1. A data transmitting apparatus comprising:
a processor;
N sets of external terminals, where N is an integer not less than 2;
N interface circuits coupled over a bus to said processor and adapted to transfer data between said processor and respective ones said N sets of external terminals; and
an interface testing circuit coupled over a bus to said processor and selectively coupled to one of said N sets of external terminals.
2. The data transmitting apparatus according to claim 1, wherein said interface circuit is adapted to receive data from or to output data to an associated set of external terminals, respectively; and wherein
said interface testing circuit includes a selection circuit that selects one of said N interface circuits, and selects whether data is to be received from or output to the set of external terminals associated with the selected interface circuit.
3. The data transmitting apparatus according to claim 1, wherein said interface testing circuit is designed and constructed so that, in case said interface circuit operates as an outputting circuit, said interface testing circuit receives and buffers data output by said interface circuit to the associated external terminals so that said data can be read out by said processor.
4. The data transmitting apparatus according to claim 1, wherein said interface testing circuit is designed and constructed so that, in case said interface circuit operates as a receiving circuit, said interface testing circuit outputs data pre-written and buffered therein by said processor so that said interface circuit will receive said data.
5. The data transmitting apparatus according to claim 3, wherein said interface testing circuit comprises a signal generating circuit; said signal generating circuit including a clock generating circuit that generates a sampling clock signal for over-sampling data output from said interface circuit; said signal generating circuit sampling the data output by said interface circuit with said sampling clock signal, receiving said sampled data and holding said received data;
said processor reading out data held by said signal generating circuit.
6. The data transmitting apparatus according to claim 3, wherein said interface testing circuit comprises a signal generating circuit, said signal generating circuit including a clock receiving circuit that receives a clock signal synchronized with data output from said interface circuit; said signal generating circuit receiving and holding data output from said interface circuit in synchronization with said clock signal;
said processor reading out data held by said signal generating circuit.
7. The data transmitting apparatus according to claim 4, wherein said interface testing circuit comprises a signal generating circuit; said signal generating circuit including a clock generating circuit that generates a sampling clock signal for over-sampling the data received by said interface circuit; said signal generating circuit outputting the data received by said interface circuit in synchronization with said sampling clock signal;
said processor pre-writing data output from said signal generating circuit in said signal generating circuit.
8. The data transmitting apparatus according to claim 4, wherein said interface testing circuit comprises a signal generating circuit; said signal generating circuit including a clock receiving circuit that receives a clock signal synchronized with the data received by said interface circuit; said signal generating circuit outputting the data received by said interface circuit in synchronization with said clock signal;
said processor pre-writing data output from said signal generating circuit in said signal generating circuit.
9. A semiconductor device including the data transmitting apparatus according to claim 1.
The claims below are in addition to those above.
All refrences to claim(s) which appear below refer to the numbering after this setence.
1. An image recognition method comprising
normalizing static image data; and
classifying the normalized static image data using a support vector machine with a defined kernel that implicitly evaluates a linear kernel on input data after a one-time precomputation of an image training data set,
wherein for some positive integers k, and m, the vector size of the normalized input data is k, the vector size of the linear features of the input data is m, with m being greater than k, and
wherein the precomputation comprises a transformation in Rkxk; and
performing an image recognition based at least in part on the classifying.
2. The method of claim 1, wherein the classified static image data is a face.
3. The method of claim 1, wherein A\u03b5Rmxk is the matrix representing the linear transformation applied to the normalized static input image data, and wherein B\u03b5Rkxk is determined by computing B:=ATA.
4. The method of claim 1, wherein A\u03b5Rmxk is the matrix representing the linear transformation applied to the normalized static input image data, and wherein B\u03b5Rkxk is determined by computing B:=ATA using a blocking scheme to split A into a set of smaller matrices A1, . . . , Ap with AT=A1T, . . . , ApT, for incremental computation of B\u03b5Rkxk.
