Hybrid transform-based compression
View Patent ↗A system implements a hybrid coding mode. The hybrid coding mode may implement a transform to decompose an input stream into frequency components. The frequency components may include frequency bands such as those resulting from a wavelet transform. The frequency components may have associated coefficients which may be determined via the transform. The hybrid coding mode may also implement a predictor-based coding mode. A predictor-based coding mode uses a set of values as predictors for another set of values. The hybrid mode may be implemented by using predictor-based coding to code a portion of the coefficients. For example, a coefficient may be used as a predictor for another coefficient of same frequency component. In some implementations, dynamic selection between a hybrid coding mode and a point coding mode may be used.
1. An encoding method, comprising:
receiving an input stream comprising a first sample;
determining a non-transform coding size estimate, the non-transform coding size estimate indicating an estimated number of bits to encode the first sample without decomposing the first sample using a transform;
determining a transform coding size estimate, the transform coding size estimate indicating an estimated number of bits to encode the first sample while decomposing the first sample using the transform; and
when the transform coding size estimate is larger than the non-transform coding size estimate, encoding the first sample without decomposing the first sample using the transform, otherwise:
decomposing the first sample via the transform to produce a first coefficient for a first frequency component and a second coefficient for a second frequency component;
encoding the first coefficient using a first coding scheme, the first coding scheme using a predictor based on a third coefficient from the first frequency component; and
encoding the second coefficient using a second coding scheme different from the first coding scheme.
2. The method of claim 1 , wherein encoding the first coefficient using the predictor comprises applying a median adaptive predictor.
3. The method of claim 1 , wherein the third coefficient comprises a vertically displaced coefficient, a horizontally displaced coefficient, or a vertically and horizontally displaced coefficient.
4. The method of claim 1 , further comprising applying rate control to the encoding of the first coefficient and the encoding of the second coefficient.
5. The method of claim 4 , wherein:
a first quantization level is applied for the encoding of the first coefficient; and
a second quantization level is applied for the encoding of the second coefficient, the second quantization level being different from the first quantization level.
6. The method of claim 1 , wherein the second coding scheme comprises a coding scheme that does not rely on correlation.
7. The method of claim 1 , wherein the transform comprises a wavelet transform.
8. The method of claim 1 , wherein the first frequency component comprises a lower frequency band than the second frequency component.
9. The method of claim 8 , wherein the first coefficient corresponds to a lowest frequency component.
10. The method of claim 8 , wherein the second coefficient corresponds to a highest frequency component.
11. The method of claim 1 , wherein decomposing the first sample comprises decomposing the input stream using a multi-dimensional wavelet transformation.
12. The method of claim 1 , wherein:
the input stream further comprises a second sample; and
the method further comprises:
determining an additional non-transform coding size estimate for encoding the second sample without decomposing the second sample using the transform; and
determining an additional transform coding size estimate for encoding while decomposing the second sample using the transform.
13. The method of claim 12 , further comprising:
responsive to the non-transform coding size estimates and the transform coding size estimates:
encoding the second sample without decomposing the second sample using the transform by applying a point coding mode to the second sample; and
encoding the first sample while decomposing the first sample using the transform.
14. The method of claim 13 , wherein the point coding mode comprises an index color history coding mode.
15. The method of claim 12 , wherein:
the input stream comprises an image; and
the first and second samples comprise groups of pixels.
16. A device, comprising:
an input interface configured to receive an input stream comprising a sample; and control circuitry configured to:
determine a non-transform coding size estimate, the non-transform coding size estimate indicating an estimated number of bits to encode the sample without decomposing the sample using a transform;
determine a transform coding size estimate for coding while decomposing the sample using the transform;
when the transform coding size estimate is smaller than the non-transform coding size estimate:
apply the transform to the sample to generate a first coefficient of a first frequency component and a second coefficient of a second frequency component;
determining a predictor for the first coefficient for the first frequency component, the predictor being based on a third coefficient from the first frequency component;
code the first coefficient using a first coding scheme, the first coding scheme using the predictor; and
code the second coefficient using a second coding scheme different from the first; and
when the transform coding size estimate is larger than the non-transform coding size estimate, code the sample without decomposing the sample using the transform.
17. The device of claim 16 , wherein the control circuitry is further configured to code the first coefficient using differential pulse code modulation based coding.
18. The device of claim 16 , wherein the control circuitry is further configured to omit, from an encoded output stream, a fourth coefficient for the second frequency component to indicate a zero value for the fourth coefficient.
19. An encoding method, comprising:
receiving an input stream comprising a sample;
determining a non-transform coding size estimate, the non-transform coding size estimate indicating an estimated number of bits to encode the sample without decomposing the sample using a wavelet transform;
determining a transform coding size estimate for encoding while decomposing the sample using the wavelet transform;
when the transform coding size estimate is smaller than the non-transform coding size estimate:
applying the wavelet transformation to decompose the input stream into a high frequency component and a low frequency component; and
at encoding circuitry:
selecting a first quantization level for rate-controlled encoding of a first coefficient for the low frequency component;
determining a prediction for the first coefficient;
selecting a second quantization level for encoding a second coefficient for the high frequency component;
coding, via differential pulse code modulation, the first coefficient using the first quantization level and the prediction; and
coding the second coefficient using the second quantization level; and
when the transform coding size estimate is larger than the non-transform coding size estimate, encoding, at the encoding circuitry, the sample without decomposing the sample using the transform.
20. The method of claim 19 , wherein the wavelet transformation comprises a multi-dimensional wavelet transformation.