CN1825354A - Method and apparatus for efficient computation of morphology operations - Google Patents
Method and apparatus for efficient computation of morphology operations Download PDFInfo
- Publication number
- CN1825354A CN1825354A CN 200510121655 CN200510121655A CN1825354A CN 1825354 A CN1825354 A CN 1825354A CN 200510121655 CN200510121655 CN 200510121655 CN 200510121655 A CN200510121655 A CN 200510121655A CN 1825354 A CN1825354 A CN 1825354A
- Authority
- CN
- China
- Prior art keywords
- pixel
- image
- computing
- structural element
- operand
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
Disclosed is an algorithm for applying a morphological operation to an image. In one embodiment, the morphological operation is iteratively applied to a focal pixel of the image and to another pixel of the image. The other pixel is located at an offset with respect to the focal pixel. The offset is based on an operation count. In another embodiment, the algorithm includes performing a morphological operation on an image using a convex structuring element. A work structuring element having dimensions corresponding to the outer-most dimensions of the convex structuring element is iteratively applied to the image. The dimensions of the work structuring element are then adjusted to correspond to the remaining outer dimensions of the convex structuring element not yet covered by the previous work structuring element. The applying and adjusting steps are repeated until a predetermined number of morphological operations have been performed.
Description
Technical field
The application requires the rights and interests of No. the 60/637653rd, the U.S. Provisional Application submitted on Dec 20th, 2004, and its full content is quoted at this as a reference.
Background technology
The present invention relates generally to Flame Image Process, more specifically, relate to the calculating of optimizing the morphology operations in the binary and grayscale application.
Mathematical morphology be applied to various Digital Image Processing such as computer vision and medical imaging are used (as segmentation and skeletonizing task) use among ().For example, the medical worker can carry out CT (computer tomography) (CT) scanning to patient.CT scan uses X-ray equipment to obtain view data from the different angles around the human body, handles these data then to show the sectional view of tissue and organ.Then, thereby can analyze the outstanding specific part of this image by the method for using morphology operations, so that it is radiologist (or other medical workers) can more easily diagnose the state of an illness relevant with patient, unusual etc. as cancer, angiocardiopathy, infectious disease, wound and muscle and skeleton.
By being applied to image, morphological operator (being also referred to as structural element (SE)) carries out such graphical analysis (()).When carrying out tracheae or segmentation of blood vessels, the SE of different proportion can be used for a step.Skeletonization method calculates the intermediate shaft of given object according to the repeated application of morphological operator.
For structural element is applied to input picture, travels through structural element and carry out computing along input picture.Basic morphology operations is to expand and corrosion.Expand by calculating by the logic OR that faces the territory of structural element (be also referred to as and filter nuclear) definition and two-value being carried out at the center that this value is distributed to operator about each pixel/voxel () of input picture.And carry out the two-value corrosion by calculating by the logic AND computing of facing the territory of structural element definition about each pixel/voxel of input picture.In grayscale image, logic OR/AND is replaced to the MAX/MIN operator to produce gray level expansion/corrosion.Logic AND/OR operator is called comparison operator with the MAX/MIN operator.
Fig. 1 illustrates 3 * 3 squares the structural element 100 that is applied to input picture 104, and the computing of execution is that two-value expands, and carries out this computing about each pixel of input picture 104 by the logic OR () that faces the territory that calculates by structural element 100 definition.Expand for carrying out, the center pixel 108 of SE 100 is placed on each pixel of input picture 104, and when OR operator non-zero, SE 100 replacement input picture pixels.In order to distinguish, which pixel of structural element 100 regulation input pictures 104 participates in the OR computing.Thereby the expansion operator produces the output image 112 that comprises than input picture 104 more non-zero pixels.Pixel 116 expressions of line are as the pixel that the result added of computing, and blank pixel 118 is then from input picture 104.State is the pixel 120 of " break " (promptly have be zero binary value and be not the pixel of the ingredient (being that state is the pixel of " leading to ") of image) encirclement image 112.
O Fig. 1 also shows the example that the two-value corrosion is applied to identical square of SE 100 of identical input picture 104 down.As mentioned above, the logic AND computing is used in the two-value corrosion between structural element 100 and input picture 104.Output image 120 comprises than input picture 104 " leading to " pixel still less.In this example, owing in the input picture that operator is surrounded by image a position is only arranged, therefore remaining is single pixel, thereby produces the true value of AND computing.State surrounds image 120 for disconnected pixel (that is, having zero binary value).
The calculating strength of directly using these morphology operations is very huge.Specifically, computing machine directly is directly proportional with the quantity and the type of the calculation operations of carrying out required computing time, and computing and SE and long-pending being directly proportional of image size.For in time obtaining the result, high-efficient algorithm is necessary.Otherwise algorithm just is subjected to the restriction of the size of SE or input picture.
