Method and apparatus for processing visual information

• US 6,115,480 A
• Filed: 03/27/1996
• Issued: 09/05/2000
• Est. Priority Date: 03/31/1995
• Status: Expired due to Fees

• Current Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Original Assignee: Canon Kabushiki Kaisha (Canon Inc.)
• Inventors: Washizawa, Teruyoshi
• Primary Examiner(s): CHANG, JON CARLTON
• Application Number:
  US08/624,803
• Time in Patent Office: 1,623 Days
• Field of Search: 382/190, 382/153, 382/156, 382/157, 382/107, 382/148, 382/291, 382/298, 382/158, 382/103, 382/151, 348/95, 348/169, 706/19, 706/20
• US Class Current: 382/103
• CPC Class Codes: G06T 5/20 using local operators

Abstract

A 2D image supplied from an image input unit including a wide view lens is sampled into a discrete form by an array sensor, and then mapped to a multi-resolution space by a 2D filter. The feature of the supplied image is detected, and then the mapped image is transformed to a local pattern about the detected feature, and then the coordinates of the position of the feature and the code word of the local pattern are formed into a set which is then encoded. The code is supplied to each cell of a stochastic automaton. The quantity of visual information is calculated in accordance with the quantity of mutual information between different cells of the stochastic automaton consisting of cells in blocks, the coordinates of the position of the feature and the distance from the feature to the optical axis so as to control the optical axis of the image input unit in such a manner that the quantity of visual information is maximized.

Claims

1. A method of controlling an image information processing apparatus comprising the steps of:

optically receiving an image from an image input unit of said image information processing apparatus;

obtaining partial images by segmenting the received image;

determining a probable local pattern and an appearance probability thereof for each of the partial images;

detecting at least one feature point from the received image;

calculating a quantity of mutual information between each feature point and a current observation point on a basis of appearance probabilities of local patterns for respective partial images including the feature point and the current observation point;

calculating a quantity of visual information for each feature point in accordance with the quantity of mutual information between respective feature points and the current observation point and with a distance between the current observation point and the respective feature points;

determining a feature point corresponding to a maximum quantity of visual information as a next observation point; and

moving the image input unit so that the optical axis matches the next observation point.

2. A method of controlling an image information processing apparatus according to claim 1, wherein zooming of the image input unit is performed in said moving step.

3. A method of controlling an image information processing apparatus according to claim 1, wherein the quantity of visual information is, in said second calculating step, calculated in a Hopfield-type neural network in accordance with entropy.

4. A method according to an image information processing apparatus according to claim 1, wherein the partial images are, in said obtaining step, obtained to the extent of a predetermined depth in a multi-resolution space.

5. A method of controlling an image information processing apparatus according to claim 4, wherein said obtaining step includes a step for causing a supplied image to be discrete and a step for masking the discrete 2D image.

6. A method of controlling an image information processing apparatus according to claim 1, wherein a wide view lens is, in said optically receiving step, used to transform the coordinates of the received image into a direction perpendicular to said wide view lens.

7. A method of controlling an image information processing apparatus according to claim 1, wherein the quantity of visual information is, in said second calculating step, calculated by a plurality of symmetry-connected-type neural networks for mutually transmitting/receiving signals.

8. An image information processing method comprising the steps of:

monitoring a supplied image;

detecting at least one feature point from the supplied image;

extracting an image information in a neighborhood of each feature point;

calculating an evaluation value of each feature point in accordance with the image information in a neighborhood of a respective feature point and a distance between an optical axis and the respective feature point;

selecting a feature point one by one, the evaluation value of which is higher than a predetermined value;

moving a direction of an optical axis to the selected feature point;

acquiring data of an image after moving the optical axis to the selected feature point; and

allotting an identifier to the acquired image data and storing a set formed by the position of the selected feature point, data of the acquired image, time of detection and the allotted identifier.

9. An image information processing method according to claim 8, further comprising the steps of:

determining a feature point, the evaluation value of which is higher than a second predetermined value;

identifying an identifier allotted to an image in accordance with the position of the determined feature point; and

storing a set formed by the position of the determined feature point, time of determination and the identified identifier.

10. An image information processing method according to claim 8 wherein in a case where a plurality of features have been detected in said detection step, the optical axis is, in said moving step, moved toward a feature nearest the present optical axis.

11. An image information processing apparatus comprising:

image input means for optically inputting an image;

obtaining means for obtaining partial images by segmenting the received image;

first determination means for determining a probable local pattern and an appearance probability thereof for each of the partial images;

detection means for detecting at least one feature point from the image received by said image input means;

first calculating means for calculating a quantity of mutual information between each feature point and a current observation point on a basis of appearance probabilities of local patterns for respective partial images including the feature point and the current observation point;

second calculating means for calculating a quantity of visual information for each feature point in accordance with the quantity of mutual information between respective feature points and the current observation point and with a distance between the current observation point and the respective feature points;

second determination means for determining a feature point corresponding to a maximum quantity of visual information as a next observation point; and

moving means for moving said image input means so that the optical axis matches the next observation point.

12. An image information processing apparatus according to claim 11 wherein said moving means performs zooming of said image input means.

13. An image information processing apparatus according to claim 11, wherein said second calculating means has a Hopfield-type neural network to calculate the quantity of visual information in accordance with entropy of said Hopfield-type neural network.

