Pattern recognition apparatus using parallel operation

• US 7,274,819 B2
• Filed: 05/30/2002
• Issued: 09/25/2007
• Est. Priority Date: 05/31/2001
• Status: Active Grant

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Matsugu, Masakazu
• Primary Examiner(s): Bella; Matthew C.
• Assistant Examiner(s): Lu; Tom Y
• Application Number:
  US10/156,942
• Publication Number:
  US 20020181775A1
• Time in Patent Office: 1,944 Days
• Field of Search: 382/141, 382/144, 382/145, 382/146, 382/147, 382/148, 382/149, 382/181, 382/209, 382/217
• US Class Current: 382/181
• CPC Class Codes: G06V 10/454 Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]
• CPC Class Codes: G06V 10/955 using specific electronic processors
• CPC Class Codes: G06V 40/171 Local features and components; Facial parts eye characteristics G06V40/18; Occluding parts, e.g. glasses; Geometrical relationships

Abstract

In a pattern recognition apparatus, a local area recognition module is constructed with operation elements having predetermined operation characteristics. Pattern data of a predetermined size in input data is acquired by time-sequentially performing inputting process at a plurality of times via a local area scanning unit, and information indicating the position of pattern data in the input data is output. The local area recognition module detects a feature of a predetermined middle-order or high-order category from the pattern data. A consolidation module time-sequentially consolidates outputs from the local area recognition module on the basis of the position information and the category of the feature thereby producing feature detection map information. A judgment unit outputs position information and category information of a high-order feature present in the input data, on the basis of the output from the time-sequential consolidation module.

Claims

1. A pattern recognition apparatus comprising:

data inputting means for inputting image data by time-sequentially inputting blocks of the image data, each of the blocks being a predetermined size;

position information inputting means for inputting position information representing a position of each of the blocks in the input image data;

pattern detection means for detecting a low-order feature pattern of a predetermined category from each block of the input image data;

prediction means for predicting a category and a position of a first low-order feature pattern to be detected on the basis of a second low-order feature pattern which has been detected;

time-sequential consolidation means for time-sequentially consolidating low-order feature patterns detected from a plurality of blocks of the input image data by said pattern detection means on the basis of the position information input by said position information inputting means and the category of each detected feature pattern, the consolidated low-order feature patterns forming a high-order feature pattern, said time-sequential consolidation means comparing a category of the first low-order feature pattern detected at the predicted position with the predicted category and determining the likelihood of presence of the high-order feature pattern formed by the first low-order feature pattern on the basis of a result of the comparison; and

judgment means for judging a position and a category of a high-order feature pattern present in the input image data, on the basis of the output from said time-sequential consolidation means.

2. A pattern recognition apparatus according to claim 1, wherein said pattern detection means or said time-sequential consolidation means includes storage means for storing a process result.

3. A pattern recognition apparatus according to claim 2, further comprising size changing means for changing the size of the blocks of pattern data, where said judgment means makes judgment on the basis of results of consolidation for different block sizes.

4. A pattern recognition apparatus according to claim 2, wherein said pattern detection means includes an operation element for detecting geometrical features with different sizes in the blocks of pattern data.

5. A pattern recognition apparatus according to claim 1, wherein said data inputting means inputs the blocks of pattern data having a predetermined size by scanning input data.

6. A pattern recognition apparatus according to claim 5, further comprising scanning control means for changing a scanning position of said data inputting means on the basis of the likelihood of presence of a high-order pattern to be detected, determined by said time-sequential consolidation means.

7. A pattern recognition apparatus according to claim 1, further comprising control means for time-sequentially changing an operation characteristic of the operation element of said pattern detection means.

8. A pattern recognition apparatus according to claim 1, wherein said pattern detection means extracts a predetermined local feature at each position in the data.

9. A pattern recognition apparatus according to claim 8, wherein said time-sequential consolidation means stores a detection result of a local feature together with associated position information into a predetermined primary storage means.

10. A pattern recognition apparatus according to claim 1, wherein said pattern detection means is parallel processing means including a plurality of operation elements arranged in parallel and connected to each other.

11. A pattern recognition apparatus according to claim 1, wherein the operation element of said pattern detection means is constructed such that a plurality of feature detection layers and a plurality of feature consolidation layers are alternately disposed and connected in a cascading fashion.

12. A pattern recognition apparatus according to claim 1, wherein said pattern detection means has the predetermined operation characteristic, said time-sequential consolidation means consolidates the outputs, associated with patterns at a plurality of scanning positions, from said pattern detection means, and said judgment means outputs information indicating the position, in the input data, of a pattern of a specified category, together with information indicating the category.

13. A pattern recognition apparatus according to claim 1, wherein said time-sequential consolidation means consolidates patterns detected at scanning positions on the basis of the position information, and said judgment means judges whether there is a high-order pattern including the detected patterns.