Can use some algorithms to carry out morphology operations.The simplest method is a strong algorithm.When every calculating finished a pixel, strong algorithm was just checked all the pixel/voxels in the SE.The result is stored on the different images.Depend on structural element " digital block " computing time of strong algorithm.This digital block is meant the quantity of the interior pixel of specified scope in two dimension (the being 2D) geometric configuration (that is its zone of dispersion) or in the three-dimensional of the appointment (that is 3D) quantity of the voxel in the geometric configuration (that is its discrete volume).Thereby in strong algorithm, each output pixel all needs () some computings, and the number of these computings equals the digital block of structural element is subtracted one (center pixel).Specifically, when foursquare structural element 100 has the length of three pixels and width, use this structural element 100 obtain the required operation number of each pixel of output image 112,120 (below be also referred to as op) be foursquare structural element 100 subtract a gained the zone (that is, 8op).And if use bigger structural elements usually to obtain different output images, then operation time, amount can increase on exponentially ground.For example, when the diameter (length of side) of square structure element is linear when increasing, the then square increase of computing time and element diameter with being directly proportional.For cube, when the cubical length of side is linear when increasing, cube increase of computing time and the cube length of side with being directly proportional.
The another kind of technology of carrying out morphology operations is factoring algorithm (for example, homotopy decomposition).Factoring algorithm uses the structural element of " substantially ".With these basic structural elements " decomposition " bigger structural element usually.Therefore, not that a big SE is offered image, and be to use a series of basic SE.Final output result is identical, but since the operand that less SE needs lack than the operand that single big SE needs, thereby reduced calculated amount.For example, suppose that structural element is that radius is 2 cube.Under strong algorithm, each output image voxel needs 5
3-1=124 op.But the use factoring algorithm, each output image voxel only needs (3
3-1) * 2=52 op.When using factorization, calculation cost is directly proportional with the cubical discrete diameter that limits the structural element of considering, rather than is directly proportional with its digital block.Like this, with respect to strong algorithm, factoring algorithm has been saved calculated amount significantly.
But factoring algorithm has also that serious hidden danger-under the situation of not doing to be similar to, employed basic structural element has seriously restricted and limited the shape than the macrostructure element that algorithm can use.For example, because the reason of organization of human body (as tracheae) uses the circular configuration element of two-dimensional digitalization or three-dimensional spherical structural element to carry out () Medical Image Processing usually.But, with respect to the shape of digitized circle or spheroid, because the shape of possible basic structural element can not be used factoring algorithm to circular or spherical structural element usually under the situation that does not cause excessive complexity.Therefore, for the purpose of practical application, best possible outcome is the approximate of circle or spheroid.
As a result, when carrying out morphology operations under the situation in restraining structure element shape not, especially, still need to reduce computing time for circle or spheroid.
Summary of the invention
The present invention is a kind of high efficiency method that is used for morphology operations is applied to image.In multidimensional image, support the binary and grayscale form.With morphology operations (as, expand or corrosion) be applied to the focal pixel of image repeatedly and be in apart from another pixel of the position of this certain side-play amount of focal pixel.This side-play amount is counted based on computing, and the latter can increase after each applied morphology computing.This computing counting is from a predetermined number that is increased to computing.Except last twice computing, the side-play amount of all computings is by 2
N-1Decision, wherein N is the computing counting.
Required operand is based on the logarithm of the predetermined length of the structural element that is applied to image.Specifically, required operand equals (log
2(Q)) get the value of maximum integer gained, wherein Q is the predetermined length of the structural element of concrete size.Except last computing, all computings (comparison) are all carried out on the concrete direction with respect to center pixel (), and last computing is carried out in the opposite direction.The side-play amount of last computing is for getting the negative value of the value of smallest positive integral gained to (Q/2).The side-play amount that penultimate computing has deducts 2 for the value of (Q/2) being got the maximum integer gained
N-2
In one embodiment, structural element also comprises preset width and/or predetermined altitude.In certain embodiments, pixel is a voxel.
This method also comprises uses the male structure element that image is carried out morphology operations.The work structuring element that will have corresponding to the size of the most external size of male structure element is applied to image repeatedly.And extra work painting canvas (image) is used for middle calculating.Then, with the adjusted size of work structuring element be size corresponding to the residue external edge of the male structure element that does not cover by before work structuring element.The result of iteration will be used for current iteration before considering, thereby this adjustment always becomes bigger.Repeat described application and set-up procedure, till having carried out the morphology operations of predetermined number, like this, just caused the morphology of given male structure element is calculated.
The male structure element can be disk, spheroid or ellipse.The work structuring element can be one-dimentional structure element or rectangular configuration element.It also can be rhombus or diagonal element.Comparison number before also comprises the current length of work structuring element and the maximal value of the logarithm of length before.
By following detailed description of reference and accompanying drawing, those of ordinary skills will be well understood to these and other advantages of the present invention.
