Content adaptive predictive and functionally predictive pictures with modified references for next generation video coding

1. A computer-implemented method for video coding, comprising:

receiving data for a sequence of frames of pixel data, the sequence having a display order, and wherein each frame is associated with a temporal position of the display order;

generating at least one current F-picture of the sequence of frames using inter-prediction, comprising providing all three options to form (1) at least one part of the current F-picture from at least one previous prediction reference frame without using subsequent prediction reference frames, (2) at least one subsequent prediction reference frame without using previous prediction reference frames, or (3) to use both previous and subsequent prediction reference frames, wherein previous and subsequent are relative to the position of the current F-picture and within the sequence, wherein the option to select among the three choices is not available for B-pictures; and

generating motion data for the part of the at least one current F-picture comprising using at least one modified prediction reference frame for generating the motion data for the part of the current F-picture, and the at least one modified prediction reference frame being modified by using a morphing technique or a synthesizing technique or both comprising analyzing a current F-picture to determine parameters of the at least one modified prediction reference frame and to decide which of the techniques to use as part of a selected prediction mode of the current F-picture, and depending, at least in part, on the data of the current F-picture,

wherein at least one morphing technique or synthesizing technique is available for the F-picture but is not available for B-pictures.

2. The method of claim 1 comprising forming the sequence with non-F prediction reference frames used to generate the motion data for the F-pictures, and wherein the F-pictures are spaced between the non-F prediction reference frames, and wherein the number of F-pictures between the non-F prediction reference frames varies along the sequence.

3. The method of claim 1 comprising generating at least one P-picture within the sequence by providing an option to generate at least part of the P-picture from at least one of multiple previous prediction reference frames along the sequence.

4. The method of claim 1 comprising providing the option to form at least one of the F-pictures from at least one additional prediction reference frame in addition to one previous prediction reference frame and one subsequent prediction reference frame along the sequence and relative to the position of the F-picture.

5. The method of claim 1 wherein the F-picture is formed by selecting the best result of a plurality of results from a plurality of reference frames provided to at least partially form the F-picture in an inter-mode.

6. The method of claim 1 wherein the F-picture is formed by using data of a plurality of reference frames used to contribute to a result to at least partially form the F-picture in a multi-mode.

7. The method of claim 1 comprising using at least one F-picture as a prediction reference frame for another F-picture.

8. The method of claim 1 comprising forming a continuous or non-continuous chain of reference frames along the sequence so that a plurality of F-pictures are reference frames to one next subsequent F-picture.

9. The method of claim 8 wherein the chain of references is provided in addition to any other reference frames for at least one of the F-pictures.

10. The method of claim 1 comprising using a pyramid reference frame structure wherein a reference F-picture has a previous reference frame and a subsequent reference frame along the sequence and is itself a reference frame to another F-picture.

11. The method of claim 1 wherein a synthesizing technique comprises projected interpolation wherein a projected interpolation frame or frame part is formed in place of the current F-picture for prediction, and by weighting data of two reference frames depending on a temporal distance along the sequence relative to the position of the current F-picture.

12. The method of claim 1 wherein a synthesizing technique comprises modifying the resolution of a frame.

13. The method of claim 1 wherein morphing technique comprises a modified reference frame modified from another reference frame of at least one of:

(1) blur compensation,

(2) registration compensation,

(3) global motion compensation

(4) dominant motion compensation,

(5) luminance gain compensation, and

all of (1) to (5).

14. The method of claim 1 wherein the modified prediction reference frame is used for prediction but is not displayed.

15. The method of claim 1 , comprising:

forming the sequence with non-F prediction reference frames used to generate the motion data for the F-pictures, and wherein the F-pictures are spaced between the non-F prediction reference frames, and wherein the number of F-pictures between the non-F prediction reference frames varies along the sequence;

generating at least one P-picture within the sequence by providing an option to generate at least part of the P-picture from at least one of multiple previous prediction reference frames along the sequence;

providing the option to form at least one of the F-pictures from at least one additional prediction reference frame in addition to one previous prediction reference frame and one subsequent prediction reference frame along the sequence and relative to the position of the F-picture;

