CN100592335C - Using corner pixels as seeds for detection of convex objects - Google Patents

Abstract

An exemplary for selecting seeds from an image for region determination is provided. The method includes determining a boundary between two areas in the image; selecting pixels on the boundary that are characterized by a salient feature that identifies the pixels as seeds for determining a region; and determining a second region from one of the selected pixels if the one of the selected pixels isnot part of a previously determined first region.

Description

Use corner pixels as seed to detect convex object

The cross reference of related application

The application requires U.S. Provisional Application No.60/549,047 right of priority, and this application all is incorporated herein by reference at this in submission on March 1st, 2004 and its.

Technical field

The present invention relates generally to the computer generated image field, and relate more particularly to use the notable feature of image to be identified for the seed of detected object.

Background technology

The detected object shape is absolutely necessary in many application, for example computer aided detection and diagnosis in two dimension (" 2D ") and three-dimensional (" 3D ") image.For example, computer aided detection and diagnostic application use usually shape localization as identification may be interested the preliminary step of (for example indicating disease potentially) ad hoc structure.Term " shape localization " refers to coordinate is associated with given position in volume or the space.

Shape localization can be carried out with two steps usually:

(1) identification and extraction jointly or individually characterize the set (zone just) of the pixel/voxel of shape; And

(2) use different shape descriptor/measure (metrics) to estimate and analyze this set to determine whether this set fully shows the shape of being considered.

The method that is used for determining in the zone includes, but are not limited to: region growing, region clustering and Region Segmentation.For example the traditional area of Tan Lan zone (greedy region) growth technology can have the standard that very simple selection is used to make the seed (starting point just) of region growing.For example, a kind of exemplary area increases technology can consider whether each pixel/voxel in the image and checking have desired characteristic (for example tight ness rating, oval structure or other), these characteristics representatives shape facility relevant with desired shape from the zone that specific pixel/voxel extracts.If the zone that this extracted has desired characteristic, the zone that this extracted just can be considered to " example " of the shape that detected so.

As used herein, term " shape " refers to space or the volume that is centered on by the border, and this border makes this space or volume separate with contiguous material or structure.This border can have clearly or fuzzy transition (edge just).The border is a kind of transition of specific type, and this transition has definite width (extent) on the direction vertical with transition.The quality of transition during according to material and image data used formation method change.As an example, if utilize the laser distance scanner of surface imaging is gathered, the edge may be (for example from deceiving to white direct transformation, perhaps vice versa) of binary so.As another example, for example material (for example suitcase) or stand the people's of regular physical examination computer tomography (" CT ") or the situation of radioscopic image under, the edge may be clearly along with luminance transition.As another example, as under the situation of ultrasonic or magnetic resonance imaging, the edge may not be well-defined partly.Yet, can be used as the seed that makes region growing no matter above-described border quality how, is determined borderline any point that desired structure is separated with the proximity structure of not expecting.

With reference now to Fig. 1,, the illustrative computer tomography of the part of colon 100 (" CT ") image illustrates border 105 (white bars zone just), and this border is two transition between the zone: inner chamber 110 (shadow region just) and the tissue 115 that separates (pattern area just).An example may wishing the projection (convex domain just) that detects is shown by the district of dotted line 120 sealing.This convex domain also can be called as area-of-interest.In the context of the CT of colon 100 image, may be polyp of colon or be attached to epipleural lung tubercle for example by the district of dotted line 120 sealing.

With reference now to Fig. 2,, shows another view of this partial C T image of colon 100 among Fig. 1.Fig. 2 clearly show that convex domain 205 and cuts apart the virtual surface 210 of this convex domain 205.If there is no projection (just convex domain 205), this virtual surface 210 smooth continuation that is borders 215 so.It should be noted that virtual surface 210 is illustrated as just vision is auxiliary in Fig. 2.

In the algorithm of traditional greediness, all surface point on the border 105 can be considered to potential seed.Particularly when big (for example order of magnitude of millions of pixel/voxel) image, this processing may excessively expend time in and efficient low.

Summary of the invention

In one aspect of the invention, provide a kind of being used for to select seed to determine the method in zone from image.This method comprises the border of determining in the image between two districts; Select borderlinely by the pixel that notable feature characterized, this notable feature is defined as these pixels to be used for to determine the seed in zone; And if the part that pixel is not previous determined first area in the selected pixel, then determine second area according to this pixel in the selected pixel.

In another aspect of the present invention, provide a kind of machine readable media, this machine readable media has storage instruction thereon, and these instructions are carried out with execution by processor and are used for selecting seed to determine the method in zone from image.This method comprises the border of determining in the image between two districts; Select borderlinely by the pixel that notable feature characterized, this notable feature is defined as these pixels to be used for to determine the seed in zone; And if the part that pixel is not previous determined first area in the selected pixel, then determine second area according to this pixel in the selected pixel.

