Method and apparatus for parametric, model-based, geometric frame partitioning for video coding
There are provided methods and apparatus for adaptive geometric partitioning for video encoding and decoding. An apparatus includes an encoder for encoding image data corresponding to pictures by adaptively partitioning at least portions of the pictures responsive to at least one parametric model. The at least one parametric model involves at least one of implicit and explicit formulation of at least one curve.
1. An apparatus, comprising:
at least one processor being configured to:
obtain values for pixels of a current block of a picture according to a parametric model;
classify said pixels of said current block into two partitions according to said obtained values, wherein pixels of the two partitions overlap in a partial surface, and wherein a first value obtained according to said parametric model indicates that a pixel of said current block is in a first partition of said two partitions, a second value indicates that a pixel of said current block is in a second partition of said two partitions, and a value in between said first and second values indicates that a pixel of said current block is in said partial surface;
predict said current block to form a predicted block by:
predicting pixels in said first partition using a first predictor,
predicting pixels in said second partition using a second predictor, and
predicting pixels in said partial surface based on a weighted linear average of said first predictor and said second predictor, wherein said first predictor is weighted by
w(x, y) and said second predictor is weighted by (1-w(x, y)) to form said weighted linear average, wherein w(x, y) is said obtained value from said parametric model for pixel at location (x, y); and
decode said current block responsive to said predicted block.
2. The apparatus of claim 1 , wherein a curve is formed based on said parametric model, wherein said first partition is on one side of said curve, said second partition is on the other side of said curve, and said partial surface is on a neighborhood of said curve.
3. The apparatus of claim 1 , wherein said first predictor is predicted from at least one of spatial neighboring blocks and temporal neighboring blocks.
4. The apparatus of claim 1 , wherein said first partition is non-square and non-rectangular.
5. The apparatus of claim 1 , wherein said obtained value w(x, y) for a pixel adapts to location (x, y) of said pixel.
6. The apparatus of claim 5 , wherein said obtained value w(x, y) is based on a first degree polynomial function.
7. The apparatus of claim 1 , wherein said first value is 1 and said second value is 0.
8. The apparatus of claim 1 , wherein said current block is partitioned based on a partition mask, wherein an element at location (x, y) in said partition mask corresponds to said obtained value w(x, y).
9. The apparatus of claim 1 , wherein said at least one processor is further configured to decode, for said current block, at least a parameter indicating a partition shape for said current block.
10. The apparatus of claim 1 , wherein said parametric model is adaptively decoded depending on a set of models when more than one parametric model is available.
11. A method, comprising:
obtaining values for pixels of a current block of a picture according to a parametric model;
classifying said pixels of said current block into two partitions according to said obtained values, wherein pixels of the two partitions overlap in a partial surface, and wherein a first value obtained according to said parametric model indicates that a pixel of said current block is in a first partition of said two partitions, a second value indicates that a pixel of said current block is in a second partition of said two partitions, and a value in between said first and second values indicates that a pixel of said current block is in said partial surface;
predicting said current block to form a predicted block by:
predicting pixels in said first partition using a first predictor,
predicting pixels in said second partition using a second predictor, and
predicting pixels in said partial surface based on a weighted linear average of said first predictor and said second predictor, wherein said first predictor is weighted by
w(x, y) and said second predictor is weighted by (1-w(x, y)) to form said weighted linear average, wherein w(x, y) is said obtained value from said parametric model for pixel at location (x, y); and
decoding said current block responsive to said predicted block.
12. The method of claim 11 , wherein a curve is formed based on said parametric model, wherein said first partition is on one side of said curve, said second partition is on the other side of said curve, and said partial surface is on a neighborhood of said curve.
13. The method of claim 11 , wherein said first predictor is predicted from at least one of spatial neighboring blocks and temporal neighboring blocks.
14. The method of claim 11 , wherein said first partition is non-square and non-rectangular.
15. The method of claim 11 , wherein said obtained value w(x, y) for a pixel adapts to location (x, y) of said pixel.
16. The method of claim 15 , wherein said obtained value w(x, y) is based on a first degree polynomial function.
17. The method of claim 11 , wherein said first value is 1 and said second value is 0.
18. The method of claim 11 , wherein said current block is partitioned based on a partition mask, wherein an element at location (x, y) in said partition mask corresponds to said obtained value w(x, y).
19. The method of claim 11 , further comprising decoding, for said current block, at least a parameter indicating a partition shape for said current block.
20. The method of claim 11 , wherein said parametric model is adaptively decoded depending on a set of models when more than one parametric model is available.