Description of drawings
Fig. 1 illustrates the block scheme that input picture is carried out the prior art of two-value expansion and corrosion by 3 * 3 squares structural element;
Fig. 2 is the high-level block diagram according to the computing machine of the embodiment of the invention;
Fig. 3 is the block scheme that is optimized the morphology algorithm on 1 dimension (1D) row according to the embodiment of the invention;
Fig. 4 is to use the block scheme that is used for the two-value expansion of 5 * 7 rectangular configuration elements according to the optimized Algorithm of the embodiment of the invention;
Fig. 5 is a process flow diagram of using rectangle or cube structure element according to the embodiment of the invention image to be carried out the optimized Algorithm of dilation operation;
Fig. 6 A-6C illustrates according to embodiments of the invention, for the cube structure element, when its radius in time when 1 is increased to 10, it is used strong algorithm, factoring algorithm and the optimized Algorithm figure of required time;
Fig. 7 illustrates the example of 2D male structure element according to an embodiment of the invention;
Fig. 8 A and 8B are the power block diagrams of application that the oval structure element is applied to the optimized Algorithm of image according to the embodiment of the invention; With
Fig. 9 illustrates a kind of diagram, and this diagram has compared to be used 10 circular configuration elements with diameter change that 512 * 512 * 50 CT subimages that limit are carried out two-value according to the embodiment of the invention and strong algorithm to expand required working time.
Embodiment
Below describe with implementing the required treatment step of the embodiment of the invention the present invention has been described.Can be by carrying out these steps through the computing machine of suitably programming, being configured in of computing machine is in the industry known.For example, can realize suitable computing machine with known computer processor, memory cell, memory storage, computer software and miscellaneous part.Fig. 2 illustrates the high-level block diagram of this computing machine.Computing machine 202 comprises processor 204, and processor comes whole operations of control computer 202 by the computer program instructions of carrying out this operation of definition.Computer program instructions can be stored in the memory storage 212 (for example, disk), () and when the needs computer program is instructed is loaded into it in storer 210.Computing machine 202 also comprises one or more being used for and other devices communicatings interface 206 of (as local or pass through network).() computing machine 202 also comprise expression allow the mutual device of user and computing machine 202 I/O 208 (as, display, keyboard, mouse, loudspeaker, button or the like).Those skilled in the art will recognize that the realization of actual computer also can comprise other assemblies, Fig. 2 is the senior expression of some assemblies that is used for this computing machine of illustration purpose.In addition, those skilled in the art will recognize that, use specialized hardware also can realize the treatment step that illustrates herein, and the circuit of hardware disposed especially be used to realize this treatment step.In addition, also can use the various combinations of hardware and software to realize treatment step.
Fig. 3 illustrates the block scheme of one dimension image canvas (canvas) 300, and highlights the pixel 304 (that is focal pixel) of concern.Image canvas 300 is pixel (or voxel) zones that comprise image.Focus pixel 304 is used for better showing by a series of computings hereinafter referred to as optimized Algorithm, which kind of influence is single pixel be subjected to, though these computings are applied to each pixel of input picture.In the present embodiment, replace strong algorithm, in image canvas 300, only between two pictorial elements, carry out a series of comparisons in each step.This is with that two elements (sparse) SE is used for these concrete steps O is similar.The each application includes the right side-play amount of increase between two pictorial elements: the adjacent image point on each pixel and its right carries out logic OR (though this can be in logic OR, logic AND, MAX or the MIN computing) (that is, each pixel being adjacent pixel expands) to the right in will going.Optimized Algorithm is utilized the idempotent characteristic (that is, element is carried out the result of once above operator gained and only used coming to the same thing of an operator gained) of comparison operator.
Computing machine 202 is carried out this computing first time (that is, result's output of the OR computing between focal pixel 304 and its right adjacent image point is in focal pixel 304) in the original place.Expand for the first time and occur in from row 308 (that is the row that, does not have op) to row 312 (row of 1op).Should be noted that all vectorizations of each computing that are used for each example described below.That is to say that when calculating was proceeded, each computing all caused substituting of focusing pixel.Thereby for the computing first time, each pixel in the painting canvas 300 substitutes by the logic OR of the right pixel of itself and its direct neighbor.
The result is that the output of each pixel all is the results that itself are adjacent computing between the pixel.Following be which pixel helps to be stored in the result of information in the pixel with this span one that is called computing.Be used for the span of focal pixel 304 shown in Fig. 3 shade in the span with shadow representation focus pixel 304 among Fig. 3.After once-through operation, the span of the focal pixel 304 in the row 312 is the adjacent image point on itself and the right thereof.
