3:2 pull down detection in video
Methods for processing of video sequences that may contain telecined (3:2 pull down) frame sequences are provided. A method for detecting 3:2 pull down is provided that measures vertical detail in frames of a video sequence and uses the variation in vertical detail over time to decide whether the video sequence contains normal interlace content or 3:2 pull down content. A method for improving the compression of detected 3:2 pull down content is also provided that controls the selection of field or frame coding mode for frames of 3:2 pull down content and the selection of reference fields for encoding duplicated fields in the 3:2 pull down content.
1. A method for processing a video sequence, the method comprising:
computing frame vertical detail and field vertical detail for a frame in the video sequence, wherein frame vertical detail is indicative of correlation between adjacent rows of pixel values in the frame and field vertical detail is indicative of correlation between adjacent rows of pixel values in fields of the frame, wherein frame vertical detail is computed as a sum of pixel differences between adjacent rows of pixels in a frame and field vertical detail is computed as an absolute sum of pixel differences between alternate rows of pixels in the frame; and
indicating 3:2 vertical pull down in the video sequence based on the frame vertical detail and the field vertical detail.
2. The method of claim 1 , wherein indicating 3:2 vertical pull down comprises:
computing an average of field vertical detail for a plurality of frames, wherein the plurality of frames comprises the frame; and
indicating 3:2 vertical pull down in the video sequence when frame vertical detail for any frame in the plurality of frames is sufficiently larger than the average of field vertical detail.
3. The method of claim 1 , wherein frame vertical detail is computed as a sum of pixel differences between adjacent rows of pixels in selected blocks of a frame and field vertical detail is computed as an absolute sum of pixel differences between alternate rows of pixels in the selected blocks.
4. The method of claim 3 , wherein the adjacent rows of pixels and the alternate rows of pixels are a subset of rows of pixels in the selected blocks.
5. The method of claim 1 , further comprising:
selecting a coding mode for the frame responsive to the indication of 3:2 vertical pull down, wherein the coding mode is one selected from a field coding mode and a frame coding mode.
6. The method of claim 5 , wherein selecting a coding mode comprises:
selecting the field coding mode when a top field of the frame and a bottom field of the frame were captured at non-consecutive time intervals;
selecting the field coding mode when the top field and the bottom field were captured at consecutive time intervals and a field in the frame is a duplicated field; and
using a picture adaptive frame-field (PAFF) algorithm to select the coding mode when the top field and the bottom field were captured at consecutive time intervals and neither field in the frame is a duplicated field.
7. The method of claim 5 , wherein selecting a coding mode comprises:
selecting the field coding mode when the frame vertical detail of the frame is sufficiently larger than an average of frame vertical detail computed for a plurality of frames comprising the frame; and
using a picture adaptive frame-field (PAFF) algorithm to select the coding mode when the frame vertical detail of the frame is not sufficiently larger than the average of frame vertical detail.
8. The method of claim 6 , further comprising:
indicating, when a field in the frame is a duplicated field, that a previously encoded version of the field should be used as a reference field when coding the duplicated field.
9. The method of claim 5 , further comprising:
determining that a top field of the frame is a duplicated field; and
indicating that a previously encoded version of the top field should be used as a reference field for the top field.
10. The method of claim 5 , further comprising:
determining that a bottom field of the frame is a duplicated field; and
indicating that a previously encoded version of the bottom field should be used as a reference field for the bottom field.
11. A method for processing a video sequence, the method comprising:
detecting that a frame is part of a 3:2 vertical pull down sequence in the video sequence, wherein frame vertical detail is computed as a sum of pixel differences between adjacent rows of pixels in a frame and field vertical detail is computed as an absolute sum of pixel differences between alternate rows of pixels in the frame; and
selecting a coding mode for the frame responsive to the detecting, wherein the coding mode is one selected from a field coding mode and a frame coding mode.
12. The method of claim 11 , wherein selecting a coding mode comprises:
selecting the field coding mode when a top field of the frame and a bottom field of the frame were captured at non-consecutive time intervals;
selecting the field coding mode when the top field and the bottom field were captured at consecutive time intervals and a field in the frame is a duplicated field; and
using a picture adaptive frame-field (PAFF) algorithm to select the coding mode when the top field and the bottom field were captured at consecutive time intervals and neither field in the frame is a duplicated field.
13. The method of claim 12 , further comprising:
indicating, when a field in the frame is a duplicated field, that a previously encoded version of the field should be used as a reference field when coding the duplicated field.
14. The method of claim 11 , further comprising:
determining that a top field of the frame is a duplicated field; and
indicating that a previously encoded version of the top field should be used as a reference field for the top field.
15. The method of claim 11 , further comprising:
determining that a bottom field of the frame is a duplicated field; and
indicating that a previously encoded version of the bottom field should be used as a reference field for the bottom field.
16. The method of claim 11 , wherein detecting that a frame is part of a 3:2 vertical pull down sequence comprises:
computing frame vertical detail and field vertical detail for a frame in the video sequence, wherein frame vertical detail is indicative of correlation between adjacent rows of pixel values in the frame and field vertical detail is indicative of correlation between adjacent rows of pixel values in fields of the frame;
computing an average of field vertical detail for a first plurality of frames, wherein the first plurality of frames comprises the frame; and
detecting 3:2 vertical pull down when frame vertical detail for any frame in the first plurality of frames is sufficiently larger than the average of field vertical detail.
17. The method of claim 16 , wherein selecting a coding mode comprises:
selecting the field coding mode when the frame vertical detail of the frame is sufficiently larger than an average of frame vertical detail computed for a second plurality of frames comprising the frame; and
using a picture adaptive frame-field (PAFF) algorithm to select the coding mode when the frame vertical detail of the frame is not sufficiently larger than the average of frame vertical detail.