Efficient encoding/decoding of a sequence of data frames

• US 8,756,482 B2
• Filed: 05/25/2007
• Issued: 06/17/2014
• Est. Priority Date: 05/25/2007
• Status: Active Grant

• Current Assignee: NVIDIA Corporation
• Original Assignee: NVIDIA Corporation
• Inventors: Goel, Anurag
• Primary Examiner(s): Rizk, Sam
• Application Number:
  US11/753,611
• Publication Number:
  US 20080294962A1
• Time in Patent Office: 2,580 Days
• Field of Search: 714/776, 375/240.01, 375/240.12, 375/240.24
• US Class Current: 714/776
• CPC Class Codes: H04N 19/11 among a plurality of spatial predictive coding modes
• CPC Class Codes: H04N 19/176 the region being a block, e.g. a macroblock
• CPC Class Codes: H04N 19/48 using compressed domain processing techniques other than decoding, e.g. modification of transform coefficients, variable length coding [VLC] data or run-length data motion estimation in a transform domain H04N19/547; processing of decoded motion vectors H04N19/513
• CPC Class Codes: H04N 19/60 using transform coding

Abstract

Encoding data by first performing a transformation of predicted data and input data, and then performing a subtraction of the resulting outputs. In an embodiment, the prediction approach is chosen such that fewer elements of different values (compared to a number of elements in the input data) are generated, and the different values are generated in a predictable position. The transformation approach is chosen such that the output expressly represents variations in the input data as well as satisfies a distributive property. The decoding may be performed based on the same concepts. As a result, the data can be encoded and/or decoded efficiently.

Claims

1. An encoder for encoding a sequence of data frames, said encoder comprising:

a first transformation block receiving a first data frame contained in said sequence of data frames and generating a transformed frame according to a discrete cosine transformation approach;

a prediction block generating a predicted frame corresponding to said first data frame from data frames prior to said first data frame in said sequence using a prediction approach selected from a group consisting of a mean prediction, a horizontal prediction and a vertical prediction, wherein a reduced number of data values for each micro block of the data frames prior to said first data frame is computed;

a second transformation block generating a transformed predicted data by processing said predicted frame also according to said discrete cosine transformation approach;

a subtractor block generating a difference of said transformed predicted data and said transformed frame to generate a difference data; and

a quantization block quantizing said difference data to generate quantized data corresponding to said first data frame.

2. The encoder of claim 1, wherein said transformation approach satisfies a distributive property.

3. The encoder of claim 2, wherein said prediction approach is designed to generate fewer elements of different values compared to a number of elements in said first data frame and the different values are generated in predictable positions, wherein said transformation approach is chosen such that the output expressly represents variations in the input data.

4. The encoder of claim 3, wherein said prediction approach comprises a mean approach, and said sub tractor block performs a single subtraction only to generate said difference data.

5. The encoder of claim 4, wherein said prediction approach comprises one of horizontal prediction or vertical prediction, wherein said first transformation block performs transformation of said first data frame in one dimension, wherein the said dimension is along the direction of the prediction, said encoder further comprising a third transformation block performing transformation in another dimension of an output generated by said sub tractor block, wherein the output of said third transformation block is provided as an input to said quantization block.

6. The encoder of claim 2, wherein the said transformation transforms the predicted data by multiplying the predicted data with a scale factor, representing a single number.

7. A decoder for decoding a sequence of encoded data frames, said decoder comprising:

an inverse quantization block generating a plurality of scaled values corresponding to the digital values in a first encoded data frame contained in said sequence of encoded data frames;

a prediction block generating a predicted frame corresponding to said first encoded data frame from data frames prior to said first encoded data frame in said sequence using a prediction approach selected from the group consisting of a mean prediction, a horizontal prediction, a vertical prediction, wherein a reduced number of data values for each micro block of the data frames prior to said first encoded data is computed;

a transformation block transforming said predicted frame according to a discrete cosine transformation approach to generate a transformed frame;

an adder adding respective values of said transformed frame and said plurality of scaled values to generate an added frame; and

a first inverse transformation block performing an inverse transform of said discrete cosine transformation approach on said added frame to generate a decoded frame corresponding to said first encoded data frame.

8. The decoder of claim 7, further comprising:

a second inverse transformation block located between said inverse quantization block and said adder, and performing a partial inverse transformation of said plurality of scaled values to generate a partially transformed frame, wherein said adder adds said plurality of scaled values in partially transformed form,

wherein said first inverse transformation block also completes said partial inverse transformation.