5. The method of claim 1, further comprising a pre-computation determination of xi\u2032=Bxi for all support vectors xi prior to classification, keeping the xi\u2032 in memory as new support vectors.
6. The method of claim 1, wherein the defined kernel is a linear kernel.
7. The method of claim 1, wherein the defined kernel is a sigmoid kernel.
8. The method of claim 1, wherein the defined kernel is a polynomial kernel.
9. The method of claim 1, wherein the defined kernel is a RBF kernel.
10. The method of claim 1, wherein the defined kernel is a RBF kernel evaluated as K(xi, xi\u2032)=exp(\u2212(si\u22122xTxi\u2032+xT Bx)c, further comprising a pre-computation determination of xi\u2032=Bxi for all support vectors xi and a pre-computation determination of si=xiTBxi for all pairs of support vectors xi and xj prior to classification, keeping the xi\u2032 as new support vectors and si in memory.
11. An article comprising a storage medium having stored thereon instructions that when executed by a machine result in:
normalizing static image data; and
classifying the normalized static image data using a support vector machine with a defined kernel that implicitly evaluates a linear kernel on input data after a one-time precomputation of an image training data set,
wherein for some positive integers k, and m, the vector size of the normalized input data is k, the vector size of the linear features of the input data is m, within being greater than k, and
wherein the precomputation comprises a transformation in Rkxk and
performing an image recognition based at least in part on the classifying.
12. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein the classified static image data is a face.
13. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein A\u03b5Rmxk is the matrix representing the linear transformation applied to the normalized static input image data, and wherein B\u03b5Rkxk is determined by computing B:=ATA.
14. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein A\u03b5Rmxk is the matrix representing the linear transformation applied to the normalized input static image data set, , and wherein B\u03b5Rkxk is determined by computing B:=ATA using a blocking scheme to split A into a set of smaller matrices A1, . . . , Ap with AT=A1T, . . . , ApT, for incremental computation of B\u03b5Rkxk.
15. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein the defined kernel is a linear kernel.
16. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein the defined kernel is a sigmoid kernel.
17. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein the defined kernel is a polynomial kernel.
18. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein the defined kernel is a RBF kernel.
19. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein a pre-computation determination of xi\u2032=Bxi for all support vectors xi is made prior to classification, keeping the xi\u2032 in memory as new support vectors.
20. The article comprising a storage medium having stored thereon instructions according to claim 11, wherein the defined kernel is a RBF kernel evaluated as
K(x, xi\u2032)=exp(\u2212(si\u22122xTxi\u2032+xTBx)c, further comprising a pre-computation determination of xi\u2032=Bxi for all support vectors xi and a pre-computation determination of si=xiTBxi for all pairs of support vectors xi and xj prior to classification, keeping the xi\u2032 as new support vectors and si in memory.
21. A system for improving speed of support vector machines comprising:
a software module to normalize static image data; and
a classification module to evaluate the normalized static image data using a support vector machine with a defined kernel that implicitly evaluates a linear kernel on input data after a one-time precomputation of an image training data set
wherein for some positive integers k, and in, the vector size of the normalized input data is k, the vector size of the linear features of the input data is m, with m being greater than k, and
wherein the precomputation comprises a transformation in Rkxk; and
performing an image recognition based at least in part on the classifying.
22. The system of claim 21, wherein the static image data is a face.
23. The system of claim 21, wherein A\u03b5Rmxk is the matrix representing the linear transformation applied to the normalized static input image data, and wherein B\u03b5Rkxk is determined by computing B:=ATA.
24. The system of claim 21, wherein A\u03b5Rmxk is the matrix representing the linear transformation applied to the input static image data set, with vector size of the input static image data being k, and wherein B\u03b5Rkxk is determined by computing B:=ATA using a blocking scheme to split A into a set of smaller matrices A1, . . . , Ap with AT=A1T, . . . , ApT, for incremental computation of B\u03b5Rkxk.