wherein at least one of:

the F-picture is formed by selecting the best result of a plurality of results from a plurality of reference frames provided to at least partially form the F-picture in an inter-mode, and

the F-picture is formed by using data of a plurality of reference frames used to contribute to a result to at least partially form the F-picture in a multi-mode;

comprising at least one of:

using at least one F-picture as a prediction reference frame for another F-picture;

forming a continuous or non-continuous chain of reference frames along the sequence so that a plurality of F-pictures are reference frames to one next subsequent F-picture; wherein the chain of references is provided in addition to any other reference frames for at least one of the F-pictures;

using a pyramid reference frame structure wherein a reference F-picture has a previous reference frame and a subsequent reference frame along the sequence and is itself a reference frame to another F-picture;

wherein at least one of:

a synthesizing technique comprises projected interpolation wherein a projected interpolation frame or frame part is formed in place of the current F-picture for prediction, and by weighting data of two reference frames depending on a temporal distance along the sequence relative to the position of the current F-picture;

a synthesizing technique comprises modifying the resolution of a frame;

wherein a morphing technique comprises a modified reference frame modified from another reference frame of at least one of:

(1) blur compensation,

(2) registration compensation,

(3) global motion compensation,

(4) dominant motion compensation,

(5) luminance gain compensation, and

all of (1) to (5); and

wherein the modified prediction reference frame is used for prediction but is not displayed.

16. The computer-implemented method of claim 1 , comprising:

generating at least one current P-picture of the sequence of frames using inter-prediction, comprising providing the option to form at least one part of the current P-picture from at least one previous prediction reference frame, wherein previous is relative to the position of the current P-picture and within the sequence; and

generating motion data for the part of the at least one current P-picture comprising using at least one modified prediction reference frame for generating the motion data for the part of the current P-picture, and the at least one modified prediction reference frame being modified by using a morphing technique or a synthesizing technique or both comprising analyzing a current P-picture to determine parameters of the at least one modified prediction reference frame and to decide which of the techniques to use as part of a selected prediction mode of the current P-picture, and depending, at least in part, on the data of the current P-picture,

wherein at least one morphing technique or synthesizing technique is available for the P-picture but is not available for B-pictures.

17. The method of claim 16 , comprising:

generating at least one P-picture within the sequence by providing an option to generate at least part of the P-picture from at least one of multiple previous prediction reference frames along the sequence;

wherein at least one of:

the P-picture is formed by selecting the best result of a plurality of results from a plurality of reference frames provided to at least partially form the P-picture in an inter-mode, and

the P-picture is formed by using data of a plurality of reference frames used to contribute to a result to at least partially form the P-picture in a multi-mode;

wherein at least one of:

a synthesizing technique comprises projected interpolation wherein a projected interpolation frame or frame part is formed in place of the current P-picture for prediction, and by weighting data of two reference frames depending on a temporal distance along the sequence relative to the position of the current P-picture;

a synthesizing technique comprises modifying the resolution of a frame;

wherein a morphing technique comprises a modified reference frame modified from another reference frame of at least one of:

(1) blur compensation,

(2) registration compensation,

(3) global motion compensation,

(4) dominant motion compensation,

(5) luminance gain compensation, and

all of (1) to (5); and

wherein the modified prediction reference frame is used for prediction but is not displayed.

18. A coder, comprising:

an image buffer;

a graphics processing unit communicatively coupled to the image buffer and comprising at least one morphing analyzer or at least one synthesizing analyzer or both, the graphics processing unit being configured to:

receive data for a sequence of frames of pixel data, the sequence having a display order, and wherein each frame is associated with a temporal position of the display order;

generate at least one of:

a current F-picture using inter-prediction, comprising providing all three options the option to form (1) at least one part of the current F-picture from at least one previous prediction reference frame without using subsequent prediction reference frames, (2) at least one subsequent prediction reference frame without using previous prediction reference frames, or (3) to use both previous and subsequent prediction reference frames, wherein previous and subsequent are relative to the position of the current F-picture and within the sequence, wherein the option to select among the three choices is not available for B-pictures, and