Next, each pixel is not to be adjacent pixel, but carries out logic OR computing with the pixel on the right of a distance one pixel.In the drawings, use from row 312 to row 316 arrow to show this point.Thereby after carrying out twice OR computing, span is four pixels.Why span is four pixels, is because in fact this logic OR computing has comprised the adjacent image point on focal pixel 304 and its right and the adjacent image point on the 3rd pixel and its right.
Four-quadrant plain 324 is carried out next operator, arrive shown in the row 320 as row 316.By carrying out logic OR with focal pixel 304 and four-quadrant element 324, it is 8 pixels that span increases.Specifically, () is similar with the result's who carries out logic OR computing as focal pixel 304 itself and its three adjacent image points (i.e. four pixels) focal pixel 304, and the 4th pixel 324 () also is the result that itself and its three adjacent image points (i.e. four pixels) carry out logic OR gained.Therefore, when finishing 3 op, span is eight pixels.It is identical with effect to the SE of image calculation 8 pixel width that image is carried out 3 op.
The maximum span that can contain in M op of each output pixel is 2
MConversely, this determines to mean the more required minimum operand M of morphology that carries out N contiguous pixels
MinDetermine by following formula:
M
min＝CEIL(log
2(N))
Powerful technology is compared with optimized Algorithm described above, and optimized Algorithm is equivalent to only carry out eight logic OR computing (promptly expanding for eight times) in three op.Yet eight op of powerful Technology Need carry out eight logic OR computing.Thereby optimized Algorithm has been saved calculated amount significantly.
As concrete example, suppose that () is with having the plain structural elements of the 1D four-quadrant that is positioned at the center that left-most pixels as follows (runic) the locates same input picture as follows that usually expands:
1111 structural elements
00000100000 input picture
Use strong algorithm, the center pixel of this structural element each pixel with input imagery is arranged in, then to carrying out the OR computing together by the territory of facing of the input imagery of structural element definition.This has caused following output image:
00111100000
Use optimized Algorithm on input imagery separately, for each computing, the result who obtains is as follows:
No op 00000100000
1op 00001100000
2op 00111100000
As implied above, optimized Algorithm has produced the result identical with strong algorithm after twice computing.Can determine the required operand of the plain 1D structural element of four-quadrant by above-mentioned equation: 2
N=4; N=2op.Therefore, need 2 op to obtain the identical output image of output image that obtains when using the plain 1D structural element of four-quadrant.In the above description, an op on the direction only is shown.This has only considered to have the SE of left side center pixel.As will be shown, use single op then to consider symmetric operator in the opposite direction.
Based on above-mentioned principle, can the rectangular configuration element be applied to image with rectangular configuration element optimized Algorithm.Fig. 4 illustrates the block scheme that uses optimized Algorithm image canvas 400 to be carried out the two-value expansion with 5 * 7 rectangular configuration element spare.The pixel that focal pixel 404 is paid close attention to, still, as mentioned above, computing is by vectorization.
Then, in step 462, computing machine 202 calculate focal pixel 404 and its below, and the pixel of its direct neighbor between logic OR.This is actually the logic OR that is positioned between 5 * 20 initial pixel values of facing on the territory.In step 466 and 470, computing machine 202 continues to carry out optimized Algorithm, remove this algorithm and be in vertical direction rather than the application in the horizontal direction, its executive mode all with step 454 and 458 in identical.Final result has been shown in step 470.Focal pixel 404 expressions have the expansion results of 5 * 7 rectangular configuration elements of image canvas 400.In addition, although carried out top description, also above-mentioned optimized Algorithm can be applied to three-dimensional rectangle structural element (for example, 5 * 5 * 7) about 2D rectangular configuration element.
Although be illustrated using two-value expansion operator (that is, logic OR) above, optimized Algorithm can be used any comparison operator.Therefore, can substitute the OR operator comparably with other morphology comparison operators (for example, AND, MIN or MAX).This ability makes and this method can be applied to the binary and grayscale field.Using this method on diagonal allows the diamond structure element is carried out morphology calculating.
Fig. 5 illustrates the process flow diagram that uses the optimized Algorithm that rectangle or cube structure element be used to expand.This process flow diagram shows the summary to the 1D situation, and 2D situation and 3D situation all are expansions (that is, being applied to different axles) of 1D situation.Therefore, Fig. 5 is illustrated in the symmetric while of noting keeping net result, obtains the logic optimized Algorithm relatively of odd length (Q) operation of two-value pixel efficiently.
At first, in step 504, determine the minimum number (N) of the op that Q pixel of comparison is required.Provide this minimum number by above-mentioned equation:
N
min＝CEIL(log
2(Q))
Because net result will be for symmetry, thereby (that is, N) in addition, all op (that is N-1 op) all will go up increase span () in a direction (being referred to as positive dirction) to remove last op in step 508.Last op will be used for expanding span on contrary direction.