25. The system of claim 21, wherein the defined kernel is a linear kernel.
26. The system of claim 21, wherein the defined kernel is a sigmoid kernel.
27. The system of claim 21, wherein the defined kernel is a polynomial kernel.
28. The system of claim 21, wherein the defined kernel is a RBF kernel.
29. The system of claim 21, wherein a pre-computation determination of xi\u2032=Bxi for all support vectors xi is made prior to classification, keeping the xi\u2032 in memory as new support vectors.
30. The system of claim 21, wherein the defined kernel is a RBF kernel evaluated as
K(x, xi\u2032)=exp(\u2212(si\u22122xTxi\u2032+xTBx)c, further comprising a pre-computation determination of xi\u2032=Bxi for all support vectors xi and a pre-computation determination of si=xiTBxi for all pairs of support vectors xi and xj prior to classification, keeping the xi\u2032 as new support vectors and si in memory.
31. An image recognition training and classification method comprising;
training a support vector machine using a one-time precomputation of a large feature data set to implicitly evaluate a linear kernel; and
classifying normalized static image data using a support vector machine with a defined kernel that implicitly evaluates a linear kernel on input data after a one-time precomputation of an image training data set
wherein for some positive integers k, and m, the vector size of the normalized input data is k, the vector size of the linear features of the input data is m, with m being greater than k, and
wherein the precomputation comprises a transformation in Rkxk and performing an image recognition based at least in part on the classifying.
32. The method of claim 31, wherein the classified static image data is a face.
33. The method of claim 31, wherein one-time precomputation sets A\u03b5Rmxk as a matrix representing the linear transformation applied to the static input image data set, and wherein B\u03b5Rkxk is determined by computing B:=ATA using a blocking scheme to split A into a set of smaller matrices A1, . . . , Ap with AT=A1T, . . . , ApT, for incremental computation of B\u03b5Rkxk.
34. An article comprising a storage medium having stored thereon instructions that when executed by a machine result in:
training a support vector machine using a one-time precomputation of a large feature data set to implicitly evaluate a linear kernel; and
classification of normalized static image data using a support vector machine with a defined kernel that implicitly evaluates a linear kernel on input data after a one-time precomputation of an image training data set
wherein for some positive integers k, and m, the vector size of the normalized input data is k, the vector size of the linear features of the input data is m, with m being greater than k, and
wherein the precomputation comprises a transformation in Rkxk; and
performing an image recognition based at least in part on the classifying.
35. The article comprising a storage medium having stored thereon instructions according to claim 34, wherein the classified static image data is a face.
36. The article comprising a storage medium having stored thereon instructions according to claim 34, wherein one-time precomputation sets A\u03b5Rmxk as a matrix representing the linear transformation applied to the input static image data set, and wherein B\u03b5Rkxk is determined by computing B:=ATA using a blocking scheme to split A into a set of smaller matrices A1, . . . , Ap with AT=A1T, . . . , ApT, for incremental computation of B\u03b5Rkxk.
37. An image recognition training and classification system comprising:
a training module for a support vector machine providing a one-time precomputation of a large feature data set to implicitly evaluate a linear kernel; and
a classification module for normalized static image data using a support vector machine with a defined kernel that implicitly evaluates a linear kernel on input data after a one-time precomputation of an image training data set
wherein for some positive integers k, and m, the vector size of the normalized input data is k, the vector size of the linear features of the input data is m, with m being greater than k, and
wherein the precomputation comprises a transformation in Rkxk; and
performing an image recognition based at least in part on the classifying.
38. The system of claim 37, wherein the classified static image data is a face.
39. The system of claim 37, wherein one-time precomputation sets A\u03b5Rmxk as a matrix representing the linear transformation applied to the input static image data set, and wherein B\u03b5Rkxk is determined by computing B:=ATA using a blocking scheme to split A into a set of smaller matrices A1, . . . , Ap with AT=A1T, . . . , ApT, for incremental computation of B\u03b5Rkxk.