IMAGE CONVERSION METHOD AND APPARATUS, AND PATTERN IDENTIFICATION METHOD AND APPARATUS

• US 20100329556A1
• Filed: 06/18/2010
• Published: 12/30/2010
• Est. Priority Date: 06/26/2009
• Status: Active Grant

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Mitarai, Yusuke, Matsugu, Masakazu, Mori, Katsuhiko
• Granted Patent: US 8,340,439 B2
• Time in Patent Office: Days
• US Class Current: 382/170
• CPC Class Codes: G06F 18/2137 based on criteria of topology preservation, e.g. multidimensional scaling or self-organising maps
• CPC Class Codes: G06V 10/28 Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
• CPC Class Codes: G06V 10/54 relating to texture
• CPC Class Codes: G06V 10/7715 Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods

Abstract

In an image conversion method, a value which reflects the mutual relationship between the classes of pixel patterns each formed from a pixel classified as one of a plurality of classes and peripheral pixels is set as a converted value corresponding to each of the plurality of classes, a pixel of interest is sequentially selected from the input image, and a pixel pattern formed from the selected pixel of interest and a predetermined number of pixels around it is classified as one of the plurality of classes in accordance with a neighboring pattern obtained based on the relationship between the value of the pixel of interest and the values of peripheral pixels located at predetermined relative positions with respect to the pixel of interest. The value of the pixel of interest is converted into a converted value set for a class to which the pixel of interest has been classified.

Claims

1. An image conversion method comprising:

a setting step of setting a value as a converted value corresponding to each of a plurality of classes, the value reflecting a mutual relationship between classes of pixel patterns each formed from a pixel classified as one of the plurality of classes and peripheral pixels;

a selection step of sequentially selecting a pixel of interest from an input image;

a classification step of classifying the selected pixel of interest as one of the plurality of classes in accordance with a neighboring pattern obtained based on a relationship between a value of the pixel of interest and values of peripheral pixels located at predetermined relative positions with respect to the pixel of interest; and

a conversion step of converting the value of the pixel of interest into a converted value set for a class to which the pixel of interest has been classified in the classification step.

2. The method according to claim 1, wherein the peripheral pixels are a plurality of pixels adjacent to the pixel of interest.

3. The method according to claim 1, wherein the peripheral pixels are a plurality of pixels located at positions spaced apart from the pixel of interest by a predetermined distance.

4. The method according to claim 1, wherein, in the classification step, the neighboring pattern is acquired based on a result of comparison between a pixel value of the pixel of interest and pixel values of the peripheral pixels.

5. The method according to claim 4, wherein, in the setting step, the converted value corresponding to each of the plurality of classes is set so as to preserve a distance or similarity between neighboring patterns for a pixel of interest classified as one of the plurality of classes and peripheral pixels.

6. The method according to claim 1, wherein, in the classification step, a distance or similarity between the neighboring pattern and a representative pattern of each of the plurality of classes is obtained, and the pixel of interest is classified as one of the plurality of classes based on the distance or similarity.

7. The method according to claim 6, wherein, in the setting step, the converted value is set so as to preserve a distance or similarity between the representative patterns of the plurality of classes.

8. The method according to claim 7, wherein the representative pattern is a representative vector obtained by vector quantization of a plurality of pattern sample groups each prepared in advance and classified as one of the plurality of classes.

9. The method according to claim 8, wherein the converted value is a vector value, and a dimension of the vector value is lower than that of the representative vector.

10. The method according to claim 1, wherein, in the setting step, the converted value corresponding to each of the plurality of classes is determined so as to preserve a distance or similarity between the classes obtained based on the pixel pattern classified as one of the plurality of classes.

11. The method according to claim 10, wherein, in the setting step, the converted value corresponding to each of the plurality of classes is set so as to preserve a cluster distance or inter-cluster similarity between the plurality of classes obtained by applying cluster analysis to class-specific pixel pattern sample groups belonging to the plurality of classes, respectively.

12. The method according to claim 11, wherein, in the setting step, the cluster analysis is performed using pixel values of each pixel pattern in the class-specific sample groups.

13. The method according to claim 11, wherein, in the setting step, the converted value corresponding to each of the plurality of classes is determined based on pixel values of each pixel pattern in the class-specific sample groups and associated information containing information for specifying an original image from which each pixel pattern is extracted.

14. The method according to claim 13, wherein the associated information further contains information representing a spatial arrangement position in the original image from which each pixel pattern is extracted.

15. The method according to claim 10, wherein, in the setting step, a map is generated by mapping each of the plurality of classes on a space having at least one dimension using distances between the classes, and a coordinate value corresponding to each class on the map is used as the converted value corresponding to the class.

16. The method according to claim 15, wherein the mapping is done using multi-dimensional scaling or Isomap.

17. The method according to claim 1, wherein, in the setting step, a self-organizing map is generated using a pixel pattern sample group prepared for each of the plurality of classes, and the converted value is set based on the self-organizing map.

18. The method according to claim 17, wherein, in the setting step,

based on a weight vector of each neuron on the self-organizing map, the neuron is classified as one of the plurality of classes,

coordinates of a position of center of gravity of the neuron classified as one of the classes are obtained in each of the plurality of classes, and

the coordinates of the position of center of gravity obtained for each of the plurality of classes are used as the converted value corresponding to each of the plurality of classes.

19. The method according to claim 17, wherein, in the setting step,

based on a weight vector of each neuron on the self-organizing map, the neuron is classified as one of the plurality of classes,

a maximum cluster on the self-organizing map is detected in each of the plurality of classes, and

coordinates of a position of center of gravity of the maximum cluster obtained for each of the plurality of classes are used as the converted value corresponding to each of the plurality of classes.

20. The method according to claim 17, wherein, in the setting step,

the generated self-organizing map is divided into a plurality of regions,

each region obtained by the division is defined as a bin of a histogram having as many dimensions as the self-organizing map, and a bin corresponding to each class is determined based on a mapping state of the plurality of classes in each of the plurality of regions on the self-organizing map, and

an index representing a position of the bin of the histogram corresponding to each class is used as the converted value corresponding to each of the plurality of classes.

21. A pattern identification method comprising:

an acquisition step of acquiring a converted image formed by replacing each pixel of an input image with a converted value by an image conversion method of claim 1; and

an identification step of identifying a pattern of the input image using the converted image acquired in the acquisition step.

24. A storage medium storing a program which causes a computer to execute an image conversion method of claim 1.

22. An image conversion apparatus comprising:

a setting unit configured to set a value as a converted value corresponding to each of a plurality of classes, the value reflecting a mutual relationship between classes of pixel patterns each formed from a pixel classified as one of the plurality of classes and peripheral pixels;

a selection unit configured to sequentially select a pixel of interest from an input image;

a classification unit configured to classify the selected pixel of interest as one of the plurality of classes in accordance with a neighboring pattern obtained based on a relationship between a value of the pixel of interest and values of peripheral pixels located at predetermined relative positions with respect to the pixel of interest; and

a conversion unit configured to convert the value of the pixel of interest into a converted value set for a class to which the pixel of interest has been classified by said classification unit.

23. A pattern identification apparatus comprising:

an image conversion apparatus of claim 22;

an acquisition unit configured to acquire, from said image conversion apparatus, a converted image formed by replacing each pixel of an input image with a converted value; and

an identification unit configured to identify a pattern of the input image using the converted image acquired by said acquisition unit.