14. A pattern recognition apparatus according to claim 1, further comprising control means for controlling the operation characteristic of said pattern detection means so that patterns of different categories with different sizes can be detected in the input pattern data.

15. A pattern recognition apparatus according to claim 1, wherein the detection map information is information about a position of the pattern and at least one of a type and a detection level of the pattern.

16. An image processing apparatus which controls a process performed on a signal of an image in accordance with a signal which is output, after being processed by a pattern recognition apparatus according to claim 1, from said pattern recognition apparatus.

17. A pattern recognition apparatus comprising:

data inputting means for inputting image data by scanning the image data of a predetermined size at a plurality of scanning positions;

detection means for detecting a predetermined feature from each of the scanned image data;

prediction means for predicting a category and a position of first scanned image data to be detected on the basis of second scanned image data which has been detected;

scanning position changing means for changing, a scanning position at which the image data is scanned by said data inputting means to the position predicted by said prediction means;

consolidation means for consolidating a plurality of features detected at different scanning positions on the basis of the scanning position of each detected feature, said consolidation means comparing a category of the first scanned image data detected at the predicted position with the predicted category and determining, on the basis of the result, the likelihood of presence of a specific pattern formed by the first scanned image data; and

judgment means for judging the position and the type of the specific pattern, on the basis of the output from said consolidation means.

18. A pattern recognition apparatus comprising:

a data inputting unit for inputting image data by time-sequentially inputting blocks of the image data, each of the blocks being a predetermined size;

a position information inputting unit for inputting position information representing a position of each of the blocks in the input image data;

a pattern detection for detecting a low-order feature pattern of a predetermined category from each block of the input image data;

a prediction unit for predicting a category and a position of a first low-order feature pattern to be detected on the basis of a second low-order feature pattern which has been detected;

a time-sequential consolidation unit for time- sequentially consolidating low-order feature patterns detected from a plurality of blocks of the input image data detected by said pattern detection unit on the basis of the position information input by said position information inputting unit and the category of each detected feature pattern, the consolidated low-order feature patterns forming a high-order feature pattern, said time-sequential consolidation unit comparing a category of the first low-order feature pattern detected at the predicted position with the predicted category and determining the likelihood of presence of the high-order feature pattern formed by the first low-order feature pattern on the basis of a result of the comparison; and

a judgment unit for judging a position and a category of a high-order feature pattern present in the input image data, on the basis of the output from said time-sequential consolidation unit.

19. A pattern recognition apparatus according to claim 18, wherein said pattern detection unit or said time-sequential consolidation unit includes storage means for storing a process result.

20. A pattern recognition apparatus according to claim 19, further comprising a size changing unit for changing the size of the blocks of pattern data, where said judgment unit makes judgment on the basis of results of consolidation for different block sizes.

21. A pattern recognition apparatus according to claim 19, wherein said pattern detection unit includes an operation element for detecting geometrical features with different sizes in the blocks of pattern data.

22. A pattern recognition apparatus according to claim 18, wherein said data inputting unit inputs the blocks of pattern data having a predetermined size by scanning input data.

23. A pattern recognition apparatus according to claim 22, further comprising a scanning control unit for changing a scanning position of said data inputting unit on the basis of the likelihood of presence of a high-order pattern to be detected, determined by said time-sequential consolidation unit.

24. A pattern recognition apparatus according to claim 18, further comprising a control unit for time-sequentially changing an operation characteristic of the operation element of said pattern detection unit.

25. A pattern recognition apparatus according to claim 18, wherein said pattern detection unit extracts a predetermined local feature at each position in the data.

26. A pattern recognition apparatus according to claim 25, wherein said time-sequential consolidation unit stores a detection result of a local feature together with associated position information into a predetermined primary storage unit.

27. A pattern recognition apparatus according to claim 18, wherein said pattern detection unit is parallel processing unit including a plurality of operation elements arranged in parallel and connected to each other.

28. A pattern recognition apparatus according to claim 18, wherein the operation element of said pattern detection unit is constructed such that a plurality of feature detection layers and a plurality of feature consolidation layers are alternately disposed and connected in a cascading fashion.

29. A pattern recognition apparatus according to claim 18, wherein said pattern detection unit has the predetermined operation characteristic, said time-sequential consolidation unit consolidates the outputs, associated with patterns at a plurality of scanning positions, from said pattern detection unit, and said judgment unit outputs information indicating the position, in the input data, of a pattern of a specified category, together with information indicating the category.

30. A pattern recognition apparatus according to claim 18, wherein said time-sequential consolidation unit consolidates patterns detected at scanning positions on the basis of the position information, and said judgment unit judges whether there is a high-order pattern including the detected patterns.