In another aspect of the present invention, provide a kind of being used for to select seed so that the method for region growing from image.This method comprises the boundary surface of determining in the image between two zones; What filter boundary surface is the pixel of low dimension angle point; To be placed in the seed list through the pixel of filtering; From this seed list, select first seed so that the first area increases; And from this seed list, select second seed, wherein do not have only when this second seed is not this first area a part of just this second seed to be used to make second area to increase.

Description of drawings

Can understand the present invention with reference to following description taken together with the accompanying drawings, wherein identical Reference numeral is represented components identical, and wherein:

Fig. 1 describes the exemplary CT image of the part of colon;

The exemplary CT image of this part of the colon in Fig. 2 depiction 1;

Fig. 3 describes the 2D neighborhood according to an exemplary of the present invention, this field be shown having the C that is positioned at the center and with the distance of C be 1 neighbors a1-a8;

Fig. 4 describes the convex domain 205 according to Fig. 2 of angle point " c " exemplary of the present invention, that have institute's mark and " x ";

Fig. 5 describes according to an image exemplary of the present invention, that have the angle point voxel, and wherein the 2D angle point in the zy plane no longer is correct 3D angle point significantly, and other angle point remains correct angle point in 3D the time; And

Fig. 6 describes according to a method exemplary of the present invention, that be used for being used in the image selection the suitable seed position of region growing.

Embodiment

Illustrative embodiment of the present invention is described below.Whole features of actual embodiment for the sake of clarity, are not described in this manual.Certainly it should be understood that in the improvement of any this actual embodiment, for the specific objective of realizing the developer, for example observe and to make the specific decision of a lot of embodiments because of the different relevant constraint relevant of system of embodiment with commerce.In addition, it should be understood that this improvement may be complicated and be time-consuming, but for those those of ordinary skills that benefit from present disclosure, remain normal work to do.

Although the present invention allows various modifications and alternative form, its specific embodiments for example is illustrated in the accompanying drawings and is described in detail at this.Yet, it should be understood that, be not intended the present invention is limited to particular forms disclosed in this description to specific embodiments, and antithesis, intention covers all and belongs to modification, equivalence and replacement scheme by the appended the spirit and scope of the present invention that claim limited.

It should be understood that system and method described herein can realize with the various forms of hardware, software, firmware, application specific processor or its combination.Particularly, at least a part of the present invention realizes preferably as the application program that comprises programmed instruction, these programmed instruction positively be included in one or more program storage device (for example hard disk, magnetic floppy disc, RAM, ROM, CD ROM or the like) and can be by any device that comprises suitable structure or machine, the universal digital computer that for example has processor, storer and an input/output interface carries out.What will be further understood that is, because forming the parts and the method step of systems, some that described in the accompanying drawing preferably realize, so the connection between the system module (the perhaps logic flow of method step) can be according to the mode of the present invention's programming and difference with software.When providing when instruction at this, those of ordinary skill in the related art can be susceptible to these and similar embodiment of the present invention.

Proposition is used for optionally pixel/voxel being regarded as the illustrative methods and the system of the possible seed that makes region growing.Be different from from all boundary positions (pixel/voxel just) and attempt the region growing method, more efficient methods can only begin region growing from selected position.For example in medical applications, size of images to be processed is big (for example order of magnitude of millions of pixel/voxel) quite.Therefore, make the possible kind period of the day from 11 p.m. to 1 a.m of region growing should take careful attention when pixel being seen act on.By careful selection seed points, can realize sizable acceleration.

Following four kinds of observations (observation) are crucial for determining seed points:

(1) any convex domain border that must be expressed the transition between this zone inside and outside surrounds.The characteristic of border transition (marginarium just) can depend on to be used and data mode (for example computer tomography, magnetic resonance, ultrasonic or the like).

(2) under the situation of strict convex object, for example disk or spheroid, always there is at least one point, this point is an angle point.Angle point is a surface point, and the prospect neighbors of this surface point strictly is arranged in half of plane/volume.

(3) angle point is a plurality of of can utilize on the surface of convex object in the point that notable feature is positioned.Utilize different notable features to make other unique position can help making the region growing that to discern.Example that allow to determine the notable feature of good seed position comprises the notable feature of track of the projection that lip-deep maximum or Gaussian minimum (Gaussian) curvature, normal intersect or the like.

(4) recessed zone also can be construed to negative convexity, and all above-mentioned considerations are suitable equally in this case.