9. The decoder of claim 8, wherein said prediction approach is designed to generate fewer elements of different values compared to a number of elements in said first encoded data frame and the different values are generated in predictable positions, wherein said transformation approach is chosen such that the output expressly represents variations in the input data.

10. The decoder of claim 9, wherein said prediction approach comprises a mean approach, and said adder performs a single addition.

11. The decoder of claim 7, wherein said transformation approach satisfies a distributive property.

12. A method of processing a plurality of data frames, said method comprising:

receiving a first data frame contained in a sequence of data frames and generating a transformed frame according to a discrete cosine transformation approach, wherein said sequence of data frames are contained in said plurality of data frames;

generating a predicted frame corresponding to said first data frame from data frames prior to said first data frame in said sequence using a prediction approach selected from a group consisting of a mean prediction, a horizontal prediction and a vertical prediction, wherein a reduced number of data values for each micro block of the data frames prior to said first data frame is computed;

generating a transformed predicted data by processing said predicted frame also according to said discrete cosine transformation approach;

generating a difference of said transformed predicted data and said transformed frame to generate a difference data; and

quantizing said difference data to generate quantized data corresponding to said first data frame.

13. The method of claim 12, wherein said transformation approach satisfies a distributive property.

14. The method of claim 13, wherein said prediction approach is designed to generate fewer elements of different values compared to a number of elements in said first data frame and the different values are generated in predictable positions, wherein said transformation approach is chosen such that the output expressly represents variations in the input data.

15. The method of claim 12, wherein said prediction approach comprises one of horizontal prediction or vertical prediction, wherein said first data frame is transformed first in one dimension, wherein the said dimension is along the direction of the prediction, said method further comprising performing transformation in another dimension of said difference data.

16. The method of claim 12, further comprising;

receiving a second data frame in an encoded format, wherein said second data frame is contained in a second sequence of frames comprised in said plurality of data frames;

generating a plurality of scaled values corresponding to the digital values in said second data frame;

generating a second predicted frame corresponding to said second data frame from data frames prior to said second data frame in said second sequence of frames using a second prediction approach selected from the group consisting of a mean prediction, a horizontal prediction, a vertical prediction, wherein a reduced number of data values for each micro block of the data frames prior to said first encoded data is computed;

generating a transformed second predicted data by processing said second predicted data also according to said discrete cosine transformation approach;

adding respective values of said transformed second frame and said plurality of scaled values to generate an added frame; and

performing an inverse transform of said discrete cosine transformation approach on said added frame to generate a decoded frame corresponding to said second encoded data frame.

17. A non-transitory machine readable medium carrying one or more sequences of instructions for causing a system to process a plurality of data frames, wherein execution of said one or more sequences of instructions by one or more processors contained in said system causes said system to perform the actions of:

receiving a first data frame contained in a sequence of data frames and generating a transformed frame according to a discrete cosine transformation approach, wherein said sequence of data frames are contained in said plurality of data frames;

generating a predicted frame corresponding to said first data frame from data frames prior to said first data frame in said sequence using a prediction approach, wherein a reduced number of data values for each micro block of the data frames prior to said first data frame is computed;

generating a transformed predicted data by processing said predicted frame also according to said discrete cosine transformation approach;

generating a difference of said transformed predicted data and said transformed frame to generate a difference data; and

quantizing said difference data to generate quantized data corresponding to said first data frame.

18. The non-transitory machine readable medium of claim 17, wherein said transformation approach satisfies a distributive property.

19. The non-transitory machine readable medium of claim 18, further comprising:

receiving a second data frame in an encoded format, wherein said second data frame is contained in a second sequence of frames comprised in said plurality of data frames;

generating a plurality of scaled values corresponding to the digital values in said second data frame;

generating a second predicted frame corresponding to said second data frame from data frames prior to said second data frame in said second sequence using a second prediction approach, wherein a reduced number of data values for each micro block of the data frames prior to said first encoded data is computed;

generating a transformed second predicted data by processing said second predicted frame also according to said discrete cosine transformation approach;

adding respective values of said transformed second predicted data and said plurality of scaled values to generate an added frame; and

performing an inverse transform of said discrete cosine transformation approach on said added frame to generate a decoded frame corresponding to said second encoded data frame.

20. The non-transitory machine readable medium of claim 17, wherein generating said predicted frame, generating said transformed predicted data and generating said difference is performed by a single instruction.