a current P-picture of the sequence of frames using inter-prediction, comprising providing the option to form at least one part of the current P-picture from at least one previous prediction reference frame, wherein previous is relative to the position of the current P-picture and within the sequence; and

generate motion data for the part of the at least one current P-picture or F-picture comprising using at least one modified prediction reference frame for generating the motion data for the part of the current P-picture or F-picture, wherein at least one of:

the morphing analyzer being configured to use a morphing technique,

the synthesizing analyzer being configured to use a synthesizing technique, and

both, and

comprising analyzing a current F-picture or P-picture to determine parameters of the at least one modified prediction reference frame and to decide which technique to use as part of a selected prediction mode of the current F-picture, and depending, at least in part, on the data of the current F-picture,

wherein at least one morphing technique or synthesizing technique is available for the P-picture and F-picture but is not available for B-pictures.

19. The coder of claim 18 wherein a synthesizing technique comprises at least one of:

projected interpolation wherein a projected interpolation frame or frame part is formed in place of the current F-picture for prediction, and by weighting data of two reference frames depending on a temporal distance along the sequence relative to the position of the current F-picture, and

modifying the resolution of a frame,

wherein morphing technique comprises a modified reference frame modified from another reference frame of at least one of:

(1) blur,

(2) registration,

(3) global motion,

(4) dominant motion,

(5) gain in luminance, and

all of (1) to (5);

the coder comprising at least one morphing analyzer or synthesizing analyzer in a decoder loop to reconstruct frames, and at least one morphing analyzer or synthesizing analyzer out of the decoding loop to generate data for motion estimation based on original non-decoded image data.

20. The coder of claim 18 ,

wherein the sequence comprises non-F prediction reference frames used to generate the motion data for the F-pictures, and wherein the F-pictures are spaced between the non-F prediction reference frames, and wherein the number of F-pictures between the non-F prediction reference frames varies along the sequence;

comprising at least one P-picture within the sequence formed by providing an option to generate at least part of the P-picture from at least one of multiple previous prediction reference frames along the sequence;

comprising at least one F-picture of the sequence being formed by providing the option to form at least part of the F-picture from at least one additional prediction reference frame in addition to one previous prediction reference frame, one subsequent prediction reference frame, or both along the sequence and relative to the position of the F-picture;

wherein at least one of:

the P-picture or F-picture is formed by selecting the best result of a plurality of results from a plurality of reference frames provided to at least partially form the P-picture or F-picture in an inter-mode, and

wherein the P-picture or F-picture is formed by using data of a plurality of reference frames used to contribute to a result to at least partially form the P-picture or F-picture in a multi-mode;

the sequence comprising at least one of:

at least one F-picture is a prediction reference frame for another F-picture,

a continuous or non-continuous chain of reference frames along the sequence so that a plurality of F-pictures are reference frames to one next subsequent F-picture, wherein the chain of references is provided in addition to any other reference frames for at least one of the F-pictures, and

a pyramid reference frame structure wherein a reference F-picture has a previous reference frame and a subsequent reference frame along the sequence and is itself a reference frame to another F-picture;

wherein a synthesizing technique comprises at least one of:

projected interpolation wherein a projected interpolation frame or frame part is formed in place of the current F-picture for prediction, and by weighting data of two reference frames depending on a temporal distance along the sequence relative to the position of the current F-picture, and

super resolution frame with a frame with a modified resolution;

wherein morphing technique comprises a modified reference frame modified from another reference frame of at least one of:

(1) blur,

(2) registration,

(3) global motion,

(4) dominant motion,

(5) gain in luminance, and

all of (1) to (5);

wherein the modified prediction reference frame is used for prediction but is not displayed; and

comprising at least one morphing analyzer or synthesizing analyzer in a decoder loop to reconstruct frames, and at least one morphing analyzer or synthesizing analyzer out of the decoding loop to generate data for motion estimation based on original non-decoded image data.