Faster transforms using scaled terms, early aborts, and precision refinements
Faster transforms using scaled terms, early aborts, and precision refinements
 CN 100,429,644 C
 Filed: 10/23/2001
 Issued: 10/29/2008
 Est. Priority Date: 10/23/2000
 Status: Active Grant
First Claim
1. image processing method comprises following steps:
 Receive the data of pending image;
Transformation equation is arranged at least one cluster, and this cluster has at least two transformation constants;
Use a proportional that each cluster is carried out scale operation to described at least two transformation constants independently, to keep the ratio of homogeneous basically between described at least two transformation constants in described at least one cluster;
AndUse described transformation equation that the data of the image of reception are handled, with the data after the generation conversion.
Chinese PRB Reexamination
Abstract
Fast transform (200) that use multiple scaled terms, early aborts, and precision refinements are disclosed. The discrete transforms are split into subtransforms (210) that are independently calculated using multiple scaling terms on the transform constants (220). The effect of the scaling for the transform coefficients may optionally be handled by appropriately scaling the quantization values or any comparison values. The possible need to perform a corrective action is detected based upon testing the incremental calculations of transform coefficients. Corrective action is then performed. The corrective action includes refining the incremental calculations to obtain additional precision and/or aborting the incremental calculations when the resulting numbers are sufficient.

3 Citations
No References
Circuit for computing the quantized coefficient discrete cosine transform of digital signal samples  
Patent #
US 4,849,922 A
Filed 12/11/1987

Current Assignee
N/A

Fast inverse discrete transform using subwords for decompression of information  
Patent #
US 5,394,349 A
Filed 07/10/1992

Current Assignee
N/A

Supercompact hardware architecture for IDCT computation  
Patent #
US 5,854,757 A
Filed 05/07/1996

Current Assignee
N/A

23 Claims

1. image processing method comprises following steps:

Receive the data of pending image; Transformation equation is arranged at least one cluster, and this cluster has at least two transformation constants; Use a proportional that each cluster is carried out scale operation to described at least two transformation constants independently, to keep the ratio of homogeneous basically between described at least two transformation constants in described at least one cluster;
AndUse described transformation equation that the data of the image of reception are handled, with the data after the generation conversion.


2. according to the process of claim 1 wherein that proportional is one of transformation constant from described at least one cluster.

3. according to the process of claim 1 wherein that each transformation constant represented by the estimation transformation constant that 2 power time sum is similar to.

4. according to the method for claim 1, further comprise data are divided at least one and by applied equation this piece is transformed into transform data, to keep the ratio of homogeneous basically between described at least two transformation constants in described at least one cluster.

5. according to the method for one of claim 1 and 4, wherein proportional is to select according to predetermined cost function.

6. according to the method for claim 5, further comprise the step of determining predetermined cost function by the selection percentage item, so that the maximum error of any conversion coefficient is not more than a predetermined error percentage.

7. according to the method for claim 5, further comprise the step of determining predetermined cost function by the selection percentage item, so that the error of predetermined transformation constant is less than or equal to a predetermined error percentage.

8. according to the method for claim 5, further comprise the step of determining predetermined cost function by the expression formula of selection percentage item and transformation constant, so that whole transformation constants of a cluster have the simultaneous binary expression formula of predetermined properties.

9. the process of claim 1 wherein that cluster representative carries out disjoint transformation equation group that part is calculated.

10. the process of claim 1 wherein that cluster is not represented carries out disjoint transformation equation group that part is calculated.

11.
11., further comprise following steps according to the method for claim 1: 
At least one number that test is obtained by the incremental computations of conversion coefficient in conversion process; Determine whether to proofread and correct according to this test;
AndWhen determining to need timing to proofread and correct.


12. according to the method for claim 11, determine whether that wherein the step of proofreading and correct comprises the incremental computations that detects conversion coefficient and whether will produce the unacceptable conversion coefficient of its precision, proofread and correct to comprise and count carefully described at least one number.

13. according to the method for claim 12, wherein conversion comprises and uses a transformation matrix, and wherein counts carefully to comprise and use one and count carefully matrix to improve the precision of transformation constant incremental computations.

14.
14., further comprise the transformation constant that goes out according to approximate treatment and produce the step that at least one counts carefully matrix according to the method for claim 11.

15. according to the method for claim 14, wherein to produce at least one step that counts carefully matrix be that offline carries out or carry out when initialization.

16.
16., wherein produce at least one step that counts carefully matrix and comprise following steps according to the method for claim 14: 
Admit that for producing the n row that conversion recovers a transformation matrix be impossible; For the part of transformation matrix is calculated pseudoinverse;
AndUse the pseudoinverse of this transformation matrix produce described at least one count carefully the approximate of matrix.


17. according to the method for claim 11, determine whether that wherein the step of proofreading and correct comprises further whether definite error that is caused by the termination incremental computations is acceptable, proofread and correct to comprise the incremental computations of ending conversion coefficient.

18.
18., wherein when the resulting several expectations of definite incremental computations will be in the preset range, stop incremental computations according to the method for claim 17.

19.
19., estimate that wherein the number that is in the preset range comprises a conversion coefficient that satisfies accuracy requirement according to the method for claim 18.

20.
20., wherein when determining not change as a result, stop incremental computations to counting carefully of conversion coefficient according to the method for claim 18.

21. method according to claim 20, wherein after the result'"'"'s who checks at least one incremental computations relative size, compare with the results of intermediate calculations of another conversion coefficient hour when the results of intermediate calculations of described at least one conversion coefficient, determine that counting carefully of this conversion coefficient will not change the result.

22.
22., wherein after the result sizes of at least one incremental computations of check,, determine that counting carefully of this conversion coefficient will not change the result when at least one results of intermediate calculations of this conversion coefficient during less than predetermined threshold according to the method for claim 20.

23. according to the method for claim 11, wherein determining step further comprises and determines that a conversion coefficient will be in zero the preset range, proofreaies and correct to comprise the incremental computations of ending conversion coefficient.
Specification(s)