Pattern recognition apparatus using parallel operation

• US 20020181775A1
• Filed: 05/30/2002
• Published: 12/05/2002
• 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
• Granted Patent: US 7,274,819 B2
• Time in Patent Office: Days
• US Class Current: 382/195
• 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:

time-division data inputting means for inputting data by time-sequentially inputting blocks of pattern data, which are part of the input data and which have a predetermined size;

position information inputting means for inputting position information of the blocks of pattern data in the input data;

pattern detection means including an operating element having a predetermined operation characteristic, for detecting a pattern of a predetermined middle-order or high-order category in the blocks of pattern data;

time-sequential consolidation means for time-sequentially consolidating the output from said pattern detection means on the basis of the position information and the category of the detected pattern and producing pattern detection map information; and

judgment means for outputting position information and category information of a high-order pattern present in the input 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 time-division 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 time-division data inputting means on the basis of a degree of consistency with 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.

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.

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 time-division 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 time-division data inputting unit on the basis of a degree of consistency with 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.

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 a degree of consistency with 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.

8. A pattern recognition apparatus according to Clam 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.

17. A pattern recognition apparatus comprising:

data inputting means for scanning pattern data of a predetermined size, which is part of input data, thereby inputting the pattern data;

detection means for detecting a predetermined feature in the pattern data;

scanning position changing means for changing, on the basis of the type of the feature, a scanning position at which the pattern data is scanned by said data inputting means;

consolidation means for consolidating a plurality of features detected at different scanning positions and determining, on the basis of consolidation result, the likelihood of presence of a specific pattern; and

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

18. A pattern recognition apparatus comprising:

a time-division data inputting unit for inputting data by time-sequentially inputting blocks of pattern data, which are part of the input data and which have a predetermined size;

a position information inputting unit for inputting position information of the blocks of pattern data in the input data;

a pattern detection unit including an operating element having a predetermined operation characteristic, for detecting a pattern of a predetermined middle-order or high-order category in the blocks of pattern data;

a time-sequential consolidation unit for time-sequentially consolidating the output from said pattern detection unit on the basis of the position information and the category of the detected pattern and producing pattern detection map information; and

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

34. A pattern recognition apparatus comprising:

a data inputting unit for scanning pattern data of a predetermined size, which is part of input data, thereby inputting the pattern data;

a detection unit for detecting a predetermined feature in the pattern data;

a scanning position changing unit for changing, on the basis of the type of the feature, a scanning position at which the pattern data is scanned by said data inputting unit;

a consolidation unit for consolidating a plurality of features detected at different scanning positions and determining, on the basis of consolidation result, the likelihood of presence of a specific pattern; and

a judgment unit for outputting position information indicating the position of the specific pattern and information indicating 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 pattern data, which are part of input data and which have a predetermined size;

inputting position information of the blocks of pattern data in the input data;

detecting a pattern of a predetermined middle-order or high-order category in the blocks of pattern data;

consolidating the output in said pattern detection step on the basis of the position information and the category of the detected pattern and producing pattern detection map information; and

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

48. A pattern recognition method comprising the steps of:

scanning pattern data of a predetermined size, which is part of input data, thereby inputting the pattern data;

detecting a predetermined feature in the pattern data;

changing, on the basis of the type of the feature, a scanning position at which the pattern data is scanned in said scanning step;

consolidating a plurality of features detected at different scanning positions and determining, on the basis of consolidation result, the likelihood of presence of a specific pattern; and

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