JP2000207550A - Image processing method containing space and time noise filtering stage and medical image pickup device for executing the same - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image processing method for reducing noise in a series of images showing a small moving object and preventing the formation of a patch and a noise pattern, which occur owing to the space filtering of the present image in a series of images. SOLUTION: An image processing method for reducing noise in an image contains the decision of time concentration containing two noises on a present in a series of three continuous images where the image to be processed is set to be a center image. The method contains a space/time filtering means for connecting the input of time concentration, the input of space concentration and at least one input of space/time random concentrations for supplying the filtered concentration to the present of the processed image. The method contains a pseudo time branch 10 connecting time input to spatial input and a pseudo space branch 20 which is connected to the branch in parallel and is connected to spatial input to timewise and recursive input. The results of the two parallel branches are connected. The method is applied to a medical X ray image pickup device.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to noise in an image, including the determination of three temporal densities for the current pixel at the same location in a series of three consecutive images centered on the image to be processed. And a method including a spatial and temporal filtering step. The invention also relates to a medical device for performing the method.

The above method is applied to preserve small moving objects and reduce noise in a series of images. Here, a small object is understood to mean, for example, an object of one to several pixels. The present invention
It can be used to process video images, especially medical images.

[0003]

2. Description of the Related Art Image processing methods for reducing noise in a series of images, including spatial and temporal filtering, are described in US Pat.
Proceedings of the 39th Meeting of IEEE Signal Processing, May 1991
Volume 5 issue "Multistage Order" by Gonzalo R. ARCE
Statistic Filters for ImageSequence Processing ''
It is known from.

The cited document describes a method of median filtering at several levels called MOS. According to this method, the outputs of elementary spatial filters operating in a cascaded filtering structure of several levels are combined. It is understood that these elementary filters correspond to structures that extend into the window of the filter. Within the spatial and temporal signal space, it is understood that each elementary filter retains features having substantially identical gray levels in a given direction (dimension). If a sufficient number of elementary filters are used, features oriented in any direction may be retained by the filters. The type of MOS basic filter is determined by the type of feature to be retained. If the feature is a unidirectional segment in a three-dimensional spatial and temporal space (cube), the MOS filter is called a unidirectional multilevel filter because the basic filter is unidirectional. If a feature has segments that extend in two orthogonal directions, one in space and the other in time,
The basic filter is bidirectional and the resulting MO
The S filter is called a two-way multi-level filter.
In another variation, the generalization of the class of filters described above is obtained by providing some smoothing control by varying the weight of the center of the window of the filter.

[0005]

According to the cited documents, these MOS filters do not include the movement of displacing the object between one image and another in a series of images. It is sometimes understood that it allows for the recovery of details on the order of pixels. The MOS filter keeps the structure of the signal without considering the movement of the object. The strength against noise is
Therefore, it depends on the width of motion in a series of images.

[0006]

SUMMARY OF THE INVENTION The present invention reduces the noise in a sequence of images representing small moving objects, and generally includes patches and patches resulting from spatial filtering of the current image in such sequence. An object of the present invention is to provide an image processing method for preventing formation of a noise pattern.

The present invention also provides a method of avoiding the dilemma between removing noise patterns involving problems with retaining moving objects and preserving moving objects involving problems with residual noise patterns. The purpose is to provide. The above objects and problems are achieved and solved by an image processing method for reducing noise in an image according to claim 1.

This filtering method has the advantage that it can be applied when the sequence of images contains very small objects subject to very large movements. In general, this filtering method can also work in particularly difficult situations for the movement of objects in a sequence of images. For example, this filtering method can be used in an image processing system included in a medical X-ray imaging apparatus that operates in a fluoroscopic mode. In fluoroscopy mode, a series of images contains only a few images per second, and the images are formed using very low doses of x-rays so are very noisy and two consecutive images The movement of an object between one image and another in a series of images is very large due to the relatively long period of time that elapses between the formation of It is a device such as a small catheter.
Even in these very difficult situations, very good results are obtained with the method of the invention. Thus, the method of the present invention provides that the noise-suppressed image is derived from a series of fluoroscopic images so that the image to be processed corresponds to an image requiring four times the x-ray exposure time required for the image to be processed. Can be extracted.