Need be the side-play amount of each op acquisition with respect to focal pixel.Because last op will be on the contrary direction, therefore for the logic op of accumulation, will have N-1 forward migration amount.() to carry out N-2 op (promptly except last forward op) with respect to the 2P side-play amount of focal pixel, wherein P ∈ [0,1,2 ... (N-2)-1].Last forward migration amount is CEIL (Q/2)-2
(N-2)
Afterwards, in step 512, computing machine 202 is expanded span on contrary direction.At first, computing machine 202 is determined its side-play amount with respect to focal pixel for last op.This side-play amount is-FLOOR (Q/2) that negative sign wherein is used for indicating this side-play amount on contrary direction.
For example, for N=2, only there is a forward migration amount (N-1).This occurs in as Q=3 when (that is, the pixel on focal pixel, a focal pixel the right and the pixel on a focal pixel left side are formed structural element).
Although abovely and following be illustrated using two-value expansion operator (that is, logic OR), optimized Algorithm can be used any comparison operator.Therefore, can substitute the OR operator with any other morphology comparison operator (for example, AND, MIN or MAX) with being equal to.And, repeating step 504-512 on remaining axle.
For the cube structure element, when its radius in time when 1 is increased to 10, Fig. 6 A-6C illustrates the time diagram that uses strong algorithm, factoring algorithm and optimized Algorithm.
Fig. 6 A shows, when the radius of cube structure element increases, uses the calculated amount exponentially ground of strong algorithm cost to increase.Thereby when the radius of cube structure element was 6, curve 602 showed that the required time of structural element of using the use strong algorithm is approximately 2000 units (for example, second).Because show the required height of time of strong algorithm, the required time of other two kinds of methods almost is sightless.
Fig. 6 B illustrates curve 604, and this curve shows that when the radius of cube structure element increased, the calculated amount of factoring algorithm cost increased linearly.When the radius of cube structure element was 6, curve 604 showed that the required time of structural element of using the use factoring algorithm is approximately 150 units (for example, second).Equally, time scale has also reduced to the 0-300 of factoring algorithm from the 0-10000 of strong algorithm.Again, the time of the method that is proposed almost is sightless.
Fig. 6 C illustrates curve 608, and this curve shows that the calculated amount that above-mentioned optimal algorithm spends is directly proportional with the logarithm of cube structure element radius.When the radius of cube structure element was 6, curve 608 showed that the required time of application structure element is approximately 12 units (for example, second).Time scale is also reduced to the 6-16 of optimized Algorithm by the 0-300 of factoring algorithm.
Also optimized Algorithm can be applied to the digital male structure element () () such as 2D circle (or oval) or 3D spheroid (or ellipsoid).() convexity herein means, can draw the line of the mid point that arrives any other pixel the same structure from the mid point of the arbitrary pixel of digital male structure (2D), and this line must be among this structure fully, perhaps, if some drops on the outside of structure this line, then by the zone that boundary segmentation defined that has common starting point and a terminal point with described line segment of line segment and numeric structure do not comprise any be not consider the mid point of pixel of the part of structure.
In one embodiment, for using the circular configuration element, use optimized Algorithm to need two image canvas to obtain the morphology result.Computing machine 202 is carried out optimized Algorithm repeatedly, and wherein, each iteration comprises the span spread step and selects/comparison step.In each iteration, computing machine 202 " is constantly removed " structural element of waiting until processing.
Top-down inspection structural element-promptly, consider that it has the outermost row/row of unique length.Because convexity, the length of each the newline/row that runs into all will be bigger than the length of front row/row.Like this, the span of facing the territory has obtained expansion, the comparison that the territory is faced in expression in each pixel of work painting canvas.This allows calculating that the iteration by before the obtains basis as follow-up work.The system of these intermediate results forms the morphology result who has produced hope.Since () () with the approximately parallel framework of dynamic programming (bottom-up) in, being decomposed into of structural element realizes that morphology created condition, thereby for algorithm (top-down), the decomposition of structural element is a kind of " dividing and ruling " method.Thereby, utilized the space overlap of the comparison of the initial painting canvas in adjacent output pixel.
Fig. 7 illustrate have line length for 5,9,11,13,15,17} and row highly be 1,5,9,11,13, the example () of the male structure element of 15} (a digital circle of 2D) 700.Do not have this structural element 700 is used for above-mentioned factoring algorithm.When each iteration, two image canvas are upgraded, wherein, at first upgrade odd-job painting canvas W with reflection rectangle span expansion (part 1).() relatively upgrades output image painting canvas O (part 2) with the vectorization between the W of output image painting canvas O itself and renewal subsequently.In one embodiment, the input picture painting canvas is as O.Perhaps, use independently image canvas.