14. An image information processing apparatus according to claim 11, wherein said obtaining means obtains the partial images to the extent of a predetermined depth in a multi-resolution space.

15. An image information processing apparatus according to claim 14, wherein said obtaining means includes discrete means for causing the supplied image to be discrete and a filter for masking the discrete 2D image.

16. An image information processing apparatus according to claim 11 wherein said input means has a wide view lens to transform the coordinates of the received image in a direction perpendicular to said wide view lens.

17. An image information processing apparatus according to claim 11, wherein said second calculating means has a plurality of symmetry-connected-type neural networks for mutually transmitting/receiving signals.

18. An image processing apparatus comprising:

monitoring means for monitoring a supplied image;

detection means for detecting at least one feature point from the supplied image;

extraction means for extracting an image information in a neighborhood of each feature point;

calculating means for calculating an evaluation value of each feature point in accordance with the image information in the neighborhood of a respective feature point and a distance between an optical axis and the respective feature point;

selection means for selecting a feature point one by one, the evaluation value of which is higher than a predetermined value;

moving means for moving a direction of an optical axis to the selected feature point;

acquiring means for acquiring data of an image after moving the optical axis to the selected feature point by said moving means; and

storage means which allots an identifier to the acquired image data so as to store a set formed by a position of the selected feature point, data of the acquired image, time of detection and the allotted identifier.

19. An image information processing apparatus according to claim 18, further comprising:

determining means for determining a feature point, the evaluation value of which is higher than a second predetermined value;

identifying means for identifying an identifier allotted to an image in accordance with the position of the determined feature point; and

second storage means for storing a set formed by the position of the determined feature point, time of determination and the identified identifier.

20. An image information processing apparatus according to claim 18, wherein in a case where a plurality of features have been detected by said detection means, said moving means moves the optical axis toward a feature nearest the present optical axis.

21. A method of controlling an image information processing apparatus, comprising the steps of:

optically receiving an image from an image input unit of said image information processing apparatus;

detecting a feature from the received image;

calculating a quantity of visual information in accordance with a position of the detected feature; and

controlling the image input unit in such a manner that the quantity of visual information is enlarged,

wherein the quantity of visual information is, in said calculating step, calculated in accordance with a quantity of mutual information between different cells in a stochastic automaton containing a plurality of cells.

22. A method of controlling an image information processing apparatus, comprising the steps of:

optically receiving an image from an image input unit of said image information processing apparatus;

detecting a feature from the received image;

calculating a quantity of visual information in accordance with a position of the detected feature; and

controlling the image input unit in such a manner that the quantity of visual information is enlarged,

wherein said detection step has a step of generating a vector field from a function expressing a supplied image and a step in which the vector field is expressed as a function of a complex variable, and then an argument principle is used so as to detect a feature as a singular point.

23. A method of controlling an image information processing apparatus, comprising the steps of:

optically receiving an image from an image input unit of said image information processing apparatus;

detecting a feature from the received image;

calculating a quantity of visual information in accordance with a position of the detected feature; and

controlling the image input unit in such a manner that the quantity of visual information is enlarged,

wherein the quantity of visual information is, in said calculating step, calculated by a plurality of symmetry-connected-type neural networks for mutually transmitting/receiving signals, and

wherein the symmetry-connected-type neural network includes one output neuron block and at least one input neuron block, and

said calculating step includes;

a step of calculating a first information quantity in accordance with an output value from the output neuron block,

a step of calculating a second information quantity in accordance with an output value from the input neuron block and a weight for connecting the input neuron block and the output neuron block with each other, and

a step of calculating the quantity of visual information in accordance with the first information quantity and the second information quantity.

24. An image information processing apparatus comprising:

image input means for optically inputting an image;

detection means for detecting a feature from the image input by said image input means;

calculating means for calculating a quantity of visual information in accordance with a position of the feature detected by said detection means; and

control means for controlling said image input means in such a manner that the quantity of visual information calculated by said calculating means is enlarged,

wherein said calculating means has a stochastic automaton including a plurality of cells to calculate the quantity of visual information in accordance with quantity of mutual information between different cells.

25. An image information processing apparatus comprising:

image input means for optically inputting an image;

detection means for detecting a feature from the image input by said image input means;

calculating means for calculating a quantity of visual information in accordance with a position of the feature detected by said detection means; and

control means for controlling said image input means in such a manner that the quantity of visual information calculated by said calculating means is enlarged,

wherein said detection means has generating means for generating a vector field from a function expressing a supplied image and singular point detection means which expresses the vector field in a form of a function of a complex variable and which uses an argument principle so as to detect a feature as a singular point.

26. An image information processing apparatus comprising:

image input means for optically inputting an image;

detection means for detecting a feature from the image input by said image input means;

calculating means for calculating a quantity of visual information in accordance with a position of the feature detected by said detection means; and

control means for controlling said image input means in such a manner that the quantity of visual information calculated by said calculating means is enlarged,

wherein said calculating means has a plurality of symmetry-connected-type neural networks for mutually transmitting/receiving signals, and

wherein said symmetry-connected-type neural network includes one output neuron block and at least one input neuron block, and

said calculating means includes;

first calculating means for calculating a first information quantity in accordance with an output value from said output neuron block,

second calculating means for calculating a second information quantity in accordance with an output value from said input neuron block and a weight for connecting said input neuron block and said output neuron block with each other, and

third calculating means for calculating the quantity of visual information in accordance with the first information quantity and the second information quantity.