CN1971616B - Multi-resolution adaptive filtering - Google Patents

Abstract

Systems and methods which analyze an image and extract features of the image therefrom for use in filtering are shown. Based on the features and structures, embodiments determine how to filter at different orientations and with different filter configurations. Filters utilized according to embodiments are adaptive with respect to spatial and/or temporal aspects of the features. Image processing according to embodiments is performed on sub-images at various levels of resolution.

Description

Multi-resolution adaptive filtering

Cross reference to related application

The title that the application requires on November 23rd, 2005 to submit to is the common unsettled U.S. Provisional Application sequence number 60/739 of " Multi-ResolutionAdaptive Filtering "; 871 right of priority, the disclosure of this U.S. Provisional Application is incorporated into this by reference.

Technical field

Present invention relates in general to Flame Image Process, relate more specifically to reduce the Flame Image Process of picture noise.

Background technology

Speckle noise comprises the coherent noise that for example results from the excusing from death image.For example, when forming ultrasonoscopy, carry out typically that the bundle forming process-it is a relevant process-to form the supersonic beam of ultrasonoscopy from its acquisition.A plurality of bundle forming processes cause a kind of " black-white point noise (salt and pepper noise) " that is superimposed upon on the true picture information.This noise is called as " speckle noise ".Similar phenomenon also occurs in the radar.

Much human all attempts to come the filtering speckle noise according to complex technique in the ultrasonic industry.That is, thus much human attempted coming together to correct speckle noise and reducing speckle noise through handling image with different frequency bands and different frequency bands being incorporated into.

The another kind of approach that reduces speckle noise that people have attempted is to use space compound.In space compound, produce two or more images from different " view direction " or different visual angles, and said image is integrated into together so that speckle noise reaches average.

The technology that reduces aforementioned two kinds of speckle noises realizes all that usually speckle noise to a certain degree suppresses.Yet these technology neither be impeccable.For example, a plurality of less and because frequency band is divided in frequency multiplexed than narrow bandwidth signal, so it is compromise to have some axial resolutions.This arrowbandization causes axial resolution compromise.Use is slow through the frame rate that the space compound that obtains a plurality of views from different view directions and realize obtains final image.Therefore, the mobile and animation quality of realtime graphic maybe be relatively poor.

Summary of the invention

The present invention relates to through analysis image and extract the local feature of image and sef-adapting filter is applied to the system and method that these characteristics are corrected speckle noise.Be based on the different characteristic in the characteristic that identifies in the image, each embodiment confirms filter configuration, so that apply filtering to improve picture quality through suppressing speckle noise effectively at different orientation and/or with the different filter parameter.With respect to just applying the special characteristic of wave filter to it, the wave filter that is applied preferably for example on the space and/or the time go up adaptive.

Embodiments of the invention use aforementioned sef-adapting filter that the subimage that is in various horizontal resolutions is handled.For example, high-definition picture can resolve into a plurality of graphical representations, and the resolution of each graphical representation is lower than next graphical representation.Embodiment is used for the local feature in each such graphical representation is discerned, and said local feature is applied wave filter, wherein to being presented in the character pair of specific image in representing, selects the wave filter that is applied according to orientation and/or parameter.Can represent that to different images the interprocedual that applies filtering shares about the information of the characteristic in the graphical representation of image.After each graphical representation being applied filtering, preferred embodiment is according to rebuilding the image through filtering through the graphical representation of filtering.For same image, can be repeatedly (for example, times without number or changing image or image is applied when changing) carry out above-mentioned picture breakdown, the filtering of decomposing graphical representation and the reconstruction of representing through filtering image.

Can utilize various Knowledge Base to apply the sef-adapting filter of embodiments of the invention.For example; The special characteristic that the Knowledge Base that various filter parameters and characteristic aspect (for example, step function, crestal line, surperficial gradient, texture, pixel intensity gradient etc.) are associated capable of using is selected with respect to identify in the image and sef-adapting filter and/or the sef-adapting filter parameter used.In addition or can be as an alternative; Capable of using make various filter parameters with typically being presented in the special characteristic that Knowledge Base that the characteristic (for example, specific anatomical structure, particular procedure etc.) in the specific image type is associated selects with respect to recognize in the image sef-adapting filter and/or the sef-adapting filter parameter of using.

Above-mentionedly summarized characteristic of the present invention and technological merit roughly, thereby following detailed description of the present invention can be understood better.To describe additional features of the present invention and advantage hereinafter, it forms the theme of claim of the present invention.It will be understood by those skilled in the art that and easily to utilize disclosed imagination and specific embodiment as other structure being made amendment or designing to realize the basis with the identical purpose of the present invention.Those skilled in the art it will also be appreciated that such equivalent constructions does not deviate from the spirit and scope of the present invention of liking enclosed in the claim to be stated.When combining accompanying drawing to consider,, all be considered to peculiar new feature of the present invention and other purpose and advantage with understanding better for its tissue and method of work according to following description.Yet, should be expressly understood especially that each accompanying drawing only provides for diagram and purpose of description, is not will be as the qualification of limitation of the present invention property.

Description of drawings

In order more completely to understand the present invention, with reference now to description, wherein below in conjunction with accompanying drawing:

Fig. 1 illustrates according to an embodiment of the invention with self-adaptation and/or can handle wave filter (steerable filter) and is applied to image;

Fig. 2 A illustrates the 2D signal of handling in the frame that utilizes according to an embodiment of the invention and handles;

Fig. 2 B illustrates the 2D signal that utilizes interframe to handle according to an embodiment of the invention and handles;

Fig. 3 illustrates three dimensional signal according to an embodiment of the invention and handles;

Fig. 4 A illustrate according to an embodiment of the invention can be by the noise step function signal of filtering;

Fig. 4 B illustrates the noise step function signal that according to one embodiment of present invention auto adapted filtering is applied to Fig. 4 A;

Fig. 5 illustrates the noise step function signal that conventional non-self-adapting wave filter is applied to Fig. 4 A;

Fig. 6 illustrates the sef-adapting filter of one embodiment of the present of invention and how in the one-dimensional space, to work;

Fig. 7 A and 7B illustrate two-dimensional case, and the one-dimensional case ground that wherein is similar to Fig. 6 applies sef-adapting filter;

Fig. 8 A and 8B illustrate the exemplary symmetrical sef-adapting filter that applies in ridge characteristic place according to one embodiment of present invention;

Fig. 9 A and 9B illustrate the exemplary symmetrical sef-adapting filter that applies in characteristic intersection place according to one embodiment of present invention;

Figure 10 A and 10B illustrate the various instances that use at least one embodiment of the present invention can handle wave filter;

Figure 11 shows the diagrammatic representation of the wave filter of one embodiment of the present of invention;

Figure 12 illustrates edge direction and the gradient direction in the image according to various embodiments of the present invention;

Figure 13 A-13C illustrates the simplified example of available according to an embodiment of the invention gaussian filtering nuclear and implements;

Figure 14 shows the exemplary signal path of the diagnostic ultrasound system that is applicable to each embodiment of the present invention;

Figure 15 shows the functional block diagram according to the processing unit that can be used for image filtering of each embodiment of the present invention;

Figure 16 shows the details about an embodiment of the decomposition frame of Figure 15;

Figure 17 shows the details about an embodiment of the processing frame of Figure 15;

Figure 18 shows the details about an embodiment of the reconstruction frame of Figure 15;

Figure 19 illustrates suitable exemplary digital signal processor Hardware configuration according to an embodiment of the invention.