31. A pattern recognition apparatus according to claim 18, further comprising a control unit for controlling the operation characteristic of said pattern detection unit so that patterns of different categories with different sizes can be detected in the input pattern data.

32. A pattern recognition apparatus according to claim 18, wherein the detection map information is information about a position of the pattern and at least one of a type and a detection level of the pattern.

33. An image processing apparatus which controls a process performed on a signal of an image in accordance with a signal which is output, after being processed by a pattern recognition apparatus according to claim 18, from said pattern recognition apparatus.

34. A pattern recognition apparatus comprising:

a data inputting unit for inputting image data by scanning the image data of a predetermined size at a plurality of scanning positions;

a detection unit for detecting a predetermined feature from each of the scanned image data;

a prediction unit for predicting a category and a position of first scanned image data to be detected on the basis of second scanned image data which has been detected;

a scanning position changing unit for changing a scanning position at which the image data is scanned by said data inputting unit to the position predicted by said prediction means;

a consolidation unit for consolidating a plurality of features detected at different scanning positions on the basis of the scanning position of each detected feature, said consolidation unit comparing a category of the first scanned image data detected at the predicted position with the predicted category and determining, on the basis of the result, the likelihood of presence of a specific pattern formed by the first scanned image data; and

a judgment unit for judging the position and the type of the specific pattern, on the basis of the output from said consolidation unit.

35. A pattern recognition method comprising the steps of:

time-sequentially inputting blocks of the image data, each of the blocks being a predetermined size;

inputting position information representing a position of each of the blocks in the input image data;

detecting a low-order feature pattern of a predetermined category from each block of the input image data;

predicting a category and a position of a first low-order feature pattern to be detected on the basis of a second low-order feature pattern which has been detected;

consolidating low-order feature patterns detected from for a plurality of blocks of the input image data by said pattern detection step on the basis of the position information input in said position information inputting step and the category of each detected feature pattern, said consolidating step comparing a category of the first low-order feature pattern detected at the predicted position with the predicted category and determining the likelihood of presence of the high-order feature pattern formed by the first low-order feature pattern on the basis of a result of the comparison; and

judging a position and a category of a high-order feature pattern present in the input image data, on the basis of the output in said consolidation step.

36. A pattern recognition method according to claim 35, wherein said detection step or said consolidation step includes storing a process result.

37. A pattern recognition method according to claim 36, further comprising a step of changing the size of the blocks of pattern data, where said outputting step outputs information on the basis of results of consolidation for different block sizes.

38. A pattern recognition method according to claim 36, wherein said detection step includes detecting geometrical features with different sizes in the blocks of pattern data.

39. A pattern recognition method according to claim 35, wherein said inputting step inputs the blocks of pattern data having a predetermined size by scanning input data.

40. A pattern recognition method according to claim 39, further comprising a scanning control step of changing a scanning position in said inputting step on the basis of the likelihood of presence of a high-order pattern to be detected, determined in said consolidation step.

41. A pattern recognition method according to claim 35, further comprising a control step for time-sequentially changing an operation characteristic in said detection step.

42. A pattern recognition method according to claim 35, wherein said detection step extracts a predetermined local feature at each position in the data.

43. A pattern recognition method according to claim 42, wherein said consolidation step further includes storing a detection result of a local feature together with associated position information into a predetermined primary storage unit.

44. A pattern recognition method according to claim 35, wherein said outputting step further includes outputting information indicating the position, in the input data, of a pattern of a specified category, together with information indicating the category.

45. A pattern recognition method according to claim 35, wherein said consolidation step further includes consolidating patterns detected at scanning positions on the basis of the position information, and said outputting step judges whether there is a high-order pattern including the detected patterns.

46. A pattern recognition method according to claim 35, further comprising a control step of controlling an operation characteristic in said detection step so that patterns of different categories with different sizes can be detected in the input pattern data.

47. A pattern recognition method according to claim 35, wherein the detection map information is information about a position of the pattern and at least one of a type and a detection level of the pattern.

48. A pattern recognition method comprising the steps of:

scanning image data of a predetermined size at a plurality of scanning positions;

detecting a predetermined feature from each of the scanned image data;

predicting a category and a position of first scanned image data to be detected on the basis of second scanned image data which has been detected;

changing a scanning position at which the image data is scanned in said scanning step to the position predicted by said prediction step;

consolidating a plurality of features detected at different scanning positions on the basis of the scanning position of each detected feature, said consolidation means comparing a category of the first scanned image data detected at the predicted position with the predicted category and determining, on the basis of the result, the likelihood of presence of a specific pattern formed by the first scanned image data; and

judging the position and the type of the specific pattern, on the basis of the output from said consolidation step.