Fig. 8 A and 8B show two image canvas 804,808 that are used for optimized Algorithm, so that the 2D circular configuration element among Fig. 7 is applied to image.Middle painting canvas W 804 comprises from initial pictures with to itself result calculated.The computing that final output image O 808 is included in W and is done between itself.Finished this application by three iteration.In three iteration of first 804, the rectangle expansion marking (footprint) of W is increased to maximum rectangle, so that when expanding, can cover remaining structural element (RSE) fully with this rectangle expansion marking with suitable structural element.In three iteration of second portion 808, W and O are systematically compared.The rectangle expansion marking to W is offset so that it is related with RSE.Therefore, with each pixel among the O itself and relatively each pixel the O is upgraded between the W value of relevant position skew, so that can constantly remove the row and column of the outermost uniqueness of RSE.
Particularly, at first, in iteration 1, computing machine 202 uses the structural elements of the row of 5 pixels usually W to be carried out the original place expansion, travels through this result then to set up O.The result is that 4 pixels among the W merge to the single central pixel among the O, and this is similar to has used sparse 4 pixel SE.In fact this " constantly removed " structural element of waiting until processing.Now, with respect to initial configuration element (as shown in Figure 7), in through the RSE that upgrades, removed the row and the most left and the rightest pixel of () top and bottom.
In iteration 2, () needs W to comprise to utilize the expansion results of the initial pictures of 9 * 5 rectangular configuration elements, because the length of these are outermost layers of RSE capable/row (i.e. the most external size that is not covered by the structural element in the step before).As the result of iteration 1, W has comprised the expansion results with rectangular configuration element of 5 * 1.Therefore, computing machine 202 needs to upgrade W, so that 5 * 1 the rectangle expansion marking is expanded to 9 * 5 the rectangle expansion marking.Notice that this step is reused the data from iteration 1, it is more efficient on calculating that this makes it calculate 9 * 5SE than starting from scratch.Then, pixel among the W and the pixel among the O are carried out strategic comparison.The expansion marking that reflects in each pixel of O when subsequently, being illustrated in iteration 2 end.
In iteration 3, computing machine 202 expands to 13 * 9 (it is the length through most external row/row of the RSE that upgrades) with the expansion results among the W, and O has been carried out selecting/comparing of final bout, with the result who obtains wishing.Carrying out through the required parameter of summarizing of algorithm for any protruding, X-Y symmetrical structure element is rectangle the span () increment (being first 804) among the W and the side-play amount (second portion 808) () that is used to select the W element with respect to the focal position of O after iterations, the each iteration.Filling (that is the adjustment of structural element, (as expansion)) required among the W is by determining than more general equation described above.This general equation is as follows:
M
min＝CEIL(log
2(L
now/L
prew))
Thereby the minimum number of required morphology operations depends on the current length of rectangular configuration element and length before.
And the optimized Algorithm of the above-mentioned 2D of being used for circular configuration element also can be applied to 3D spherical structure element.Under the situation of 3D spheroid, need to use 6 or more voxel operator.Another example of male structure element is oval.
Fig. 9 illustrates and uses 10 circular configuration elements with diameter change 512 * 512 * 50 CT subimages that limit to be carried out the diagram 900 of the working time that two-value expands.Diagram 900 illustrates curve 904, and this curve illustrates when using the algorithm of prior art (for example MATLAB tool box), the computing time that increases with the increase of the circular configuration element radius that disperses.908 of curves illustrate when using above-mentioned optimized Algorithm, the increase of oblong structural element radius and computing time of increasing.In diagram 900, be approximately 1100 seconds the highest working time that prior art reaches, and be about 200 seconds the highest working time that optimized Algorithm reaches.
Aspect each, all above detailed description should be understood as illustrative and illustrative rather than restrictive, scope of the present invention disclosed herein also be can't help to describe in detail and is determined, but is determined by the claim that the four corner that allows according to Patent Law is explained.Should be appreciated that the embodiment that illustrates and illustrate only is an explanation of the principles of the present invention herein, and those skilled in the art can carry out various modifications to these embodiment under the prerequisite that does not deviate from scope and spirit of the present invention.Those skilled in the art also can realize various other combination of features under the prerequisite that does not deviate from scope and spirit of the present invention.
Claims (44)
1, a kind of method that is used for morphology operations is applied to image comprises:
Described morphology operations is applied to the focal pixel of described image and another pixel of described image repeatedly, and described another pixel is in apart from the position of a certain side-play amount of described focal pixel, and described side-play amount is counted based on computing.
2, the method for claim 1, the number of times that wherein said computing counting has been employed for described morphology operations.
3, the method for claim 1, wherein said computing counting is increased to from one described morphology operations is applied to the required operand of described image.
4, the method for claim 1, wherein said side-play amount is by 2
N-1Determine that wherein N is described computing counting.
5, method as claimed in claim 3 also comprises the operand of determining described needs.
6, method as claimed in claim 5, the operand of wherein said needs depends on the predetermined length of structural element.
7, method as claimed in claim 6, wherein said operand is based on the logarithm of described predetermined length.