Embodiment

With reference to figure 1, show the expression that according to one embodiment of present invention sef-adapting filter is applied to image 100.Particularly, Fig. 1 comprises round 111-114 and the oval 121-127 of expression according to the exemplary filters of the embodiment of the invention.Said circle for example covers on Figure 100 top with respect to being presented in the various character pairs (for example, structure, texture, gradient, slope, function etc.) in the image with oval.In operation according to the preferred embodiment of the invention, different size, put on respect on the different orientation of image and/or adopt the above-mentioned wave filter of various parameters to be developed and be applied on the image so that smoothly fall or filter out speckle noise.Such wave filter preferably is applicable to local feature so that the filtering performance of best quality is provided.For example, the filter processor of each embodiment uses above-mentioned wave filter to make the pixel that is in diverse location reach average to avoid average not similar pixel simultaneously through average similar pixel.Correspondingly, optimum system choosing ground determines whether in filtering, to comprise certain pixel adaptively.Can utilize one or more technical foundation to select with respect to such as the special characteristic of the image of image 100 and the wave filter and/or the filter parameter that use.

With respect to image filtering is provided, embodiments of the invention are also implemented subimage and are handled.For example; The multiresolution subimage treatment step of each embodiment becomes the subimage or the graphical representation of different resolution, wherein one or more sef-adapting filters to be applied on each characteristic of being presented in each this subimage to suppress speckle noise picture breakdown.As stated, with respect to the employed wave filter of any this subimage can be different size, put on respect on the different orientation of image and/or adopt various parameters.That is, embodiments of the invention are implemented multi-filter, and said wave filter comprises and depends on the proterties that is in the characteristic in the corresponding subimage partly.In operations according to the instant invention, in case use above-mentioned sef-adapting filter that subimage is handled, the subimage of then handling is combined rebuilding said image, and said image is presented to the terminal user then or used or store as the image of filtering.

Embodiment

Example characteristic according to the sef-adapting filter of each embodiment can comprise: to flat surfaces carry out filtering and level and smooth, to inclined surface carry out between filtering, the retention surface sharp edges, the noise at ridge characteristic place is carried out filtering, keeps sharp corner etc.For example, can the pixel that same lip-deep pixel promptly has similar characteristic be averaged about the wave filter that the special characteristic with flat surfaces is implemented.Yet, if the surface is not a flat surfaces, but having the curved surface of the slope related with it, embodiment can implement along same slope pixel to be divided into groups not upset this slope with filtering so.If between the surface like top surface and side surface between (the cubical view that for example several surfaces are visible) have sharp edges; The wave filter of implementing according to embodiment is so preferably worked; To keep the acutance at said edge, make the deterioration at edge and distortion minimum thus.When characteristics of image or structure comprised ridge (like linear structure), the wave filter of implementing according to the embodiment of the invention kept the shape of this ridge, and also kept the acutance of this ridge.If, there be intersecting (intersecting) of line like two ridges, cause the turning of intersection, the wave filter of then implementing according to the embodiment of the invention keeps this turning makes it very sharp-pointed.In the operation of the preferred embodiment of the present invention, the afore-mentioned characteristics of discernible feature association is developed one or more wave filters in utilization and the image, can realize the high image quality in the spot minimizing process thus.

Note Fig. 2 A, 2B and 3, can find out that embodiments of the invention can be applicable to multidimensional.For example, the algorithm of embodiment of the present invention notion can on be applied to that one-dimensional signal is handled, two dimensional image is handled, 3-D view is handled and four-dimensional Flame Image Process.Four-dimensional Flame Image Process according to embodiment means three dimensions and time (like X, Y, Z and time).Handling according to the 3-D view of embodiment can be two-dimensional space and time (like X, Y and time) or three dimensions (like X, Y, Z).

Image sequence about Fig. 2 A shows the instance that 2D signal is handled, and wherein the t axle is the time, and X and Y axle are the Spatial Dimensions of two dimensional image.In the instance shown in Fig. 2 A, provide the Flame Image Process of the filtering of front to be provided (being to handle in the frame) individually with time series to each frame, shown here is frame 201-209.That is to say, this processing be by on the basis of frame, do not utilizing under the situation of the inter-frame information (like the information of obtaining from next frame) that can obtain and accomplishing from the time series of frame.Various embodiments of the present invention can use in interframe (spatial information in this example) and the frame (temporal information in this example) filter characteristic the two or wherein any.

Fig. 2 B shows the instance that three dimensional signal is handled.Particularly, instance shows the utilization to inter-frame information shown in Fig. 2 B.Shown in interframe signal handle, comprise continuous frame in this processings, as organizing the frame of 211-214, to improve the performance of wave filter.For example, the wave filter that the characteristic that exists in the successive frame about the group 211 of Fig. 2 B is used can be along t axle orientation, so that utilize information available in the successive frame.This frame group can wait the relevant series of frames of characteristic, the frame of any amount, mobile window and limit.Embodiments of the invention can implement to comprise the frame group of different frame, so that about various characteristics of image filtering (as be used for the group 211 of first characteristic and be used for the group 212 of second characteristic) is provided.Yet, utilize this inter-frame information about various characteristics, selected the frame group of front, can be used to provide the Flame Image Process of improvement, wherein interframe provides relevant information.

Should be appreciated that aforementioned concepts can be used about the spatial information of different dimensions, and therefore be not limited to the two-dimensional space Flame Image Process.The instance of Fig. 3 shows the application of aforementioned concepts about the three dimensional object sequence.For example, this blocks of data obtains at different time if image comprises the three-dimensional bits data, then can obtain piece image sequence (being depicted as 301 and 302 here), and applied aforementioned concepts can be improved picture quality.

Following equality provides describes the wave filter formula according to two types of wave filters of each embodiment.

$G_s(r(x,y,z,t),g(x,y,z,t),\&dtri;g(x,y,z,t),f(x,y,z,t),...)=e^{-\left(\frac{{\mathrm{<figure-callout\; id="2"\; label="r"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">r</figure-callout>}}^2}{{\mathrm{\&sigma;}}_r{(g,.)}^2}\right)}{e}^{-\left(\frac{{\left|{\&dtri;}_g\right|}^2}{{\mathrm{\&sigma;}}_g{(g,.)}^2}\right)}{e}^{-\left(\frac{{|\&dtri;f|}^2}{{\mathrm{\&sigma;}}_f{(g,.)}^2}\right)}...\left(1\right)$

$G_{\mathrm{\&alpha;}}(u,v,w,t,g(x,y,z,t),\&dtri;g(x,y,z,t),f(z,y,z,t),...)=e^{-(\frac{u^2}{{\mathrm{\&sigma;}}_{\mathrm{\&mu;}}{(g,.)}^2}+\frac{{\mathrm{<figure-callout\; id="2"\; label="v"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">v</figure-callout>}}^2}{{\mathrm{\&sigma;}}_{\mathrm{\&nu;}}{(g,.)}^2}+..)}{e}^{-(\frac{{|\&dtri;g_u|}^2+|\&dtri;g_v{|}^2}{{\mathrm{\&sigma;}}^{2}g(g,.)}+..)}{e}^{-\left(\frac{{|\&dtri;f||}^2}{{\mathrm{\&sigma;}}^{2}f(g,.)}\right)}...\left(2\right)$

Above shown in first kind wave filter (equality (1)) comprise balanced-filter.Above shown in second type of wave filter (equality (2)) comprise asymmetric filters, wherein u is the dominant orientation of characteristic.The two all can be adaptive for symmetry and asymmetric filters.The wave filter formula of example is the function of a plurality of different parameters.Here, r, x, y and z comprise spatial information, and t comprises temporal information, and g comprises gray scale (grayscale) information (like, difference half-tone information), and f is generic items (like, other relevant informations).