Another advantage of the present invention is that it is not necessary to know a priori the variable of the noise type relative to the average noise, ie the standard deviation. To implement many other noise reduction methods, it is necessary to know the standard deviation. Another advantage of the present invention is that no initial conditions are required. The invention can be applied directly to a sequence of images.

A medical imaging device comprising means for performing the method according to the invention is defined in claim 10.

[0011]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below in detail with reference to the accompanying drawings. FIG. 1A shows a simplified image processing method for reducing noise in a series of images and is shown in the form of a series of functional blocks, and FIG. 2 shows the same method in more detail. FIG. First, the method of the present invention comprises:
Three consecutive images that are inherently noisy and have not previously been smoothed in terms of noise peaks, namely J _t−1 , J _t , J
capturing (100) _{t + 1} . These images are not smoothed because such smoothing is not necessary since the method performs a complete filtering of the noise of the images themselves. Each captured image is causal
(CAUSAL) image, current (PRESENT) image and anti-causal (ANTIC
AUSAL) image and is shown in FIG. According to the method of the present invention, the current image _Jt is processed to achieve noise reduction by its own density data and by the density data of the causal image _Jt-1 and the anticausal image _{Jt + 1.} . For processing, consider a pixel P (x, y) with spatial coordinates x, y in the image system X, Y for all images, and also the density It _-1 (x, y) in the causal image, concentration I _t (x, y) of the current image and the density _{I t + 1 (x, y} ) in the anti-causal image consider. For simplicity, hereafter these concentrations and I _t-1, I _t and I _{t + 1.}

Referring to FIG. 1A, the basic functions used are shown, and it can be seen that the method is based on three important principles. According to the first principle, the method is performed according to two main branches. The two branches are a first branch 10 that performs substantially temporal filtering, ie, a pseudo-time branch, and a second branch 20 that performs substantially spatial filtering, ie, a pseudo-spatial branch.

According to this first principle, any branch 1
Neither 0 nor 20 performs pure temporal filtering or pure spatial filtering. Each branch 10 or 20 is 3
One of providing a balanced noise suppression in a series of central image J _t of the successive images. The filtering method involving these two branches 10 and 20 includes, between the pseudo-space branch and the pseudo-time branch, a series of images including moving objects or objects having lost motion. Establishes a continuous movement.

According to a second principle, the method consists of operations to prevent patches or noise patterns that would appear due to pure spatial filtering. Referring to FIG. 2, according to a second principle, these operations use a mechanism for random selection of pixels 111, 112, 113 in a series of images, which is purely a random selection mechanism. Not spatial. Indeed, such a random selection mechanism is not suitable to achieve a sufficient level of noise suppression if it is purely spatial, or leaves the noise pattern present, or removes the noise pattern. From time to time, if the mechanism reaches a sufficient noise suppression level and succeeds in removing the noise pattern, noise suppression corresponding to the density plane due to motion between the two images is left.

According to a third principle, the method is based on two types of pixels distinguished in a sequence of images. One type is the so-called reliable pixels, are arranged from the current pixel P to be processed in the current image J _t (x, y) in a short distance, is given by calculating 121, 122, 123. Other types are not necessarily reliable pixels and may be located in the causal and anti-causal images, or may be located in the current image but compared to the current pixel P (x, These pixels are arranged at a long distance from y) and are generated by the operations 111, 112, and 113.

Second, referring to FIGS. 1A and 2, in order to implement the present invention based on the above-mentioned important principles, the present invention employs filtering in two parallel operations 10 and 20. Including stages. This stage comprises a first pseudo-time branch 10 and a second pseudo-space branch 20, these 2
One of the following operations are median filter 140 is followed by the input of the median filter 140 receives density I _t of the current image J _t represented by two outputs 8,9 and reference number 1 from branch 10 and 20.