8, method as claimed in claim 7, wherein said operand equals (log
2(Q)) get the value of maximum integer gained, wherein Q is the described predetermined length of described structural element.
9, method as claimed in claim 8, the last computing of wherein said operand is carried out on the contrary direction with respect at least some other computings.
10, method as claimed in claim 6, the described side-play amount of wherein said last computing is for getting the negative value of the value of smallest positive integral gained to (Q/2), and wherein Q is the described predetermined length of described structural element.
11, method as claimed in claim 6, the side-play amount that time computing wherein second from the bottom has deducts 2 for the value of (Q/2) being got the maximum integer gained
N-2, wherein Q is the described predetermined length of described structural element, and N is described computing counting.
12, the method for claim 1, wherein said pixel are voxel.
13, method as claimed in claim 6, wherein said structural element also comprises predetermined width.
14, method as claimed in claim 6, wherein said structural element also comprises predetermined height.
15, a kind ofly be used to use the male structure element that image is carried out the method for morphology operations, comprise:
The work structuring element that will have corresponding to the size of the most external size of described male structure element is applied to described image repeatedly;
Adjust the size of described work structuring element, so that it is corresponding to the residue external dimensions of the described male structure element that is not covered by the work structuring element before described; With
Repeat described application and described adjustment till the morphology operations of finishing predetermined number.
16, method as claimed in claim 15, wherein said male structure element comprise at least a in circle, ball and the ellipse.
17, method as claimed in claim 15, wherein said work structuring element also comprises the one-dimentional structure element.
18, method as claimed in claim 15, wherein said work structuring element also comprises the rectangular configuration element.
19, method as claimed in claim 15, the predetermined number of wherein said morphology operations also comprise the current length of described structural element and the logarithm of the length value of getting the maximum integer gained before.
20, method as claimed in claim 15, wherein the described adjustment that the described size of described work structuring element is carried out occurs on the work painting canvas.
21, method as claimed in claim 15, wherein at least one focal pixel by described morphology operations being applied to described image and at least one other pixel of described image are applied to described image with described work structuring element.
22, method as claimed in claim 21, described at least one other pixel of the described focal pixel of wherein said image and described image also comprise at least one voxel.
23, a kind ofly be used to use the male structure element that image is carried out the system of morphology operations, comprise:
Be used for to have the device that is applied to described image corresponding to the work structuring element of the size of the most external size of described male structure element repeatedly;
The described size that is used to adjust described structural element is so that its device corresponding to the residue external dimensions of the described male structure element that is not covered by the structural element before described; With
Be used to repeat described application and the device of described adjustment till the morphology operations of finishing predetermined number.
24, system as claimed in claim 23, wherein said male structure element comprises at least a in circle, ball and the ellipse.
25, system as claimed in claim 23, wherein said work structuring element also comprises the one-dimentional structure element.
26, system as claimed in claim 23, wherein said work structuring element also comprises the rectangular configuration element.
27, system as claimed in claim 23, the predetermined number of wherein said morphology operations also comprise the current length of described structural element and the logarithm of the length value of getting the maximum integer gained before.
28, system as claimed in claim 23, wherein the described adjustment that the described size of described work structuring element is carried out occurs on the work painting canvas.
29, system as claimed in claim 23, wherein at least one focal pixel by described morphology operations being applied to described image and at least one other pixel of described image are applied to described image with described work structuring element.
30, system as claimed in claim 29, described at least one other pixel of the described focal pixel of wherein said image and described image also comprise at least one voxel.
31, a kind of morphology operations is applied to the system of image, comprises:
Be used for described morphology operations is applied to repeatedly the device of another pixel of the focal pixel of described image and described image, described another pixel is in apart from the position of a certain side-play amount of described focal pixel, and described side-play amount is counted based on computing.
32, system as claimed in claim 31, the number of times that wherein said computing counting has been employed for described morphology operations.
33, system as claimed in claim 31, wherein said operand are increased to from one described morphology operations are applied to the required operand of described image.
34, system as claimed in claim 31, wherein said side-play amount is by 2
N-1Determine that wherein N is described computing counting.
35, system as claimed in claim 33 also comprises the device that is used for determining required described operand.
36, system as claimed in claim 35, wherein required described operand depends on the predetermined length of structural element.
37, system as claimed in claim 36, wherein said operand is based on the logarithm of described predetermined length.
38, system as claimed in claim 37, wherein said operand equals (log
2(Q)) get the value of maximum integer gained, wherein Q is the predetermined length of described structural element.
39, system as claimed in claim 38, the last computing of wherein said operand is carried out on the contrary direction with respect at least some other operations.
40, system as claimed in claim 39, the described side-play amount of wherein said last computing is for getting the negative value of the value of smallest positive integral gained to (Q/2), and wherein Q is the predetermined length of described structural element.