Should be appreciated that, the spot that takes place according to the embodiment of the invention reduce coordinate system in the operation can be polar coordinate system (as, use radial coordinate, for example radius or apart from r), perhaps can be the Di Kaer coordinate system (as, use coordinate along X, Y and Z axle).Ultrasound information is to visit the data that direction is obtained from difference basically, so that said data set is synthetic according to polar image.Yet most of display modes all are Di Kaer, because linear array rectangle normally.Scanner head uses polar coordinates usually; And display uses rectangular coordinate usually, usually exists according to the polar coordinate system of using about the view data of being obtained through the ultrasonic system scanner head and according to about come the process of reconstruction of the rectangular coordinate system that the images displayed data use through ultrasound system display.That is to say that most ultrasound systems converts polar coordinates to rectangular coordinate (being called scan conversion process).Said filtering can be applicable to polar coordinate space and/or rectangular coordinate space.

Equality (1) right-hand side is described to be to use cascade (like, a plurality of different Gaussian filters).Gaussian filter is cascaded into single filter, is called G _sTherefore shown filter equations is that symmetry is with consistent in all dimensions about apart from the r symmetry.Delta g (Δ g) is the embodiment of equality (1), comprises gradient information, as in the different dimensions about the gradient of gray scale.For example, the f function is provided in example embodiment, to adapt to general purpose function.Should be appreciated that,,, can use more than such f function as when having a plurality of different additional correlation information according to embodiments of the invention.

Equality (2) provides asymmetric filters, and this equality is not consistent in all dimensions like this.That is to say that this wave filter is with respect to selected orientation, as along applying filtering with this feature axis of comparing of axle with the feature axis quadrature differently.Use such asymmetric filters according to the embodiment of the invention, particularly useful on the aspects that keeps characteristic in the image of filtering.The same with above-mentioned equality (1), above-mentioned equality (2) use cascade (as, a plurality of different Gaussian filters).Should be appreciated that the asymmetric filters of example can resolve into different orientations.The wave filter orientation can be confirmed so that extract one or more characteristic orientation through analyzing local feature.Then, can preferably use said wave filter along the dominant orientation of characteristic.

Note Fig. 4 A and 4B, show the illustrative examples of the application of auto adapted filtering according to an embodiment of the invention.In order to simplify described notion, the instance of Fig. 4 A and 4B provides the explanation of one-dimensional signal being handled auto adapted filtering.Shown in Fig. 4 A, having one is the signal of step function, is shown step function signal 401.If noise is introduced into this step function signal, shown in noise signal 402, then this step function signal becomes noise step function signal, shown in noise step function signal 403.Should be appreciated that, for example, the single characteristic that the signal indication among Fig. 4 extracts from picture signal.In given signal, can exist and treat other possible characteristics of extracting based on similar tracking mode.

Be shown the sef-adapting filter of one embodiment of the invention of wave filter 410 among Fig. 4 B; Preferably be applied to noise step function signal 403; So that removal noise; Thereby present the signal through filtering, as directed step function signal 404 through filtering is similar to original signal (the step function signal 401 shown in Fig. 4 A).Application is in order to suppress noise according to this wave filter of embodiment, and keeps the edge of step function.Therefore, the sharp edges of step function remain in illustrated embodiment in the step function signal 404 of filtering.Therefore, be useful especially according to the sef-adapting filter of the embodiment of the invention about characteristic boundary, as keeping sharp features edge, turning, lines etc.

Fig. 5 shows traditional non-self-adapting filter applies in aforementioned noise step function signal, so that the advantage of the sef-adapting filter of the embodiment of the invention is shown.Wave filter among Fig. 5 does not fit into signal.The gaussian kernel that is shown filter kernel 510 and is referred to herein as G (h) is represented traditional non-self-adapting wave filter.At the edge of step function, there is gaussian kernel G (h).Equally, leaving several points at said edge, also there is identical gaussian kernel G (h).Therefore, apply gaussian kernel and signal is averaged, conventional filter provides the step function signal 504 through filtering, has wherein not only smoothly fallen noise, has also smoothly fallen the edge of step function, and is therefore no longer sharp-pointed.Particularly, level and smooth through what wave filter provided, the acutance at the edge at 521 and 522 places, zone has reduced now, and the ultrasonoscopy that therefore forms according to this signal will provide the edge that seems obscurity boundary.

Compare with the traditional non-self-adapting wave filter among Fig. 5, the sef-adapting filter of Fig. 4 B has kept the sharp edges of step function characteristic.Fig. 6 shows the sef-adapting filter 410 of one embodiment of the invention and how can in the one-dimensional space, work.Particularly, Fig. 6 shows the characteristic of being extracted (being basic step function) here and how to be used for controlling applied wave filter and to reduce noise.

Represent by the filter kernel 611-615 among Fig. 6 about noise step function signal 403 employed filter kernel at any specified point.Different with the filter kernel of Fig. 5 is that the filter kernel of Fig. 6 is adaptive.Therefore, when filter kernel during near step function, filter kernel is suitable for corresponding with the step function characteristic.Notice that the shape of filter kernel 613 and 612 represented gaussian kernel has sharp-pointed edge in the left side, then level and smooth as filter kernel shown in Figure 5 on the right side.Particularly, in an illustrated embodiment, half is set at 0 to filter kernel 613 with the pact of filter kernel, and so only half the gaussian kernel is applied to the filtering noise signal.Yet filter kernel 612 is set at 0 with about 1/3rd of filter kernel, and so only about 2/3rds gaussian kernel is applied to the filtering noise signal.When wave filter when the step function edge is removed, filter kernel become with Fig. 6 in similar.At the sef-adapting filter of step function edge use illustrated embodiment, the left-hand side of signal is with the right-hand side of inequality to signal.

In according to operation embodiment illustrated in fig. 6; When the step function edge by near the time; Sef-adapting filter will reduce weight automatically, or change filter coefficient so that the signal on the step function edge right-hand side the signal on step function edge left-hand side is not average.Should be appreciated that the aforesaid filters function can be applicable to any signal of being checked, no matter whether there is enough the conversion of signals that comprises " edge " greatly.Yet in case characteristic or structure have enough comprised the edge greatly, the weight of wave filter is preferably so handled and is made the pixel of either side at edge not be averaged in to together.In other words, embodiment operates and seeks this edge, and when finding an edge, locates to adjust the weight of wave filter on the edge of, makes that weight at this this wave filter of edge is zero or near zero.Should be appreciated that this edge can be limited in the hyperspace, be not limited to top illustrated one dimension instance so be operated in the above-mentioned notion at edge.In addition, the latitude that wherein limits this edge is not limited to space boundary, and can be time qualified.

Fig. 7 A and 7B show the situation of two dimension, and be wherein similar with the one-dimensional case of above-mentioned Fig. 6, used auto adapted filtering.Under the situation of two dimension, although be two-dimentional, embodiments of the invention also carry out as described above.For example, shown in Fig. 7 A, can go up and use complete Gaussian filter (being shown filter kernel 711) at the flat surfaces of two-dimentional noise step function signal (being shown noise step function signal 703) (like, upper and lower plateau).But, near the step function edge time, filter kernel preferably reduce filter weight (as, when wave filter during near the step function edge, filter kernel 712 makes about 1/3rd of filter kernel be set to 0), right-hand side will only carry out smoothly the top like this.Therefore, the information on the left-hand side will be not on the edge information on the right-hand side average.Fig. 7 B shows after having used sef-adapting filter, has kept the sharp edges of step function through the step function signal 704 of filtering.

As stated, by cascade (in above-mentioned instance, being the Gaussian function of cascade), this provides a kind of difficult relatively wave filter about the employed equality of the embodiment of aforementioned sef-adapting filter (1) and (2).But if local feature is an edge, filter kernel preferably will reduce weight automatically so.When weight reduces; The wave filter of embodiment is located not being a complete nuclear on the edge of, shown in the filter kernel 612-615 of Fig. 6, therefore; Compare with complete filter kernel, see that drawing applied wave filter is so not overcritical wave filter for signal Processing.