Here in the exemplary of being discussed, notable feature will will be characterized by lip-deep angle point based on the characteristic and the selected position of neighborhood.

In order to limit corner location better, introduce connective symbol.If a left side and the right neighbors of position " C " all exist, then this position just has and equals two (2) connectedness (being marked as N2 is communicated with).If position " C " is N2 to be communicated with and to have a upper and lower neighbors, then this position just has and equals four (4) connectedness (being marked as N4 is communicated with).As one of ordinary skill as expected, can use other label symbol.For example, with reference now to Fig. 3, " C " has 8 distances with " C " is 1 neighbors a1-a8.This is the example of matrix neighborhood; Yet, as one of ordinary skill as expected, can use other layout of different topological structures.

Now, given " C " is positioned at the neighborhood at center, when existence be adjacent to each other and be N4 be communicated with half or still less during neighbors, have angle point.For example, refer again to Fig. 3, have 8 neighbors, wherein 4 neighbors have N4 connectedness (for example by clockwise order a1-a4) at the most.The N4 connectedness of neighbors is requirement, and will to make C be not angle point but bridge because have a1, a2, a3 and a5.Term N4, N6, N8 and N26 refer to the quantity with the adjacent neighbors of surface point.For example, N4 relates to center pixel and in each pixel of this center pixel left side, right side, top and bottom.As another example, N26 relates to each adjacent with the center voxel in 3 * 3 * 3 neighborhoods voxel.

Although do not limited like this, for simplicity, term " pixel " will be used to describe exemplary below.Yet, should be appreciated that in an alternative embodiment, these embodiments can comprise voxel rather than pixel.

Because target is to extract the convex domain that comprises boundary pixel/voxel effectively from image, therefore will need a kind of technology that can optionally increase from boundary pixel/voxel.

For example under the situation of the polyp of colon be projected into inner chamber (air just) from colon wall, when using region growth technique to detect the shape of polyp, any point on the colon surface can be counted as effective seed points.Particularly, as observed among Fig. 1 and Fig. 2, interesting areas (convex object just) has frontier district and projection when detecting.For outstanding pixel/voxel, these pixel/voxel must be crooked and be expanded outwardly, thus the introducing angle point.These angle points are a part and (b) part of convex domain on (a) border normally.

With reference now to Fig. 4,, illustrates and have the angle point " c " that is labeled and the convex domain 205 of " x ".Because Fig. 4 is the 2D image, the angle point that is labeled " c " and " x " are the same good candidates of beginning region growing.All outstanding border angle points are the same good candidates that make the 2D region growing.Yet, when considering the 3D voxel of adjacency, among Fig. 3 by many among these candidates of " x " institute mark no longer be among Fig. 3 by the good candidate angular of " c " institute mark, and be borderline voxel.

With reference now to Fig. 5,, the example images 500 with angle point voxel is shown, wherein the 2D angle point in zy plane (plane that for example comprises x2 and c1) obviously no longer is correct 3D angle point, and other angle point remains correct angle point in 3D the time.This is important consideration because when the dimension of object increases (just from pixel (2D) to voxel (3D)) it can reduce candidate's quantity.Is angle point by the angle point voxel of " x " institute mark when considering the 2D plane, but have only those by the angle point voxel of " c " institute mark when more high-dimensional or angle point.

Therefore, as can be seen, in Fig. 5, although x2 is an angle point in the zy plane, x2 is not angle point in the zx plane.Therefore 3D (perhaps more high-dimensional) angle point can be considered to pixel, so that C is an angle point in all process planes (perhaps lineoid) of this position.Utilize this characteristic, having only those angle points that are marked as C as can be seen in Fig. 5 just really is the 3D angle point.This characteristic is important, because it further reduces candidate's quantity.Plane (perhaps lineoid) by this position is not limited in quadrature xy, zy and zx plane, but extends to other possible plane.When for example considering 3 * 3 * 3 neighborhoods, 13 discrete Different Plane are passed this point.In another embodiment of the invention, neighborhood can be bigger or different shapes is arranged, thereby further allows the discrete feature on plane.In another embodiment, these lineoid can not be disperse just in time to pass through the center of voxel.In another embodiment of the present invention, intersect and can characterize by 3D (or more high-dimensional) angle point, this angle point is not by plane or lineoid but determine by line or line segment.

With reference now to Fig. 6,, a kind of illustrative methods that is used for selecting at image the suitable seed position that is used for region growing is shown.Between two zones, determine (in 605) boundary surface.An exemplary of this method can be carried out the Canny rim detection, and this detection can change according to feature and the transient characteristic between two zones.The pixel for low dimension angle point of border surface is filtered (in 610).These are placed (in 615) in seed list through the pixel of filtering.From this seed list, select (in 620) first seed so that first convex domain increases.From this seed list, select (in 625) second seed.Having only when this second seed just increases convex domain from this second seed when not being first convex domain a part of.

