Pattern similarity metric for image search, registration, and comparison
First Claim
1. A method for comparing a model image with a run-time image so as to provide a quantitative measure of image similarity, the method comprising:
- at train-time,extracting sample pixels of the model image so as to provide a pixel position and a gray value for each sample pixel of the model image;
computing gray value differences between overlapping pairs of sample pixels of the model image;
computing a compact binary representation for each overlapping pair;
computing an expanded binary representation for each compact binary representation;
associating a corresponding pixel position with each expanded binary representation to provide a model element; and
grouping all model elements together to provide a model, and at run-time,determining gray values at a plurality of selected pixel locations in a run-time image in accordance with pixel positions of the elements of the model;
computing gray value differences between overlapping pairs of the selected pixels of the run-time image;
computing a compact binary representation for each overlapping pair of the selected pixels of the run-time image;
computing a run-time bit-expanded representation for each run-time compact binary representation;
comparing each run-time bit-expanded representation with a corresponding model bit-expanded representation to provide a match count increment; and
combining all match count increments to provide a total match value.
1 Assignment
0 Petitions
Accused Products
Abstract
The present invention provides a method for the calculation of a pattern similarity metric that is locally normalized with respect to illumination intensity, and is invariant with respect to rigid body preserving gray scale variations, such as scale, rotation, translation, and non-linear intensity transformations. In one aspect, the invention provides a method for comparing a model image with a run-time image so as to provide a quantitative measure of image similarity. In another general aspect of the invention, a method is provided for searching a run-time image with a model image so as to provide at least one location of the model image within the run-time image.
-
Citations
19 Claims
-
1. A method for comparing a model image with a run-time image so as to provide a quantitative measure of image similarity, the method comprising:
-
at train-time, extracting sample pixels of the model image so as to provide a pixel position and a gray value for each sample pixel of the model image; computing gray value differences between overlapping pairs of sample pixels of the model image; computing a compact binary representation for each overlapping pair; computing an expanded binary representation for each compact binary representation; associating a corresponding pixel position with each expanded binary representation to provide a model element; and grouping all model elements together to provide a model, and at run-time, determining gray values at a plurality of selected pixel locations in a run-time image in accordance with pixel positions of the elements of the model; computing gray value differences between overlapping pairs of the selected pixels of the run-time image; computing a compact binary representation for each overlapping pair of the selected pixels of the run-time image; computing a run-time bit-expanded representation for each run-time compact binary representation; comparing each run-time bit-expanded representation with a corresponding model bit-expanded representation to provide a match count increment; and combining all match count increments to provide a total match value. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18)
-
-
19. A method for searching a run-time image with a model image so as to provide at least one location of the model image within the run-time image, the method comprising:
-
at train-time, extracting sample pixels of the model image so as to provide a pixel position and a gray value for each sample pixel of the model image; computing gray value differences between overlapping pairs of sample pixels of the model image; computing a compact binary representation for each overlapping pair; computing an expanded binary representation for each compact binary representation; associating a corresponding pixel position with each expanded binary representation to provide a model element; and grouping all model elements together to provide a model, and at run-time, determining gray values at a plurality of selected pixel locations in a run-time image in accordance with pixel positions of the elements of the model; computing gray value differences between overlapping pairs of the selected pixels of the run-time image; computing a compact binary representation for each overlapping pair of the selected pixels of the run-time image; computing a run-time bit-expanded representation for each run-time compact binary representation; comparing each run-time bit-expanded representation with a corresponding model bit-expanded representation to provide a match count increment; combining all match count increments to provide a total match value; determining a plurality of total match values at a plurality of positions in the run-time image to provide a total match value space; and evaluating the total match value space so as to find local maxima.
-
Specification