As stated, the filter class of equality (1) and (2) expression comprises balanced-filter (equality (1)) and asymmetric filters (equality (2)).Difference is that balanced-filter has an orientation that puts on it, and asymmetric filters is isotropic to all directions.Therefore use wave filter that equality (1) is provided with filtering to be provided, and use wave filter that equality (2) is provided with filtering to be provided as the function of gradient as the function of distance.Above-mentioned notion can provide about arbitrary filter class (as, according to characteristic about distance and/or adjust filter weight about character gradient).

Fig. 8 A and 8B show according to an embodiment of the invention the typical symmetrical sef-adapting filter at the ridge place.According to the filter adaptation of illustrated embodiment in characteristics of image.Shown in Fig. 8 A, complete Gaussian filter (being shown filter kernel 811) can be applicable to the flat surfaces (like, background) of noise ridge signal (being shown noise ridge signal 803).But; Filter kernel preferably reduce at the ridge place filter weight (as; When filter applies during in the ridge characteristic, filter kernel 812 makes each of the right about 1/3rd of the left side of filter kernel about 1/3rd and filter kernel be set at 0), so only the top of ridge characteristic is by smoothly.Therefore, from the information of background surface on the ridge characteristic left side and the right will be not with ridge on information average.Fig. 8 B shows after having used sef-adapting filter, in the ridge signal of filtering, has kept the sharp edges of ridge.

Fig. 9 A and 9B show according to an embodiment of the invention at the characteristic infall typical symmetrical sef-adapting filter of (like, the intersection of ridge).The filter adaptation of illustrated embodiment is in the characteristics of image near the center.Shown in Fig. 9 A, on the flat surfaces (like background) of noise ridge crossbar signal (being shown noise ridge crossbar signal 903), can use complete Gaussian filter (being shown filter kernel 911).But, filter kernel preferably ridge and ridge infall reduce filter weight (as, filter kernel 912 makes the corresponding part of ridge cross section this filter kernel and filter application be set at 0).Should be appreciated that, compare with filter kernel 812 that the shape of the filter kernel 912 of illustrated embodiment is more complicated, for example, it keeps beyond the edge of ridge, also keeps the turning that ridge intersects.Therefore, from the information of the background surface on the ridge characteristic left side and the right will be not with ridge on information average.Fig. 9 B shows after having used this sef-adapting filter, in the ridge crossbar signal 904 of wave filter, has kept the sharp edges of ridge and the sharp edges at ridge intersection turning.

As stated, according to preferred embodiment, sef-adapting filter attempts to keep the turning, and the acutance of the ridge of the cross correlation of reservation and characteristic.Although generally speaking sef-adapting filter is worked well in the described process of face in realization, also there is limitation in some applications.Particularly, utilize balanced-filter, filtering weighting is the function of local feature, and wherein when weight was suppressed, effectively filtering core was big or small less.As a result, near the filter effect of edge feature less than lip-deep filter effect.This phenomenon is found in Fig. 8 A, and wherein complete Gaussian filter has been used in the background area, and has only used the subclass of Gaussian filter along the edge of ridge.Therefore, the applied filtering amount of the different piece of image is different, cause in the signal of filtering near the zone of ridge with compare away from the zone of ridge keeping more noise.Fig. 6 A and 9A show about this phenomenon of individual features wherein.

Therefore, embodiments of the invention have been implemented a kind of steerable filtering, to compensate the unequal filter effect at aforementioned edge feature place.Steerable wave filter can be extracted or classify, and optimum system choosing ground with these filter applies in interested orientation, to add filtering around on the edge of.For example, can provide steerable wave filter to use, provide steerable wave filter to be applied to background and can intersect the edge that is limited along the background plane of noise ridge signal 803 and ridge along the ridge of the noise ridge signal 803 of Fig. 8 A.In order to improve performance, except balanced-filter, various embodiments of the present invention also comprise steerable asymmetric filters.In this embodiment, system is applied to this characteristic based on the orientation of characteristic with asymmetric filters.

Figure 10 A and 10B show the steerable wave filter of various instances that is used at least one embodiment of the present invention.According to preferred embodiment, confirm to treat local feature in the signal of filtering orientation (as, can confirm the dominant angle θ of characteristic).The steerable wave filter that preferably establishment orientation is consistent with the orientation of local feature (as, with the asymmetric steerable wave filter of angle θ orientation).Then, on determined orientation, use steerable wave filter, so that the image through filtering is provided.

Although the diagram of Figure 10 A and 10B is illustrated in the XY space and carries out space (spatialcentering) placed in the middle, identical notion can expand to other dimensions (as, Z axle and time shaft).For example, each embodiment applicable to four-dimensional (as, X, Y, Z and time).Should be noted that one or more aforementioned dimensions need not to be on the space, therefore can comprise time, intensity etc.When having a plurality of dimension (as: 3 dimensions, the XYZ that is surveyed), system can make different filter on different dimensions, be orientated differently.

According to various embodiments of the present invention, following equality is represented steerable asymmetric filters, is used for relative simple two-dimensional situation.

$G(u,v,g,\&dtri;g)=e^{-(\frac{{\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">u</figure-callout>}}^2}{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">u</figure-callout>}}^2}+\frac{{\mathrm{<figure-callout\; id="2"\; label="v"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">v</figure-callout>}}^2}{{\mathrm{\&sigma;}}_{\mathrm{\&nu;}}^2})}{e}^{-\left(\frac{{|\&dtri;g_u|}^2+{|\&dtri;g_v|}^2}{{\mathrm{\&sigma;}}_g{(g,.)}^2}\right)}---\left(3\right)$

Wherein, the v in the equality (3) is the gradient direction vertical with edge feature, and the u in the equality (3) is parallel with edge feature, and is shown in figure 11.In the embodiment of equality (3), two exponential expression are arranged, be two gaussian representation formulas here, derive from above-mentioned asymmetric filters equality (2).The first gaussian representation formula comprises u and v, and the orientation of characteristic is shown.Next gaussian representation formula represent along u direction gray scale (g) gradient and along the character gradient of v direction.The expression that ellipse shown in Figure 11 provides shows the wave filter that obtains.In the embodiment shown, wave filter stretches along u direction, and in the lesser extent upper edge vertical v direction.Suppose that this is a Gaussian filter, Gauss's stretching, extension is described through Sigma u and Sigma v.The Sigma v of illustrated embodiment is littler than Sigma u.Sigma is the orientation of this u axle about the XY axle, and the XY axle is the quality of image, and u is the direction of characteristic.Sigma g is the stretching, extension of this particular gaussian function.

The uv space of illustrated embodiment is a feature space, is rotational transform, can be shown below.

$\left[\begin{array}{c}u\\ v\end{array}\right]=\left[\begin{array}{cc}\cos \mathrm{\&theta;}& \sin \mathrm{\&theta;}\\ -\sin \mathrm{\&theta;}& \cos \mathrm{\&theta;}\end{array}\right]\left[\begin{array}{c}x\\ y\end{array}\right]---\left(4\right)$

Like this, according to following equality, steerable wave filter can be illustrated in the image space, wherein supposes λ _v＞＞λ _uAnd |

g _v|＞＞|

g _u|.

$G(x,y)-e^{-(\frac{{(x\cos \mathrm{\&theta;}+y\sin \mathrm{\&theta;})}^2}{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">u</figure-callout>}}^2}+\frac{{(-x\sin \mathrm{\&theta;}+y\cos )}^2}{{\mathrm{\&sigma;}}_{\mathrm{\&nu;}}^2})}{e}^{-\left(\frac{{|\&dtri;g_u|}^2+{|\&dtri;g_v|}^2}{{\mathrm{\&sigma;}}^{2}g\left(g\right)}\right)---\left(5\right)}$

According to more than, but should understand orientation according to system's recognition feature of embodiment, and for example use equality (3) or (5) defined filter kernel that filter adaptation is become the specific direction work along this characteristic.This direction can be the direction of this characteristic itself.According to embodiments of the invention, the surface can be a characteristic, and gradient itself can be a characteristic, and the position of structure can be a characteristic, or the like.

In the uv space, function G can be expressed as:

$G(u,v)=e^{-(\frac{{\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">u</figure-callout>}}^2}{{\mathrm{\&sigma;}}_{\mathrm{<figure-callout\; id="2"\; label="u"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">u</figure-callout>}}^2}+\frac{{\mathrm{<figure-callout\; id="2"\; label="v"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">v</figure-callout>}}^2}{{\mathrm{\&sigma;}}_{\mathrm{\&nu;}}^2})}{e}^{-\left(\frac{{|\&dtri;g_v|}^2}{{\mathrm{\&sigma;}}^{2}g(g,.)}\right)---\left(6\right)}$

Suppose λ _v＞＞λ _uAnd v is a gradient direction, and makes σ _u＞＞σ _v, the gradient of equality (6) can be represented as follows.

$G(u,v)\&ap;e^{-\frac{{|\&dtri;g_v|}^2}{{\mathrm{\&sigma;}}^{2}g(g,.)}}=e^{-\frac{{\left(\stackrel{\&OverBar;}{n}(x,y){\stackrel{\&OverBar;}{n}}_{\left|\right|}\right)}^2}{{\mathrm{\&sigma;}}_u^2}}{e}^{-\frac{{|\&dtri;g_v|}^2}{{\mathrm{\&sigma;}}^{2}g(g,.)}---\left(7\right)}$

Wherein

is the vector that is parallel to edge feature; And is (x, the point of y) locating in the filtering district.

Gradient direction shows that steepness changes in two-dimensional space.In other words, when gray level changes, can resemble and describe it the landform.Gradient is bigger in steep side.Typically, on said specific direction, coming smoothed image is unfavorable (for example avoiding " falling down from steep cliff ").Therefore, embodiments of the invention are used smooth function on the direction different with the direction of the gradient of steepest.For example, if the greatest gradient direction is gradient G v, system comes filter application along u direction, because maximum gradient is on the v direction.In other words, smoothing filter is used with the maximum vertical direction of direction of gradient in various embodiment edge.

According to embodiments of the invention, can obtain the edge feature orientation through from the defined Hai Sai of Jacobian (Jacobian) (Hessian) matrix of brightness step, obtaining eigenvector.Hessian matrix M representes as follows.

$M=\left[\begin{array}{cc}\frac{d^{2}J}{d{x}^2}& \frac{d^{2}J}{\mathrm{dx\; dy}}\\ \frac{d^{2}J}{\mathrm{dy\; dx}}& \frac{d^2\mathrm{<figure-callout\; id="2"\; label="J\; d\; y"\; filenames="H061E4939020061205E000131.png,H061E4939020061205E000151.png"\; state="\{\{state\}\}">J</figure-callout>}}{d{y}^{}}\end{array}\right]=\left[\begin{array}{cc}{J}_{\mathrm{xx}}& J_{\mathrm{xy}}\\ J_{\mathrm{yx}}& J_{\mathrm{yy}}\end{array}\right]---\left(8\right)$

Before differentiation, can at first come regularization input picture I (J=G*I) through Gaussian filter G.The eigenvalue of hessian matrix M and eigenvector can be through calculating and expression to get off.

$v_{\&perp;}=\left[\begin{array}{c}\cos \mathrm{\&theta;}\\ -\sin \mathrm{\&theta;}\end{array}\right]u_{\left|\right|}=\left[\begin{array}{c}\sin \mathrm{\&theta;}\\ \cos \mathrm{\&theta;}\end{array}\right]---\left(9\right)$

In equality (9), at λ _v＞λ _uThe time, v is perpendicular to the vector at the edge that is orientated with the angled θ of axle, and u is parallel to said edge.

Although above-mentioned instance utilizes eigenvector to come the location feature edge, the technology that embodiments of the invention can be realized being used for the additional of location feature edge or replace choosing.For example, can utilize various known digital image processing techniques, computer vision signal processing technology, form Flame Image Process to wait location feature according to embodiments of the invention.For example, embodiments of the invention can implement to be used for the fuzzy logic of location feature, and wherein fuzzy logic controller is analyzed the various attributes of inferring characteristic so that draw best features coupling conclusion.

Figure 12 illustrates according to the edge direction of various embodiments of the invention and gradient direction.Image shown in Figure 12 is with reference to two dimensional image more complicated shown in Fig. 7 B, 8B and the 9B than above.In the embodiment of Figure 12, the third dimension is a gray scale.Some steep variations on the gradient direction indication two-dimensional surface shown in the parallel arrow.Perpendicular to this gradient direction is edge direction.According to preferred embodiment, along edge direction application self-adapting wave filter.Should be understood that and to come the calculating filter orientation based on the mathematics of equality discussed above (6)-(9).

Should be understood that the combination of the various primitive features of describing when the object of representing among Figure 12 comprises early (primitive feature).Particularly, the object of Figure 12 has edge, ridge and slope.When object when (locality) has such characteristics combination in a place; The sef-adapting filter of implementing according to embodiments of the invention can be the combination that is adapted to the various filter configuration of each characteristic in the characteristic, the for example combination of above-described filter configuration.

Figure 13 A-13C illustrates the simple examples property enforcement of the Gaussian filter nuclear that can utilize according to the embodiment of the invention.Figure 13 A illustrates one dimension Gaussian filter nuclear, and Figure 13 B illustrates 2-d gaussian filters device nuclear.Shown in instance in, in continuum (continuum), implement Gaussian filter nuclear and comprise a large amount of points.Gaussian filter nuclear utilizes central limit theorem to be approximately binomial expansion.Till a large amount of points that central limit theorem has instructed figure can be extended to Gaussian function are determined.According to this theorem, Gaussian function (Gaussian) can be come approximate by binomial nuclear, and this binomial is endorsed through the person of closing on (neighbor) is asked on average repeatedly and produced.

Figure 13 C illustrates, and the two dimensional filter of Figure 13 B can handled on the four direction at least.Particularly, Figure 13 C representes a simple wave filter, and it describes the wave filter handled on the two-dimentional four direction.Certainly, this notion can be applicable on the direction (for example six direction) of any number.The value of a in the filter kernel of Figure 13 C, b, c, d, e, f and g can be according to the rectilinearity of Gaussian function (for example by σ _uAnd σ _vParameter) define.The main direction of embodiment shown in Figure 13 C is aa.The direction of three points will be bab, and all the other directions are inserted according to Gaussian function and different coefficients different sigmas.

In operation according to the embodiment of the invention, system check characteristic and to the information block of the criterion (for example pixel similarity) that meets a certain type so that filtering.If meet this criterion, then corresponding pixel preferably is included in this filter process.Otherwise this specific pixel is not included in this filter process.In other words, algorithm of the present invention can be operated and check pixel, and if this pixel enough not similar with its pixel of next-door neighbour, then those pixels are not together and are asked average.Yet if edge orientation is very similar, embodiment can ask average with pixel together.Usually, if pixel near the surface, it belongs to flat site more, with these pixels together filtering be desirable.If there is steep variation, for example they are in not same district, and then filtering is normally unfavorable together with pixel.

Although with reference to single characteristic above-mentioned instance has been discussed, should have been understood picture signal and can comprise a plurality of characteristics.Therefore, embodiments of the invention can be used for discerning the various characteristics in such characteristic and are directed against such feature selecting as previously discussed and/or use one or more wave filters.And for optimizing image filtering, embodiments of the invention are with respect to providing image filtering to realize the subimage processing.As previously discussed, embodiments of the invention become picture breakdown the subimage or the graphical representation of different resolution.Above-mentioned self-adaptation with can handle the various characteristics in the one or more subimages that are applied to as being present in each in the wave filter.Although to any such subimage and the wave filter that uses can have different sizes, to use with different orientation, and/or adopt various parameters with respect to image, each such wave filter can that kind as discussed above be selected and is used.

Figure 14 illustrates the exemplary signal path of the diagnostic ultrasound system 1400 that is suitable for various embodiment of the present invention.Yet should be understood that the present invention is not limited to any specific signal path.Shown signal path comprises scanner head 1401, as comprising ultrasound transducer array well-known in the art.Other embodiment can replace to various circuit with scanner head 1401, like the aerial array among the radio frequency embodiment.Front-end circuit 1402 can for example provide the conversion of modulus and digital and analogue signals, wave beam to form and/or other front-end processing such as being provided as front end special IC (ASIC).Signal processor 1403 can for example provide signal filtering, synthetic aperture formation, frequency compounding, the Doppler of certain level to handle and/or other senior characteristic such as being provided as digital signal processor (DSP).Back-end circuit 1405 can for example provide scan conversion, vision signal output etc. such as being provided as rear end ASIC.Display 1406 is such as comprising CRT display system, liquid crystal display systems etc.; User interface is provided so that information is shown to the user, like the video image that produces by the ultrasonic signal of handling through scanner head 1401, front-end circuit 1402, signal processor 1403 and back-end circuit 1405.About the additional detail of the ultrasonic system that is suitable for using with the signal path that comprises scanner head, front-end circuit, signal processor and back-end circuit according to the embodiment of the invention at United States Patent (USP) the 5th; 722; Shown in 314 and describe, it is open to be incorporated into this by reference.

In exemplary signal path shown in Figure 14, the auto adapted filtering of one embodiment of the invention is externally carried out among the DSP 1404.Particularly; The outside DSP 1404 of illustrated embodiment docks with back-end circuit 1405 so that therefrom receive digital image information; No matter be before or after the scan conversion of back-end circuit 1405, and will be provided to back-end circuit 1405 through the digital image information of filtering.Yet, for selecting embodiment in other circuit, to realize auto adapted filtering, no matter be inner or outside, and/or whether use with compuscan at the diagnostic ultrasound system signal path with interrelating.For example, if desired, the auto adapted filtering of the embodiment of the invention can be provided as the part of signal processor 1403.

According to one embodiment of present invention, outside DSP 1404 works under aforesaid self-adaptation and the control of software that can handle filter kernel realizing.Particularly; The outside DSP 1404 of an embodiment has implemented to be used for the algorithm of one or more characteristics of discriminating digit picture signal; Confirm the orientation of these characteristics, select and/or the configuration filter kernel so that be applied to said characteristic, and these filter kernel are used for said picture signal.According to embodiments of the invention, when the subimage that Flame Image Process is provided was handled, outside DSP 1404 can additionally provide the multiresolution to picture signal to decompose and the multiresolution through the signal of filtering is rebuild.

The embodiment of outside DSP 1404 can comprise knowledge base 1414 or communication with it, this knowledge base memory filter configuration information, filter kernel parameter selection criterion, filter kernel parameter and/or useful out of Memory at exploitation, configuration and application self-adapting and can handle wave filter the time.For example, knowledge base 1414 can be stored one or more filter kernel configurations, parameter etc. and the information that is associated like the ad hoc structure that can in picture signal, discern.The information of using when additionally, or can be as an alternative, knowledge base 1414 can be stored in identification ad hoc structure, structural approach etc.

Can utilize the higher level knowledge storehouse according to embodiments of the invention, information wherein or its certain part are by index or can be according to contextual access.For example, diagnostic ultrasound system 1400 can be used to a plurality of predetermined operating process or pattern, like heart scanning, kidney scanning, UGI scanning etc.Knowledge base 1414 can be stored the various operating process in these operating process or the pattern or pattern customization or unique information; Make and to dispose ultrasonic system 1400 so that when being used in the selected process as the user, the related part of visit knowledge base 1414 so as to obtain to be used for being identified in the typical ad hoc structure of such process, the information of typical structural approach, the filter parameter that customizes for one or more filter kernel configurations of such process customization, for such process etc. in such process.For example, can come the characteristic in the recognition image signal such as the above-mentioned fuzzy logic of use, and the specific filter that addressable knowledge base is selected to use to this characteristic is examined and/or filter parameter.Which type of if there is characteristic might be present in priori in the picture signal (for example through selected operator scheme or the particular procedure carried out), then can this information be considered in feature identification and/or wave filter are selected to confirm).Certainly, the knowledge base 1414 of embodiment can additionally or can comprise the information that has wide applicability or do not customize to any specific context as an alternative, so that adapt to not predetermined purposes.

Figure 15 illustrates the functional block diagram of processing unit according to various embodiments of the present invention, as can be corresponding to the outside DSP 1404 of Figure 14.In an illustrated embodiment, pretreatment component 1500 provides the processing that input image data is carried out, like the pre-filtering that can comprise certain level, mapping treatment or can and can handle other processing of carrying out before the filtering in self-adaptation of the present invention.

In an illustrated embodiment, after pre-service, picture signal is resolved into the multi-resolution representation (subimage) of image by block of decomposition 1501.An embodiment can have the N of reaching number of sub images, makes input picture can resolve into N number of sub images (should be understood that the original image that therefrom decomposites other subimage can be used as " subimage " that is used for said filtering and comprise).For example; Block of decomposition 1501 can begin from the high resolution graphics image signal; This signal decomposition is become the first half the exploded view image signal of original signal resolution; This first exploded view image signal is resolved into first decompose the second exploded view image signal of image signal resolution half the (original signal resolution 1/4th), or the like, so that the N number of sub images of a half-resolution that respectively has the next son image resolution ratio is provided.For example, consider that the image that has 128 pixels in each dimension, next level of resolution will be 64 * 64, be 32 * 32 then, then 16 * 16, then 8 * 8, or the like.Decompose according to the preferred embodiment of the invention and carry out (for example from the highest resolution to the lowest resolution) from top to bottom.

Should be understood that the present invention is not limited to level of resolution or the number of decomposition levels between the subimage.Equally, the present invention's mode of being not limited to decompose.Therefore, can use the whole bag of tricks of exploded view picture, comprise the various alternate manners of wavelet decomposition and known now or later exploitation.

Notion according to the multi-resolution image of the embodiment of the invention is handled can be explained through human eye.If the observer stands in the place of 10 feet of range images, resolution will be lower, and what seen is structure or the global characteristics in the image.Yet, in case the observer near in 1 foot of range image for example, the observer will see the more details in the image, cost possibly be to can't see global characteristics.

Realize the abstract of varying level according to embodiments of the invention, be used to discern the various characteristics in said level abstract, for example global characteristics, the more characteristic and the characteristic of localization highly of localization, and to its application filtering.For example, the side that embodiment can detected characteristics, and use speckle reduction filter (speckle reduction filter) at the different aspect of characteristic.System can use the subimage of low resolution to extract the global characteristics of image and use the subimage of high-resolution to extract the details that will keep.

Refer again to Figure 15, block of decomposition 1501 provide image to the decomposition of subimage so that carry out filter process.Be shown H in the embodiment shown ₀To L _N-1The output of block of decomposition 1501 be provided for processing block 1502, it provides according to the self-adaptation of the embodiment of the invention and/or can handle filtering.Therefore, according to shown in embodiment, the filter applies that in processing block 1502, will discuss than early the time just is in image.

The processing block 1502 of preferred embodiment has been explained a kind of dependence.That is to say, except the corresponding subimage that offers processing block, but about also being provided for processing block (for example the information about the characteristic handled by processing block 1502a is sent to processing block 1502b) from the information of the characteristic of low resolution piece in the time spent.The processing basis that this additional information provides guiding that the high-resolution subimage is carried out from the low resolution subimage.Therefore, the preferred embodiments of the present invention provide the image filtering (for example from the lowest resolution to the highest resolution) that up uses processing block 1502 to carry out the end of from.Up the processing end of from like this provides the economy of advantage and processing when getting into local characteristic with the height local characteristic at the identification global characteristics.

Be shown P in the embodiment shown ₀To P _N-1The processing subimage output of processing block 1502 be provided for and be used for the reconstructed block 1503 that multi-resolution image is rebuild.That is to say that the reconstructed block 1503 of embodiment provides the combination (inverse process of for example decomposing) of subimage.The system of illustrated embodiment is used for the image through filtering of human user with generation from the output of each processing block 1502 with the aptitude manner combination.Image reconstruction according to the embodiment of the invention can be implemented up-sampling and combination.For example, the low resolution subimage resolution and two number of sub images that can be up-sampled to next high-resolution subimage can be combined.Such up-sampling and combination can repeated reaching be arrived till the resolution of original image.Therefore, the preferred embodiments of the present invention provide the image reconstruction (for example from the lowest resolution to the highest resolution) that up uses reconstructed block 1503 to carry out the end of from.Yet should be understood that the mode that the present invention is not limited to make up, because any mode known or later exploitation now can be used in one or more embodiments.

After-treatment components 1504 can be used for providing as required additional signals to handle after image reconstruction.For example, after handling image, possibly it is desirable to increase brightness, (remap) gray scale that remaps, add filtering so that look after medical treatment or the like according to the present invention.Therefore, the sub-fraction of the aftertreatment that provided of after-treatment components 1504 during normally overall signal is handled.

With reference to what Figure 14 discussed, knowledge base 1414 can be used for storing various exclusive datas as above.The processing of in processing block 1502, carrying out usually depends on characteristic, expection from the various attributes of image etc., and therefore knowledge base 1414 can provide Useful Information when coming customized treatment to various characteristics.It also possibly be desirable that certain control of the processing of type that the type that depends on employed scanner head, employed emission are used etc. is provided.For example, cardiology can be used the cardiology specialized apparatus.Use is carried out the adjusting specific to application to this processing, the additional parameter that said knowledge base provides processing block to utilize from the information of knowledge base 1414.Knowledge base can comprise the priori of for example forming about available image.For example, when known image will be the image about heart, but embodiment application specific algorithm is used for to cardiac imaging this application better.

Figure 16 illustrates the details about an embodiment of block of decomposition 1501.In the embodiment of Figure 16, use decimation filter (decimation filter) to produce the subimage of a half-resolution of resolution with next higher-order image or subimage.For example, to the original subimage 1602 that has the picture signal of making an uproar to use decimation filter 1600 to have a half-resolution with generation.This subimage is used as the input of next decimation filter so that produce another subimage.This subimage also is used as the input (being shown subimage 1603) of interpolation filter 1601.This interpolation filter is used to rebuild the level and smooth version of original image.The version that this is level and smooth deducts from original image to produce the high pass version of original image.

Figure 17 illustrates the details about an embodiment of processing block 1502.As discussed above, the sef-adapting filter of embodiments of the invention is applied to processing block 1502.Therefore, input signal is one of subimage.But in the time spent, local feature can be extracted from the information of input of this subimage and the relevant characteristic from the processing of low resolution subimage by feature extraction piece 1701.The information of features relevant is provided to up-sampler piece 1704 by the feature extraction piece 1701 of illustrated embodiment, so that characteristic information is provided to the employed processing block of higher-order resolution subimage.The information of features relevant also is provided to filter configuration piece 1703 by feature extraction piece 1701.Filter configuration piece 1703 preferably with characteristic information with can use from the information combination that knowledge base obtains so as to select, configuration and/or calculate and be used to be applied to one or more self-adaptations and/or can handle wave filter, go through as above.Filter configuration piece 1703 determined one or more wave filters are applied to subimage by filter block 1702.Should be understood that filter block 1702 applied self-adaptations and/or space-time wave filter need not be single filter.For example, according to embodiments of the invention, after connect the balanced-filter of asymmetric filters cascade can be applicable to subimage.

As shown in, 1502 outputs of the processing block of Figure 17 are about the information (in due course) of the characteristic of extracting that is used for using at next high-resolution processing block, and are provided for processing block 1502 about the information (in due course) from the characteristic of low resolution processing block.Preferably be used in combination so that calculate or select adaptive filter coefficient about the information of the characteristic of low resolution piece with information about the characteristic extracted in current level of resolution.In this example, employed characteristic is from two mid-resolution grades, and this helps to provide the consistance between the different disposal piece.

Figure 18 illustrates the details about an embodiment of reconstructed block 1503.The embodiment of Figure 18 illustrates how the output of subimage processing block 1502 to be made up and forms combination image.Particularly, piece 1801 allows to readjust where necessary brightness of image.Obtain in the knowledge base of this employed mapping function from Figure 14.The output of piece 1801 of remapping is provided for up-sampler 1802 so that subimage is upsampled to the resolution of next higher subimage.Combiner 1803 makes up subimage and next higher subimage of up-sampling then, or the like.

Figure 19 illustrates according to one or more embodiment and revises so that the above exemplary DSP Hardware configuration that is directed against the functional block of outside DSP1404 discussion is provided.Shown embodiment comprises I/O port one 904, is used for DSP is docked with other circuit, the picture signal of making an uproar is arranged, output through the picture signal of filtering is provided, docks with knowledge base 1414 etc. so that receive.The DSP 1401 of illustrated embodiment comprises DMA engine 1903, frame buffer 1905, HSM 1902 and DSP core 1901.The DSP 1404 of illustrated embodiment comprises DMA engine 1903, frame buffer 1905, HSM 1902 and DSP core 1901.DSP core 1901 comprises ALU (ALU).HSM 1902 is provided at the storer of computing interval by 1901 uses of DSP core.DMA engine 1903 be convenient to HSM 1902 and frame buffer 1905 the back-end data transmission between the slower external memory storage of resident typical case.Although the customized configuration of DSP has been shown, should understand how specific the present invention be not limited to DSP or the processor of other kind is realized above-described multi-resolution adaptive filtering in Figure 19.In fact, such processing can for example be waited and carried out by field programmable gate array (FPGA), special IC (ASIC), general purpose microprocessor.

The advantage of some embodiment is that sef-adapting filter can provide The better resolution through preserving edge.When with the multiresolution decomposition and combination, can more effectively carry out high-performance treatments.Another advantage of some embodiment is that multiresolution is handled can provide more effective mode to come from signal extraction information, from high-grade characteristic to the lower grade details.In fact, some embodiment can realize in mancarried device, because comparatively effectively handle with lower power use and less computing power higher performance are provided.

Major part calculating related in the imaging is to find the solution the differential equation.For example, if extract the advantageous characteristic of striding big image area, can use traditional treating apparatus to set up very big wave filter.Yet various embodiment of the present invention become the low resolution subimage with signal decomposition, and it allows to utilize less nuclear to come recognition feature, and this is because will handle less pixel or point.For example, adopt 30 * 30 or 50 * 50 the nuclear be applicable to different pixels.Yet, through using the notion of the present invention wherein adopt multiresolution to decompose, can use less filter kernel, for example 3 * 3 or 5 * 5.Can provide additional performance to improve through the technology of use simplifying, as precalculated question blank etc. being used for processing block 1502.High-performance filtering with lower-wattage use and lower cost can provide high-quality portable imaging device, like Vltrasonic device.

Although described the present invention and advantage thereof in detail, should understand and in the spirit and scope of the present invention that accompanying claims limits, to carry out various changes, replacement and modification.And, be not the specific embodiment of wanting scope of the present invention is confined to process, machine, manufacturing, material composition, device, method and step described in the instructions.According to disclosure of the present invention; It will be appreciated by those of ordinary skill in the art that to utilize according to the present invention and carry out with the said essentially identical function of corresponding embodiment or realize essentially identical result's current existence or process, machine, manufacturing, material composition, device, method and the step of developing later on.Therefore, intention is that accompanying claims comprises such process, machine, manufacturing, material composition, device, method and step in its scope.

Claims (42)

1. method that is used to handle image, said method comprises:

Said picture breakdown is become a plurality of subimages;

Confirm the one or more characteristics in each subimage in said a plurality of subimage;

Access stored is used for the knowledge base of the filter parameter information of a plurality of scanning processes;

Based on confirming one or more sef-adapting filter parameters with the part filter parameter of the ongoing specific scan procedure correlation of host computer system; And

Sef-adapting filter is applied to each subimage in said a plurality of subimage respectively; Wherein said sef-adapting filter is adaptive to the aspect of the linked character in said one or more characteristic, and wherein said sef-adapting filter is implemented said one or more sef-adapting filter parameter.

2. method according to claim 1, wherein said a plurality of subimages comprise the subimage of different resolution.

3. method according to claim 2, the subimage in wherein said a plurality of subimages respectively are the half the of next high-resolution subimage resolution.

4. method according to claim 2, wherein said decomposition is carried out to lowest resolution from the highest resolution of said different resolution.

5. method according to claim 1, the one or more characteristics in wherein said definite each subimage comprise:

Acceptance is about the characteristic information of another subimage in said a plurality of subimages, so that in one or more characteristics of confirming specific subimage, use.

6. method according to claim 5, the one or more characteristics in wherein said definite each subimage are carried out from the lowest resolution to the highest resolution.

7. method according to claim 1, the one or more characteristics in wherein said definite each subimage comprise:

Edge feature in the identified sub-images.

8. method according to claim 1, the one or more characteristics in wherein said definite each subimage comprise:

Gradient in the identified sub-images.

9. method according to claim 1, the one or more characteristics in wherein said definite each subimage comprise:

The visit knowledge base, said knowledge base is also stored the information about the characteristics of image related with the certain operational modes of host computer system.

10. method according to claim 1, the one or more characteristics in wherein said definite each subimage comprise:

Visit knowledge base, said knowledge base are also stored the information about the characteristics of image related with the particular procedure that uses host computer system to carry out.

11. method according to claim 1, wherein said application self-adapting wave filter comprises:

, the characteristic of subimage uses inter-frame information when being carried out filtering.

12. method according to claim 1, wherein said application self-adapting wave filter comprises:

, the characteristic of subimage uses the frame internal information when being carried out filtering.

13. method according to claim 1, wherein said application self-adapting wave filter is adaptive to the aspect, space of the linked character of subimage.

14. method according to claim 1, wherein said application self-adapting wave filter is adaptive to the time aspect of the linked character of subimage.

15. method according to claim 1, wherein said application self-adapting wave filter is adaptive to the edge of the linked character of subimage.

16. method according to claim 1, wherein said application self-adapting wave filter is adaptive to the slope of the linked character of subimage.

17. method according to claim 1, wherein said application self-adapting wave filter is adaptive to the gradient of the linked character of subimage.

18. method according to claim 1 confirms that wherein one or more said filter parameters further comprise:

Confirm one or more said filter parameters according to said one or more characteristics.

19. method according to claim 1 further comprises:

Confirm the orientation of at least one characteristic in said one or more characteristic.

20. method according to claim 19, wherein said application self-adapting wave filter comprises:

Said orientation according to said at least one characteristic is handled at least one in the said sef-adapting filter.

21. method according to claim 19, wherein said orientation comprises spatial orientation.

22. method according to claim 19, wherein said orientation comprise the time orientation.

23. method according to claim 1 further comprises:

After sef-adapting filter being applied to each said subimage respectively, rebuild image through filtering according to said a plurality of subimages.

24. method according to claim 23, the said image through filtering of the said image of wherein said decomposition, said application self-adapting wave filter and said reconstruction is carried out repeatedly.

25. method according to claim 24, wherein said several times in repeatedly with introduce new process points to said image and be associated.

26. a method that is used to handle image, said method comprises:

Said picture breakdown is become a plurality of subimages;

Confirm the one or more characteristics in each subimage in said a plurality of subimage;

Access stored is used for the knowledge base of the filter parameter information of a plurality of scanning processes;

Confirm one or more sef-adapting filter parameters according to said one or more characteristics, said one or more sef-adapting filter parameters are also based on confirming with the part filter parameter of the ongoing specific scan procedure correlation of host computer system;

Sef-adapting filter is applied to each subimage in said a plurality of subimage respectively; Wherein said sef-adapting filter is adaptive to the aspect of the linked character in said one or more characteristic, and wherein said sef-adapting filter is implemented one or more in the said sef-adapting filter parameter; And

After sef-adapting filter being applied to each said subimage, rebuild image through filtering according to said a plurality of subimages.

27. method according to claim 26, wherein said subimage comprises the subimage of different resolution, and each in the said different resolution is the half the of next high-resolution.

28. method according to claim 26, the said image of wherein said decomposition carries out from the highest resolution to the lowest resolution, and the said image through filtering of wherein said reconstruction carries out from the lowest resolution to the highest resolution.

29. method according to claim 26, one or more characteristics of wherein said definite each said subimage comprise:

Confirm first one or more characteristics in the lowest resolution subimage;

To offer the processing that is used for confirming the one or more characteristics in the high-resolution subimage about the information of said one or more characteristics; And

Use is confirmed second one or more characteristics in the said high-resolution subimage about the said information of said first one or more characteristics.

30. method according to claim 26, wherein said application self-adapting wave filter is adaptive to the aspect, space of the linked character of subimage.

31. method according to claim 26, wherein said application self-adapting wave filter is adaptive to the time aspect of the linked character of subimage.

32. method according to claim 26 further comprises:

Confirm the orientation of at least one characteristic in said one or more characteristic.

33. method according to claim 32 is wherein handled at least one in the said sef-adapting filter according to the said orientation of said at least one characteristic.

34. a system that is used to handle image, said system comprises:

The multi-resolution image resolver can operate being used for receiving view data and therefrom producing a plurality of subimages, and each subimage in the said subimage has different resolution;

Storage is used for the filter parameter database of information of a plurality of scanning processes;

A plurality of processing blocks; Each processing block can be operated one or more characteristics of being used for confirming in the related subimage in said a plurality of subimages and according to said one or more characteristics the filtering to said related subimage is provided; One or more processing blocks in the wherein said processing block receive about be directed against the information of the determined characteristic of another subimage by another processing block in said a plurality of processing blocks; Wherein, said processing block is based on confirming to be used to provide one or more filter parameters of filtering with the part filter parameter of system ongoing specific scan procedure correlation; And

Image reconstructor can operate being used for receiving the output from said a plurality of processing blocks, and therefrom produces the image of combination.

35. system according to claim 34, wherein said a plurality of processing blocks are provided by digital signal processor.

36. system according to claim 35, wherein said multi-resolution image resolver and said image reconstructor are provided by said digital signal processor.

37. system according to claim 34, wherein said filter parameter information is associated with the confirmable special characteristic of said processing block.

38. system according to claim 34, wherein said filtering comprises auto adapted filtering.

39. according to the described system of claim 38, wherein said sef-adapting filter comprises that the aspect, space according to the linked character in the said characteristic adapts to one or more filter parameters.

40. according to the described system of claim 38, wherein said sef-adapting filter comprises that the time aspect according to the linked character in the said characteristic adapts to one or more filter parameters.

41. system according to claim 34, wherein said filtering comprises controlled filtering.

42. according to the described system of claim 41, wherein said controlled filtering comprises according to the orientation of the linked character in the said characteristic handles said wave filter.

Images