[0017] Pseudo Time branch 10, whereas in the image _{J t-1, J t,} J t + 1 concentration of each obtained from the _{I t-1, I t,} I t + 1 of three consecutive in Three time inputs 1, 2,
3 On the other hand, this pseudo-time branch 10 has a spatial input 7 which is evaluated on the basis of reliable pixels. These reliable pixels are used for small objects or small details,
That is, it is detected by the operations 121, 122 and 123 executed when it is assumed that a very short segment is to be detected.

Referring to FIG. 3A, during this operation, a given number of directions .theta. Is analyzed using a filter having a linear support region, on which the support region is analyzed. Is formed. A system of centered axes having a direction indicated by a discontinuous angle θ measured with respect to a reference axis, for example the axis X of the abscissa x, is defined. The discontinuous angle θ is θ ₁ = 0, θ ₂ = π / 4, θ ₃ = π /
It takes different values regularly distributed in space, such as 2, θ ₄ = 3π / 4. The axes located in opposite directions are
It is shown as the same angle notation plus π, such as θ ₁ + π, θ ₂ + π, θ ₃ + π, θ ₄ + π. In this axis system centered on the current pixel P (x, y), the support regions of the linear filter are arranged in various discontinuous directions, denoted by θ or θ + π.

For the detection of small details, the length of the support area or mask is preferably chosen to be two pixels. In fact, the larger the support area, the lower the possibility of detecting fine details. If the size of the support area or mask is increased, for example by using three or four aligned pixels, it will be necessary to increase the number of directions of the axis system. Thus, while noise suppression is increased, there is a risk of breaking details and computational costs are also increased. Referring to FIG. 3A, the mask is located at some distance from the center pixel, so that the spatial correlation of the noise is reduced, and thus better noise suppression may be achieved. For example, a circle of one pixel is left. Thus, the short distance for reliable pixel detection is two or three pixels with respect to the center pixel.

[0020] Reliable operation for detection pixel is currently performed for the image J _t. This operation involves the selection of a particular direction indicated by the angle value α of the various angles θ and θ + π of the axis system. This particular direction, the average concentration level calculated for the mask support region of the linear filter disposed to the direction α is chosen as close to the density level I _t of the current image P (x, y) as possible .

For this reason, for all the filter-supported masks arranged in all directions θ of the axis system,

[0022]

[Outside 7]

Is obtained. Subsequently, when θ is changed, the following equation:

[0024]

(Equation 1)

The specific direction corresponding to the angle θ at which is minimum is determined, and α is referred to as the corresponding specific angle θ. Formed in the support region of the linear filter, the above equation (1)
Minimize

[0026]

[Outside 8]

The specific value of is

[0028]

[Outside 9]

[0029] Referring to FIG. 3B,
After determining the direction α as described above, one of the two reliable pixels, i.e., one of the respective support regions corresponding to the above-mentioned average.
Operations 121 and 122 are performed to select one pixel. The first reliable pixel selected by operation 121 is located in the support region of the filter located in direction α.

[0030]

[Outside 10]

And the second reliable pixel selected by operation 122 is the pixel in the support region of the filter located in the opposite direction α + π.

[0032]

[Outside 11]

A pixel having The method also includes a supplementary selection performed by operation 123 to determine which of the two previously selected reliable pixels has the minimum density. This minimum concentration is

[0034]

[Outside 12]

Is shown. In many applications of the method of the invention, the object is reproduced dark on a light background,
It is correct to select a reliable pixel from the pixels of the minimum density. This applies to applications such as medical imaging, radiography, and fluoroscopy. In these applications, the aim is to always hold small, dark objects. If the image to be processed represents a bright object on a dark background, the reliable pixels are determined among the pixels with the highest density. Obviously, for television applications, it is not necessary to search for the reliable pixel or point from the darkest pixel or point. In that case, the pixel or point having the density closest to the current pixel or point is simply searched.

[0036] Referring to (A) and FIG. 2 FIG. 1, the pseudo-time branch 10, three time indicated by 1, 2, 3 in the input density _{I t-1, I t,} I t + 1 And indicated by 7

[0037]

[Outside 13]

And an averaging operator 131 which combines Hereinafter, the above operation will be analyzed. Referring to FIG. 4A, the movement of a small object OBJ present in the image between J _t−1 and J _{t + 1} is illustrated. Three successive input images _{J t-1, J t,} J t + 1 in unless caused motion of the object OBJ between the concentration _{I t-1, I t,} 3 one time, which is formed by the I t _{+ 1} input 1, 2 and 3 all three have substantially the same value, this value because substantially equal to the concentration I _t of pixel currently said to be "correct". On the other hand, reliable pixels

[0039]

[Outside 14]

The input 7 formed by
Always "correct" according to the principle of pixel selection. Previously
The trusted pixel issued is the current pixel to be processed.
Concentration I _tAre arranged in the direction α having the same concentration as
Therefore, it is declared reliable. This looks like
Can be summarized. The pseudo time branch 10 is a time image
J_{t- 1}, J_t, J_{t + 1}Three images of the time origin obtained from
From elementary and purely spatial detections 121,122,123
The resulting so-called reliable pixels are processed. No movement
The pseudo-time branch 10 is thus suitable noise suppression
And appropriate time integration.

Referring to FIG. 4B, the object OBJ
Movement occurs between three consecutive images J _t−1 , J _t , J _{t + 1} , and when considering the so-called “plateau movement”, either density I _t−1 or density I _{t + 1} is considered. or obviously now it deviates from the concentration I _t of pixel. The corresponding time input 2 or 3 of the pseudo time branch can be said to be "wrong". The other three inputs are said to be "correct" and the result of the average operator represents a small distortion caused by only 75% of the inputs being valid.

Referring to FIG. 4A, the object OBJ
If the motion of the object occurs between three consecutive images J _t−1 , J _t , J _{t + 1} , and if the furtive motion of small objects (passing motion) is taken into account, the density it may occur that I _t-1 and the concentration I _{t + 1} deviates from the concentration I _t apparently simultaneously the current pixel. The two corresponding time inputs 2 and 3 of the pseudo time branch 10 are "wrong." Only the two inputs 1 and 7 of branch 10 are "correct". In this case, since only 50% of the inputs are valid, the averaging operator 131
The result is unreliable. These errors are corrected by the pseudo-space branch 20.

Referring to FIGS. 1A and 2, the pseudo-space branch 20 first has three reliable spatial inputs 1, 4 and 5, namely a current pixel density I _t and a direction α. Pixel

[0044]

[Outside 15]

And the reliable pixel in the opposite direction α + π

[0046]

[Outside 16]

[0047] However, by those skilled in the art,
It will be appreciated that purely spatial filters are likely to produce patches and noise patterns. Thus, according to the method of the invention, this risk is eliminated by providing the pseudo-spatial branch with a fourth input 6 which is not purely spatial.

Referring to FIG. 5, the formation of this input 6 is based on a random selection mechanism and has a recursive nature. This fourth input 6 is called R _t, and the current image J _t
Is the density of a pixel randomly selected around the current pixel P (x, y). From “around the current pixel”, R
_t is understood to be the concentration of the pixel selected in the vicinity V _t around the current pixel. Neighborhood V _t is e a square dimension of the pixels on both sides in the 6 or 8 Current pixel, the current image is excluded from the vicinity V _t. Because of the selected neighborhood dimensions, random pixels are located farther from the center pixel than trusted pixels. In general, the density R _t is the density of random pixels in the neighborhood V _t , but not the density of reliable pixels.

For this reason, a temporal concept and a recursive concept are introduced and applied to R _t . According to the method of the invention, the value R _t-1 has already been evaluated for the causal image J _t-1 . Therefore, referring to FIG. 2, when the image J _t arrives, the values R _t−1 and R _t are held by the operations 111 and 112 according to the principle described above, respectively. Subsequently, in operation 113, the density 6, which is the fourth input 6 to the pseudo space branch 20, that is,

[0050]

(Equation 2)

Is selected. If the image contains a bright object on a dark background, MAX instead of MIN in equation (2)
Is substituted. The error of the pseudo time branch 10 occurs in the case of stealthy or passing movement of a small object OBJ,
It is correct to select a random pixel at the time origin to provide a fourth input 6 to the pseudo-space branch 20. This error cannot be compensated for by the input formed by the density of the pixels near the center pixel, because the density of the pixels near the center pixel is very different in the current image processed by the pseudo-space branch. This error may be in the vicinity of the average dimensions, but sometimes in images other than the current image, ie even in a counter-causal image where it is more likely to encounter small moving objects OBJ. Can also be compensated only by the density of pixels located far from the current pixel.

The four inputs 1, 4, 5 of the pseudo-space branch 20
And 6 are the pixels of such a reliable pixel.

[0053]

[Outside 17]

And the pixel density MIN (R _t−1 , R _t−1 ,
R _t ). These four density values (three of which are always correct) are applied to the inputs of the median 132, including the four inputs. Under the influence of the median 132, the four concentrations represented are classified by value,
The two densities with intermediate values are averaged. Median 132
Gives at output 9 a density value located between the other two remaining densities. The median 132 has a property of removing an abnormal value and outputting a median value as a result. That is, the incorrect input of the three inputs applied to the median is automatically removed and may not be relevant to the result.
Therefore, the median result of the pseudo-space branch 20 is 100
%correct. If the results are not as accurate as possible and are some distance from the best density, remove the patches and noise patterns from the image.

Third, an operation for combining by the output median 140 will be described. FIG. 1A and FIG.
Referring to the concentration I _t of the current pixel in the current image J _t indicated output 8, the output 9 of the pseudo-spatial branch 20, and by the reference numeral 1 in the pseudo-time branch 10 is referred to as the output median Applied to the three inputs of the median 140.

[0056] Two of these three inputs 1,8 and 9, i.e., the output 9 an output of the input 1 and the pseudo-spatial branch 20 is the concentration I _t of the current pixel is always right. As a result, STRMI _t called output 40 of the output median 140 is always 100% correct on the nature of the median. The method according to the invention is performed automatically by a standard and systematic scan of the noisy raw image, each pixel undergoing all steps of the method. For example, the image can be scanned from left to right and from top to bottom, as is known by those skilled in the art. Each pixel being processed is called a current pixel.

The method according to the invention essentially comprises the determination of three temporal densities for the current pixel at the same location in a series of three consecutive images centered on the image to be processed, Input of the concentration of the property (1, 2, 3);
A spatial and temporal processing means (1) for combining an input of spatial property density (4, 5, 7) and at least one input of spatial, temporal and random property density (6);
Comprising filtering step for providing the processed image (filtered density current for the pixels of J _t) (40) by 0,20).

In general, the method of the invention comprises a pseudo-spatial processing branch 20 arranged in parallel with the pseudo-time processing branch 10 and means 140 for combining the results 8 and 9 of these branches. The pseudo time branch 10 comprises a majority input 1, 2, 3 which is the density selected on the time axis t-1, t, t + 1 considered for the current pixel having the coordinates x, y, and Means 121, 12 for the selection of reliable pixels in
2, means 131 for combining with a small number of inputs 7, which are the concentrations of the reliable pixels provided by 2,123.

The pseudo-space branch 20 comprises a majority input 1,4,5 which is the density of reliable pixels provided by the means for the selection of reliable pixels in the current image, and a temporal and recursive input. Mechanism 1 for random selection with different properties
Means 132 for combining with a small number of inputs 6 the pixel densities obtained by 11, 112, 113.

The means for the selection of reliable pixels are described only by way of example. Many other means for reliable pixel or point selection are known to those skilled in the art. Reliable pixels are, for example, providing a threshold for the density level of pixels in the vicinity of the current pixel in the current image, and that pixels having a density close to the density of the current pixel within the threshold limit are reliable pixels. Can be selected by declaring.

Thus, the selection of reliable pixels is based on a given probability criterion, eg, an approximation of the density level, at a short distance in the mask or neighborhood that is most likely to belong to the same object. This can be performed by selecting a pixel. The method of the present invention can be performed by a simple system. This type of system consists of a means for the selection of reliable pixels and random pixels of recursive nature and three combining means, preferably the ME in the pseudo-space branch.
Includes DIAN-4, a 4-input averaging operator in the pseudo-time branch, and a median for combining the outputs of the pseudo-space and pseudo-time branches.

The spatial combination can be performed using an averaging operator, a linear combination operation device, etc. instead of using a median. FIG. 6 shows, by way of example, a medical imaging device including digital radiography provided with means for implementing the noise reduction method described above. The apparatus includes an X-ray source 31, a patient mount 32, a device for receiving X-rays passing through the patient, and an image intensity coupled to a camera tube 34 for providing data to an image processing system 35 including a microprocessor. And a fire device 33. Image processing system 3
5 has a plurality of outputs, one output 36 of which is coupled to a monitor 37 for displaying the processed image or sequence of images.

The digital radiographic image has 8 bits or 1 bit.
512 × 512 or 1024 × encoded with 0 bits
It may include 1024 pixels. Each pixel can thus be assigned a density level from 1 to 256 or 1024. For example, dark areas of the image have low density levels,
Light areas can have high density levels. Digital images can be captured in fluoroscopy mode. The invention can be used in particular for processing arterial photographic images.

The various steps and operations of the digital image processing method described above can be implemented in system 35. Data may be stored in a memory area (not shown). Recording means (not shown) can also be used.

[Brief description of the drawings]

1A shows in block diagram form the functions used in the method of the invention, and FIG. 1B shows the capture of three consecutive time images.

2 shows a sequence of steps of the method of the invention in more detail in the form of a block diagram in comparison with FIG. 1 (A).

3A is a diagram showing a stage of searching for a reliable pixel, and FIG. 3B is a diagram showing a stage of detecting a reliable pixel.

FIG. 4A is a diagram illustrating a case where a series of images represents a small moving object, and FIG. 4B is a diagram illustrating a case where the series of images is a moving object having a plateau shape. It is.

FIG. 5 is a diagram showing a random pixel determination.

FIG. 6 shows a medical X-ray apparatus including means for processing a digital image.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 10 Pseudo-time branch 20 Pseudo-space branch 31 X-ray source 32 Patient mount 33 Image intensifier device 34 Camera tube 35 Image processing system 36 Output 37 Monitor 131 Average operator 132 Median 140 Median

Continuation of the front page (71) Applicant 590000248 Groenewoodseweg 1, 5621 BA Eindhoven, The Netherlands (72) Inventor Christel Swye France, 75005 Paris, Boulevard de l'Opital 14 F-term (Reference) 5B057 AA08 CE06 CH09 5C021 PA39 PA40 PA56 PA62 PA66 PA76 RB03 RB08 RC01 SA24 YA01 YC08 ZA00

Claims (10)

[Claims] An image for reducing noise in an image, comprising determining three temporal densities for a current pixel at the same location in a series of three consecutive images centered on the image to be processed. A processing method, comprising: inputting a density of a temporal property (1, 2, 3) and providing a density of a spatial property to provide a filtered density (40) for a current pixel of the processed image (J _t ). A spatial and temporal processing means (10, 20) for combining the density input (4, 5, 7) and at least one density (6) input of a spatial, temporal and random nature; A method comprising a spatial and temporal filtering step. 2. The majority of the inputs of the concentration of the temporal nature (1,
2,3) and a small number of inputs (7) of spatial property density, a first processing branch (1) called pseudo time branch
0) and the majority input (1,4,5) of the spatial property density and the minority input (1,4,5) of the density of the spatial, temporal and random property are connected in parallel to the first branch. 6) and
A second processing branch (20), called a pseudo-space branch, and the results of the two parallel branches (8, 9) are combined and filtered for the current pixel of the processed image (J _t ). Means (140) for producing a determined concentration (40). 3. The majority of the inputs (1, 2, 3) in the pseudo time branch (10) are on the time axis (t-1, t, t + 1) of the series of images considered at the current pixel. concentration selected _{(I t-1, I t} , I t + 1) a small number of inputs (7)
Is substantially close to the density of the current pixel and is said to be reliable and is spatially selected in the image to be processed (J _t ). And the majority of the inputs (1,4,4) in the pseudo-space branch (20)
5) is substantially close to the density of the current pixel and is said to be reliable and is spatially selected in the image to be processed (J _t ) And a small number of inputs (6) are selected by the mechanism of random and recursive features among the images preceding the image to be processed and the images to be processed (J _t−1 , J _t ). 3. The method of claim 2, wherein the density is a pixel density. 4. The pseudo-time branch (10) comprises an average operator (1
31) form a connection of its inputs (1, 2, 3, 7) and the pseudo-space branch (20) forms a median filter (1).
32) form the combination of its inputs (1, 4, 5, 6) and the outputs (8, 9) of the two branches are the current pixel P
(X, y) is coupled by a median filter (140) having an output for providing attributes and filtering to be concentrations of (STRMI _t) (40), The method of claim 3. 5. If the object is represented in the image as a dark object on a light background, a random and random number of pseudo-space branches (20) for performing temporal selection of a small number of densities (6). Mechanisms of recursive nature (111, 112, 11
3) takes into account the fact that the neighborhood has already been determined during the processing of the preceding image (Jt _-1 ) in the preceding image (Jt _-1 ) and the current image in the image ( _Jt ) to be processed. Determining a neighborhood (V _t ) around the current pixel, excluding [P (x, y)]; random pixels are already determined during processing of the preceding image in the preceding image (J _t-1 ) Determining (111, 112) a random pixel in the neighborhood of the temporal image to be processed, taking into account that the preceding image and the two images obtained from the image to be processed. Determining the density (R _t−1 , R _t ) corresponding to the random pixel; and determining the lowest density [MIN (R _t−1 , R _t )] of the two random pixels from the two densities. Selection stage (11
Or if the object is represented as a light object on a dark background, the mechanism may select the highest density instead of the lowest to perform the same operation. Including,
A method according to claim 3 or claim 4. 6. If the object is represented in the image as a dark object on a light background, a small number of reliable densities (5) in the image (J _t ) to be processed in the pseudo time branch (10). , 7) comprises the steps of defining a support region of a linear filter which is distributed and distributed and arranged along an axis centered on the current pixel; and a filter which is diametrically opposite. Is called a reliable pixel in the support area of (121, 122) determining two pixels with the following, ie, one pixel for each support region: From [outside 5] (123) and as spatial inputs (7) for pseudo-time branches Or, if the object is represented as a light object on a dark background, selecting the highest density instead of the lowest density and performing the same operation. , Claim 5
The described method. 7. The random pixels are selected such that the spatial distance between the random pixel and the current pixel is greater than the spatial distance between the trusted pixel and the current pixel. 7. The method of claim 6, wherein the method is performed. 8. The support region of the linear filter consists of two pixels separated from the current pixel by a circle of one pixel, the axes being separated and distributed by forming an angle of π / 4 therebetween, and the neighborhood ( V _t) is 6 pixels to ± 8 pixels ±, the method of claim 7 wherein. 9. The method according to claim 1, wherein the image is scanned automatically and standardly on a pixel-by-pixel basis, and the steps are performed automatically for each pixel. 10. A system for capturing a series of digital images, and a system for image processing having access to image data and a display system of the capture system. A medical imaging device comprising a processor for performing the described method.