Image analysis – Image transformation or preprocessing – Changing the image coordinates
Reexamination Certificate
1999-07-27
2003-03-25
Mehta, Bhavesh (Department: 2625)
Image analysis
Image transformation or preprocessing
Changing the image coordinates
C382S276000
Reexamination Certificate
active
06539127
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to the field of the processing of pairs of “two-dimensional” (2D) or “three-dimensional” (3D) comparable digital images, and more particularly the registration of a first image with respect to a second image.
BACKGROUND TO THE INVENTION
The term registration is understood to mean determining, from data sets which represent the two comparable digital images, a transformation making it possible to go from one of them, termed the “reference” image, to the other, termed the “floating” image. Stated otherwise, it involves carrying out the “best possible” superposition of the structures contained in two comparable images.
Here, the word “comparable” should be understood within its widest acceptance. It therefore refers, in particular, to so-called “monomodal” images captured with the aid of the same acquisition technique, either over substantially identical regions of the same “object” at different instants, or over substantially identical regions of two distinct “objects”. However it also refers to so-called “multimodal” images captured with the aid of two different acquisition techniques, either over substantially identical regions of the same “object” possibly at different instants, or over substantially identical regions of two distinct “objects”.
PRIOR ART
A number of registration techniques are already known to the person skilled in the art. In particular, this is the case for certain so-called “oriented voxel” techniques which involve determining the registration transformation which maximizes a certain measure of similarity between the superposable voxels of two comparable images. A voxel is the smallest volume element (3D) or surface element (2D) of an image. In general, a datum, from the set of data of an image, is associated with each element (or voxel) making up this image. A datum comprises at least one position component making it possible to identify the corresponding element and one intensity component representing the value at this position of a chosen physical quantity.
Such oriented voxel techniques entail, for example, the WOODS criterion, or the VIOLA criterion, or else the so-called “mutual” information cue. These techniques are not entirely satisfactory, especially when they are used for the registration of multimodal images.
OBJECT OF THE INVENTION
It is a principal object of the present invention to provide an alternative to the aforementioned prior art techniques.
SUMMARY OF THE INVENTION
To this end it proposes an electronic image processing device furnished with a registration module comprising on the one hand a first module capable of calculating a main function representing the correlation ratios (as defined below) between the data of a first data set representative of a first image termed the reference image, and the data of a second data set representative of a second image termed the floating image, and on the other hand a second module having the function of determining a registration transformation making it possible to superimpose the floating image onto the reference image on the basis of the main function calculated by the first module.
The expression “function representative of the correlation ratios” is understood to mean the function from probability theory, which is not usually symmetric, and which for a first data set measures its degree of resemblance to a second data set, independently of the intensity levels of one of the two data sets. With the aim of simplifying the language, in the following description the main function will be regarded as the correlation ratio. Of course, whenever the value of the correlation ratio is to be designated rather than the function which it represents, mention will be made of this.
Moreover, the correlation ratio (in the guise of a function) should be understood both in its conventional acceptance, that is to say accompanied by a Euclidean norm referred to as “order
2
”, and in a very general acceptance in which it is accompanied either by a non-Euclidean norm of any order different from
2
, or by any quadratic form of order
2
.
In a particularly advantageous embodiment, the registration module is furnished with a third module so as to estimate a joint probability density function (or two-dimensional histogram), representative of the intensity of the second (floating) image as a function of the intensity of the first (reference) image, from the first and second data sets, the first module then being arranged so as to calculate the correlation ratio from this joint probability density function (JPDF). This arrangement makes it possible to ease the calculation of the correlation ratio.
According to another characteristic of the invention, mechanical or electronic means are provided which make it possible to designate, out of the two images, the one which serves as reference image (first image) and the one which serves as “floating” image (second image), given that the floating image is the one which is to be registered with regard to the reference image by the registration transformation to be determined.
In a variant, intended to ease the determination of the correlation ratio (in the guise of a function), there is provided a fourth module for transforming the first or the second data set (reference set or floating set) into a third data set comprising data whose position components correspond to each of the position components of the data of the second set (floating set) or of the first set (reference set), respectively. The intensity components of the third and second (or first) sets may thus be compared directly at each voxel, given that the associated position components correspond. According to the variants, the fourth module directly feeds either the first module for the calculation of the correlation ratio, or the third module for the calculation of the joint probability density function.
Advantageously, the transformation of the first (or second) data set into a third data set is performed by applying an initialization registration transformation followed by an interpolation, preferably of trilinear type, to the first (or second) set.
According to yet another characteristic of the invention, the second module determines the main function (correlation ratio) by selecting according to a chosen criterion a maximum of this function from among the maximum or maxima which it possesses, the registration transformation is then determined from this selected maximum.
The registration transformations are preferably chosen from among any n-dimensional geometric transformations, n being chosen at least from the values 2 and 3 according to the dimension of the images.
More preferably still, the registration transformation is determined within the class of affine n-dimensional transformations, and in particular those referred to as “rigid”. However, it could also be determined from among the so-called “non-rigid” transformations, in particular.
In a particularly advantageous variant, the registration transformation is calculated iteratively. To do this, the second module is arranged so as to determine whether the registration transformation which has just been determined from the main function (correlation ratio), satisfies a chosen criterion. In the event of the non-satisfaction of this criterion, the registration transformation having just been checked is addressed to the fourth module with a view to replacing the transformation previously used for obtaining the third data set. A new correlation ratio is then recalculated, followed by a new registration transformation. This operation is then repeated until a registration transformation fulfils the satisfaction criterion, or else until the operation has been repeated a number of times equal to a predetermined value, possibly adaptable.
With the aim of simplifying the calculations, the correlation ratio may be estimated from part(s) of the first and/or second sets which satisfy a chosen criterion, for example of intensity or of belonging to a region.
The invention applies very particularly, although not exclusively
Ayache Nicholas
Malandain Gregoire
Pennec Xavier
Roche Alexis
Inria Institut National de Recherche
Mehta Bhavesh
Patel Kanji
Rabin & Berdo P.C.
LandOfFree
Electronic device for automatic registration of images does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Electronic device for automatic registration of images, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Electronic device for automatic registration of images will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3082533