of a known progressive film mode and the generated pattern values, e(t), for any t within the selected range differ from an expected pattern value, x(t), determined from the known progressive film mode;
- or
- the standard deviation and mean of d(t), over the selected time range, are less than a threshold standard deviation, Ts, and a threshold mean, Tm, respectively;
- and
- determining the selected frames, F(t), over the selected range of times, to be either a part of a known progressive film mode or video based upon a comparison of the generated pattern to patterns of a plurality of known progressive film modes.
A method of detecting a film mode of a plurality of frames of a video sequence, comprising:
- computing a difference, d(t), between a selected frame, F(t), and a temporally adjacent frame, F(t−1), where t is time, over a selected range of times;
- generating a pattern value, e(t), for each frame, comprising a first pattern value and a second pattern value, different than the first, using the measured differences;
- and
- adjusting one or more of the pattern values to a third value, different than the first and second pattern values, if either:
- the frame F(t−1) has been determined to be part of a known progressive film mode and the generated pattern values, e(t), for any t within the selected range differ from an expected pattern value, x(t), determined from the known progressive film mode;
- or
- the standard deviation and mean of d(t), over the selected time range, are less than a threshold standard deviation, Ts, and a threshold mean, Tm, respectively;
- wherein, if F(t−1) is not detected to be part of a known progressive film mode, the film mode of F(t) is determined to be a progressive film mode if each pattern value in the generated pattern follows the pattern of the known progressive film mode and wherein if F(t−1) is detected to be part of a known progressive film mode, the film mode of F(t) is determined to be the known progressive film mode if each e(t) in the generated pattern not equal to the third pattern value follows the pattern of the known progressive film mode.
An apparatus for detecting progressive video sequences from a video sequence of unknown film mode, comprising:
- a first component which computes a difference, d(t), between a selected frame, F(t), and a temporally adjacent frame, F(t−1), where t is time, over a selected range of times;
- a second component, coupled to the first component, which calculates a difference threshold, T(t), for each d(t) as the average of the maximum and minimum values of d(t−1), d(t), and d(t+1) and generates a pattern value, e(t), for each d(t), wherein a first pattern value is generated when d(t) is greater than T(t) and a second pattern value different than the first is generated when d(t) is less than T(t) and e(t) is adjusted to a third pattern value if either:
- the frame F(t−1) has been determined to be part of a known film mode and the generated pattern values, e(t), for any t within the selected range differ from an expected pattern value, x(t), determined from the known film mode;
- or
- the standard deviation and mean of d(t), over the selected time range, are less than a threshold standard deviation, Ts, and a threshold mean, Tm, respectively;
- and
- a third component, coupled to the first component, which determines that the selected frames, F(t), over the selected range of times, to be either part of a selected progressive film mode or video based upon a comparison of the generated pattern to patterns of a plurality of known progressive film modes.
Dependencies are inferred from the claim text.