Radiant energy – Photocells; circuits and apparatus – With circuit for evaluating a web – strand – strip – or sheet
Reexamination Certificate
1998-12-16
2001-07-31
Le, Que T. (Department: 2878)
Radiant energy
Photocells; circuits and apparatus
With circuit for evaluating a web, strand, strip, or sheet
C382S284000, C382S255000, C382S294000, C358S450000
Reexamination Certificate
active
06268611
ABSTRACT:
FIELD OF THE INVENTION
This invention relates to registration (alignment) of images and more particularly, to a system for measuring similarity between images (similarity metric) which allows for robust registration of dissimilar images, that is, images in which some of the details are not identical to both images.
BACKGROUND OF THE INVENTION
Registration is useful for many reasons: for example, to compare two or more images obtained at different times, or to align images taken by sensors having different spectral responses, such as sensitivity to different coloured lights.
One type of registration combines the two images to be aligned into a composite image to obtain information as to how to better align the images. Such a composite image can be processed by considering the following: (1) the feature space (what features within the images to look for) ; (2) the search space (how one image can be ‘moved’ relative to the other); (3) the search strategy (in what order to ‘move’ one image relative to another to better align the images); and (4) the similarity metric (the measure of how closely the images are aligned).
Conventional registration systems often use features. By using the features in an image, the whole image does not have to be used for alignment. Instead, significant details, for example, edges of objects in the image can be extracted and aligned. The feature space defines the features which are extracted. Depending on the application, using features for registration is robust. However, registration using features requires prior knowledge about what kind of image can be expected so that the important features will be included in the feature space. In addition, if the features can not be detected reliably, special features (markers), may have to be added to the objects or to the images in order to provide reliable registration. Also, in many applications, the images to be registered are dissimilar, that is, the features are not quite the same. These dissimilarities can, for example, be caused by: changes between exposures in the depicted object; or use of different sensors to obtained images, for example, images captured from the red chip and the green chip in a 3-chip CCD (charge-coupled device) camera or images captured from sensors responding to infrared light and visible light.
The search space defines what kind of geometric transformations can be used to align the images. Examples of geometric transformations include: (1) translation (shifting); (2) rotation, and (3) relative magnification.
The search strategy is used to identify (often iteratively) the parameters of the geometric transformation required to align the images. Typically, the search strategy optimizes a similarity metric, that is, a measure of the similarity between images. If the similarity metric measures differences between images, the desired geometric transformation can be obtained by minimizing the measure. Similarly, if the similarity metric measures the sameness of images, the desired geometric transformation can be obtained by maximizing the measure. Some similarity metrics are calculated using: (1) correlation, that is, the summing of those portions from one image which are the same or similar to the corresponding portion of the other image after the portions have been normalized, weighted, statistically adjusted, or phase adjusted; (2) summing of absolute differences of: the intensity of portions of the images, contours in the images, or surfaces in the images; (3) matched filters (similar to correlation); and/or (4) summing of sign changes between portions of a difference image, that is, for example, the resulting image when one image is subtracted pixel-wise from the another image.
However, without feature extraction, only the ‘summing of sign changes’ (sign summing) has been found to be especially useful for dissimilar images. The other similarity metrics: (1) are sensitive to the dissimilarities; (2) require features; (3) are sensitive to noise; and/or (4) are sensitive not only to dissimilarities but also to changes in illumination of the object. Some implementations of sign summing can take advantage of a characteristic of certain noise distributions in certain images wherein the number of sign changes in a pointwise intensity difference image have a maximum when the images are perfectly aligned, but such implementations require knowledge about the type of noise distribution. Also, if the noise distribution is modified (to fit within certain parameters for sign summing), the speed of the registration can be relatively slow.
SUMMARY OF THE INVENTION
This invention relates to systems (devices and procedures) for solving the problem of feature-free registration of dissimilar images by maximizing a new measure of similarity, that is, a measurement of the sharpness of the focus of the composite image. The maximization is performed by transforming (‘moving’) one image relative to another image until the sharpest focus for the composite image is achieved.
An important aspect of this invention is that no features are calculated. In other words, all of the pixels of the chosen images are used. Accordingly, no knowledge about any feature is needed and the system can be applied generally. In addition, by not requiring features, quick iterations using the whole images can be performed and the system is statistically more efficient since all available information is used.
One goal of this invention is to apply image content autofocusing technology to the field of image registration.
An object of this invention is to provide a system for obtaining registration information to align a reference image with an unregistered image having the steps of: combining the reference image and the unregistered image to generate a composite image; and determining alignment parameters for at least one of the reference image or the unregistered image from the composite image, the alignment parameters selected to correspond to a focus or optimal focus for the composite image. The system can also include the steps of: determining an energy level from the composite image; and selecting the alignment parameters in accordance with the energy level. In addition,the system can also include the steps of: filtering the composite image using a focus filter to generate a filtered image; determining an energy level from the composite image; and selecting the alignment parameters in accordance with the energy level. The focus filter can be a Wiener filter, a modified second order difference filter, and a bandpass filter.
Another object of this invention is to provide a system for aligning a reference image with an unregistered image having: a transformer for receiving the unregistered image and a transformation, and for generating a transformed image; a compositor for combining the reference image and the transformed image to generate a composite image; an energy calculator for receiving the composite image and for determining an energy level of the composite image; and a transformation generator for receiving the energy level and for generating the transformation in accordance with the energy level such that the transformation is selected to correspond to a focus for the composite image. The transformation generator can include a maximizer for determining which parameters for the transformation result in an optimal focus. The transformer can include an interpolator for increasing resolution of parameters of the transformation. The system can also include a filter calculator for modifying the composite image such that the noise of the composite image has been reduced, the noise being reduced by increasing energy contributions from selected parts of the composite image which contribute a relatively larger proportion of alignment information than noise, and by decreasing energy contributions from other parts of the composite image which contribute a relatively larger proportion to noise than to alignment information. The filter calculator can include a convoluter; and a filter selected from the group consisting of: a Wiener fi
Almers Martin
Heyden Anders
Pettersson Magnus
Rosenqvist Anders
CellaVision AB
Cooper & Dunham LLP
Le Que T.
Luu Thanh X.
LandOfFree
Feature-free registration of dissimilar images using a... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Feature-free registration of dissimilar images using a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Feature-free registration of dissimilar images using a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2482467