Refer again to Fig. 4, the example of the outstanding convex domain that comprises seed c2 and c3 is shown.In this case, if select c2 as first seed after handling c1, the convex domain that is extracted will also comprise c3 probably.Therefore, after having finished the region growing that begins with c2, c3 will be counted as new seed, and observing it has been the part of convex domain, and the region growing process will skip to c4, or the like.

The example protruding or recessed zone that increases from the seed position can be the surface of the convex domain of reality.For example, refer again to Fig. 2, the zone that increases from seed points can be a ribbon area, up to line and the crossing pixel of this band.Therefore, this convex domain can comprise from whole interior zone and surface self that this seed points increases or can only be characterized by the surface.

Above disclosed particular only be illustrative because but the present invention can be modified and put into practice in mode different equivalence, these modes are conspicuous for the those of ordinary skills of instruction that have benefited from this.In addition, except as described in the following claims, the structure that goes out shown here and the details of design are not intended as any restriction.Therefore, obviously can change or revise, and all such variations all be considered within the scope and spirit of the invention top disclosed particular.Therefore, this protection of looking for such as below claim described in.

Claims (20)

1. one kind is used for being the definite method of selecting seed in zone from image, and this method comprises:

Determine in the image border between two districts;

By selecting this borderlinely to form seed list by a plurality of pixels that notable feature characterized, this notable feature is the seed that is used for determining the zone with these pixel logos;

From seed list, select first seed to determine the first area; And

From seed list, select second seed of the part of non-first area, to determine second area.

2. the method for claim 1 is wherein determined first and second zones, wherein each zone comprise growth, cluster and cut apart in a kind of.

3. the method for claim 1 determines wherein in the image that the border between two districts comprises the edge that detects between described two districts.

4. the method for claim 1, the border between wherein said two districts are the surfaces.

5. the method for claim 1, wherein said second area comprises described border.

6. method as claimed in claim 5, wherein said border comprise the fuzzy transition between described first area and the described second area.

7. the method for claim 1 wherein selects the pixel by notable feature characterized on the boundary surface to comprise to utilize this notable feature to select angle point on this boundary surface.

8. method as claimed in claim 7 wherein utilizes this notable feature to select the angle point on this boundary surface to comprise the angle point of selecting as angle point in the process any plane of selected pixel.

9. the method for claim 1, wherein said notable feature comprise (a) first order derivative, (b) second derivative and (c) a kind of during derivative is represented.

10. the method for claim 1 is wherein determined from first seed that the first area comprises from first seed convex domain is increased.

11. the method for claim 1 is wherein determined from first seed that the first area comprises from first seed recessed zone or hole is increased.

12. the method for claim 1 wherein selects select at least one to be used to make the pixel with the region growing that is approximately circular or oval-shaped xsect borderline being comprised by the pixel that notable feature characterized.

13. the method for claim 1 is wherein selected the pixel by notable feature characterized on the boundary surface to comprise and is selected at least one to be used to make the pixel with the region growing that is approximately cylindrical or paraboloidal xsect.

14. the method for claim 1 wherein selects the pixel by notable feature characterized on the boundary surface to comprise to select at least one to be used to make the pixel of most of smooth region growing.

15. the method for claim 1, wherein the border between first area and the second area is wide and blurs.

16. method as claimed in claim 15 further comprises according to subregion wide and fuzzy edge and determines described notable feature.

17. method as claimed in claim 16 is wherein selected the pixel by notable feature characterized on the boundary surface to be included in the subregion at described wide and fuzzy edge and is discerned angle point.

18. the method for claim 1, wherein said first area and described second area all are convex domain or recessed zone.

19. one kind is used for being the definite equipment of selecting seed in zone from image, this equipment comprises:

In order to determine in the image device on the border between two districts;

In order to by selecting this borderlinely to form the device of seed list by a plurality of pixels that notable feature characterized, this notable feature is the seed that is used for determining the zone with these pixel logos;

In order to from seed list, to select first seed to determine the first area device; And

In order to from seed list, to select second seed of the part of non-first area, to determine the device of second area.

20. one kind is used for selecting seed so that the method for region growing from image, this method comprises:

Determine in the image boundary surface between two zones;

By what filter boundary surface is that the pixel of hanging down the dimension angle point also will be placed in the seed list through the pixel of filtering, and forms seed list;

From this seed list, select first seed so that the first area increases; And

From this seed list, select second seed of the part of non-first area, second area is increased.

Images