41, system as claimed in claim 39, the side-play amount that time computing wherein second from the bottom has deducts 2 for the value of (Q/2) being got the maximum integer gained
N-2, wherein Q is that the described predetermined length and the N of described structural element are described computing counting.
42, system as claimed in claim 31, wherein said pixel is a voxel.
43, system as claimed in claim 36, wherein said structural element also comprises predetermined width.
44, system as claimed in claim 36, wherein said structural element also comprises predetermined height.
Applications Claiming Priority (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US63765304P | 2004-12-20 | 2004-12-20 | |
| US60/637653 | 2004-12-20 | ||
| US11/301393 | 2005-12-13 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN1825354A true CN1825354A (en) | 2006-08-30 |
Family
ID=36936016
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 200510121655 Pending CN1825354A (en) | 2004-12-20 | 2005-12-20 | Method and apparatus for efficient computation of morphology operations |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1825354A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102622735A (en) * | 2012-04-18 | 2012-08-01 | 江苏技术师范学院 | Secondary de-noising image processing method |
| CN104766085A (en) * | 2014-12-30 | 2015-07-08 | 沈阳理工大学 | Multi-scale figure recognition method |
| CN110097525A (en) * | 2019-04-23 | 2019-08-06 | 厦门美图之家科技有限公司 | A kind of image rendering method, device and calculate equipment |
| CN112528209A (en) * | 2020-12-08 | 2021-03-19 | 四川蓉信开工程设计有限公司 | Method for rapidly calculating volume rate of irregular box based on divide-and-conquer thought |
-
2005
- 2005-12-20 CN CN 200510121655 patent/CN1825354A/en active Pending
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102622735A (en) * | 2012-04-18 | 2012-08-01 | 江苏技术师范学院 | Secondary de-noising image processing method |
| CN104766085A (en) * | 2014-12-30 | 2015-07-08 | 沈阳理工大学 | Multi-scale figure recognition method |
| CN104766085B (en) * | 2014-12-30 | 2019-01-29 | 沈阳理工大学 | A kind of multiple dimensioned pattern recognition method |
| CN110097525A (en) * | 2019-04-23 | 2019-08-06 | 厦门美图之家科技有限公司 | A kind of image rendering method, device and calculate equipment |
| CN110097525B (en) * | 2019-04-23 | 2021-01-29 | 厦门美图之家科技有限公司 | Image rendering method and device and computing equipment |
| CN112528209A (en) * | 2020-12-08 | 2021-03-19 | 四川蓉信开工程设计有限公司 | Method for rapidly calculating volume rate of irregular box based on divide-and-conquer thought |
| CN112528209B (en) * | 2020-12-08 | 2024-04-12 | 四川蓉信开工程设计有限公司 | Irregular box volume rate rapid calculation method based on divide-and-conquer idea |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Haris et al. | Deep back-projection networks for super-resolution | |
| CN104700451B (en) | Point cloud registering method based on iterative closest point algorithm | |
| CN1934587A (en) | Method, computer program product and apparatus for enhancing a computerized tomography image | |
| CN112508808B (en) | CT double-domain combined metal artifact correction method based on generation countermeasure network | |
| CN101052991A (en) | Feature weighted medical object contouring using distance coordinates | |
| CN106683144A (en) | Image iteration reconstruction method and device | |
| CN110223376B (en) | Three-dimensional particle reconstruction method based on single accumulated particle material image | |
| US7974499B2 (en) | Method and apparatus for efficient computation of morphology operations | |
| CN104866868A (en) | Metal coin identification method based on deep neural network and apparatus thereof | |
| CN1819621A (en) | Medical image enhancing processing method | |
| CN109671061B (en) | Image analysis method and device, computing equipment and storage medium | |
| Schouten et al. | Fast exact Euclidean distance (FEED): A new class of adaptable distance transforms | |
| CN1825354A (en) | Method and apparatus for efficient computation of morphology operations | |
| JP6643821B2 (en) | Image processing device | |
| CN110335322B (en) | Road recognition method and road recognition device based on image | |
| WO2018096122A2 (en) | System and method for determining the presence of objects in an image | |
| CN107316282A (en) | A kind of system and method for image noise reduction | |
| CN109584368B (en) | Method and device for constructing three-dimensional structure of biological sample | |
| CN1207673C (en) | Half-tone dot eliminating method and its system | |
| CN1497695A (en) | Graphic measuring method, manufacturing method of semiconductor device using it and graphic measuring device | |
| US11562231B2 (en) | Neural networks for embedded devices | |
| CN1236728C (en) | Tomographic image reconstructing process and radiographic apparatus | |
| CN113532320B (en) | Image-based facula diffraction ring analysis method, storage medium and chip | |
| CN1366275A (en) | Technology for removing false image of storage of liquid crystal projector | |
| CN113240015A (en) | Image matching method combined with deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |