Generalized object recognition for portable reading machine
Techniques for operating a reading machine are disclosed. The techniques include forming an N-dimensional features vector based on features of an image, the features corresponding to characteristics of at least one object depicted in the image, representing the features vector as a point in n-dimensional space, where n corresponds to N, the number of features in the features vector and comparing the point in n-dimensional space to a centroid that represents a cluster of points in the n-dimensional space corresponding to a class of objects to determine whether the point belongs in the class of objects corresponding to the centroid.
1. A method of operating a reading machine, the method comprising:
capturing an image by an image capture device on the reading machine;
operating the reading machine in a mode that performs generalized object recognition, by:
forming by the reading machine an N-dimensional features vector, the features vector comprising a header portion having fields including a field identifying the image used to produce the features vector, a field identifying the object, and a field identifying the class of the object and an array portion comprising fields that correspond to values based on features of the image, the features corresponding to characteristics of the object depicted in the image;
representing by the reading machine the values portion of the features vector as a point in n-dimensional space, where n corresponds to N, the number of features in the features vector;
retrieving centroid points, a tolerance, and identification of the class of objects for the retrieved centroid points;
comparing by the reading machine the point in n-dimensional space to centroid points that represent clusters of points in the n-dimensional space corresponding to classes of objects to determine, which of the classes of objects correspond to the point; and
rendering the reading machine to a user, identification information based on the determined class of objects.
2. The method of claim 1 wherein the image is a received image from an image input device associated with the reading machine.
3. The method of claim 1 further comprising:
populating by the reading machine the fields in the header portion of the features vector to identify the object that appears in the captured image based on the determined class of object; and
normalizing feature data that populates the array portion to account for spatial variations.
4. The method of claim 1 further comprising:
capturing an image or series of images to form a series of arrays.
5. The method of claim 1 further comprising:
retrieving an identification corresponding to a centroid that matches the point in n-dimensional space.
6. The method of claim 1 wherein rendering further comprises:
producing an auditory output to a user to render the identification information based on the determined class of objects.
7. The method of claim 1 further comprises:
receiving a tolerance associated with the centroid; and wherein comparing determines whether the point is within a distance from the centroid that is less than the tolerance associated with the centroid.
8. The method of claim 7 wherein the tolerance is at least one of variance, standard deviation.
9. The method of claim 1 wherein the centroid that represents a cluster of points is stored in a database and includes a value of the centroid and identification information that identifies a class of objects that the centroid represents.
10. The method of claim 1 wherein forming the N-dimensional features array uses the Scale Invariant Feature Transform (SIFT) to transform image data into scale-invariant coordinates relative to local features.
11. A reading machine comprising:
a processing device;
a computer program product residing on a computer readable medium for controlling operation of the reading machine, comprising instructions for causing the machine to:
capture an image by an image capture device on the reading machine;
operate the reading machine in a mode that performs generalized object recognition, by instructions to:
form an N-dimensional features vector, the features vector comprising a header portion having fields including a field identifying the image used to produce the features vector, a field identifying the object, and a field identifying the class of the object and an array portion comprising fields that correspond to values based on features of the received image, the features corresponding to characteristics of the object depicted in the image;
represent the values portion of the features vector as a point in n-dimensional space, where n corresponds to N, the number of features in the features vector;
retrieve centroid points, a tolerance, and identification of the class of objects for the retrieved centroid points;
compare the point in n-dimensional space to centroid points that represent clusters of points in the n-dimensional space corresponding to classes of objects to determine, which of the classes of objects correspond to the point; and
render by the reading machine to a user, the identification information based on the determined class of objects.
12. The reading machine of claim 11 further comprises:
an image input device that captures the received image.
13. The reading machine of claim 11 wherein the computer program product comprises instructions to:
populate the fields in the header portion of the features vector to identify the objects that appear in the captured image based on the determined class of object; and
normalize features data that populates the array portion to account for spatial variations.
14. The reading machine of claim 11 wherein the computer program product comprises instructions to:
retrieve an identification corresponding to a centroid that matches the point n-dimensional space.
15. The reading machine of claim 11 wherein the computer program product comprises instructions to:
produce an auditory output to a user to render the identification information based on the determined class of objects.
16. The reading machine of claim 11 wherein the computer program product comprises instructions to:
receive a tolerance associated with the centroid; and wherein the instructions to compare,
determine whether the point is within a distance from the centroid that is less than the tolerance associated with the centroid.
17. The reading machine of claim 11 further comprises:
a database of centroids that represent classes of objects.
18. A computer program product tangibly embodied on a computer readable storage device for performing object recognition by a reading machine, the computer program product comprising instructions for causing the reading machine to:
receive an image captured by an image capture device on the reading machine;
operate the reading machine in a mode that performs generalized object recognition, by instructions to:
form an N-dimensional features vector, the features vector comprising a header portion having fields including a field identifying the image used to produce the features vector, a field identifying the object, and a field identifying the class of the object and an array portion comprising fields that correspond to values based on features of the received image, the features corresponding to characteristics of the object depicted in the image;
represent the values portion of the features vector as a point in n-dimensional space, where n corresponds to N, the number of features in the features vector;
retrieve centroid points, a tolerance, and identification of the class of objects for the retrieved centroid points;
compare the point in n-dimensional space centroid points that represent clusters of points in the n-dimensional space corresponding to classes of objects to determine, which of the classes of objects correspond to the point; and
rendering by the reading machine to a user, the identification information based on the determined class of objects.
19. The computer program product of claim 18 further comprises instructions to:
populate the fields in the header portion of the features vector to identity the objects that appear in the captured image based on the determined class of object; and
normalize features data that populates the array to account for spatial variations.
20. The computer program product of claim 18 further comprises instructions to:
retrieve an identification corresponding to a centroid that matches the point in n-dimensional space.
21. The computer program product of claim 18 further comprises instructions to:
produce an auditory output to a user to render the identification information based on the determined class of objects.
22. The computer program product of claim 18 further comprises instructions to:
receive a tolerance associated with the centroid; and wherein the instructions to compare,
determine whether the point is within a distance from the centroid that is less than the tolerance associated with the centroid.
23. A reading machine comprising:
circuitry to receive an image; and
circuitry to form an N-dimensional features vector, the features vector comprising a header portion having fields including a field identifying the image used to produce the features vector, a field identifying the object, and a field identifying the class of the object and an array portion comprising fields that correspond to values based on features of the received image, the features corresponding to characteristics of the one object depicted in the image;
circuitry to represent the values portion of the features vector as a point in n-dimensional space, where n corresponds to N, the number of features in the features vector;
circuitry to retrieve centroid points, a tolerance, and identification of the class of objects for the retrieved centroid points;
circuitry to compare the point in n-dimensional space to centroid points that represent clusters of points in the n-dimensional space corresponding to classes of objects to determine, which of the classes of objects corresponds to the point; and
circuitry to render the identification information based on the determined class of objects.
24. The reading machine of claim 23 further comprises:
an image input device that captures the received image.
25. The reading machine of claim 23 further comprises:
circuitry to retrieve an identification corresponding to a centroid that matches the point in n-dimensional space.
26. The reading machine of claim 23 further comprises:
circuitry to produce an auditory output to a user to identify the object class or object associated with the received image.
27. The reading machine of claim 23 further comprises:
circuitry to receive a tolerance associated with the centroid; and
circuitry to determine whether the point is within a distance from the centroid that is less than the tolerance associated with the centroid.
28. The reading machine of claim 23 further comprises:
a database of centroids that represent classes of objects.