CN105677986A - Stamping die good in contour recognition filtering performance - Google Patents

Abstract

The invention discloses a stamping die good in contour recognition filtering performance. The stamping die comprises a common stamping die body and a target recognition device installed on the stamping die body; the recognition device comprises a modeling module, a segmentation module, a combination module and a filtering module. Due to the fact that the target recognition device is additionally arranged on the stamping die body, the stamping efficiency of the stamping die body can be effectively improved, the stamping die body recognizes a target through a target profile, target profile noise can be effectively filtered out in the recognition process, and therefore the category of the target can be correctly recognized.

Description

The diel that a kind of outline identification filtering performance is good

Technical field

The present invention relates to diel field, be specifically related to the diel that a kind of outline identification filtering performance is good.

Background technology

Diel, is in Cold Press Working, material (metal or nonmetal) is processed into a kind of special process equipment of part (or semi-finished product), is called cold punching die. Increasing diel is manufactured, and along with the progress of society and science and technology, the various aspects of the life of people are created tremendous influence by the application of diel. But, target can not be identified by current diel, not only the application of diel is created restriction, and greatly reduces ram efficiency.

Objective contour identification is as the important means of target recognition, owing to practical application being subject to the impact of the factor such as noise, quantization error, objective contour inevitably results from distortion, and for accurate description contour feature, it is very necessary that the filtering of objective contour processes. At present, scholars propose the filtering algorithm of many noisy profiles, but ubiquity amount of calculation is huge, noise reduction is undesirable, be susceptible to excessively filtering causes the problems such as target distortion.

Summary of the invention

For the problems referred to above, the present invention provides the diel that a kind of outline identification filtering performance is good.

The purpose of the present invention realizes by the following technical solutions:

The diel that a kind of outline identification filtering performance is good, including common diel and the Target Identification Unit that is arranged in diel, this diel has significantly high ram efficiency, target can be identified by Target Identification Unit according to objective contour, it is characterized in that, including MBM, segmentation module, merging module and filtration module;

MBM, for setting up the parametrization equation of objective contour: for given objective contour G (t), its arc length parameterized equation is expressed as G (t)=(x (t), y (t)), wherein x (t) and y (t) represents the coordinate of profile point respectively, t represents the parameter of contour curve equation and t ∈ [0,1];

The arc length parameterized equation of noisy profile is expressed as: G_N(t)=G (t)+N₁(t)+N₂(t) G (t), wherein additive noise part N₁(t)=N₁(x₁(t),y₁(t)), multiplicative noise part N₂(t)=N₂(x₂(t),y₂(t));

Segmentation module, the segmentation for profile: objective contour G (t) and noisy profile G_NCurvature corresponding to (t) respectively k (t) and k_N(t); Window function W (n) selecting width x width to be D, to curvature k_NT () carries out neighborhood averaging, obtain average curvature k_1NT (), sorts to the curvature value in window simultaneously, selected intermediate value curvature k_2NT (), by average curvature k_1N(t) and intermediate value curvature k_2NThe absolute value of (t) difference and selected threshold value T₁Compare, determine noisy contour curvature k ' according to comparative result_N(t), it may be assumed that

When | k_1N(t)-k_2N(t)|>T₁Time, k '_N(t)=k_1N(t)

Otherwise, k '_N(t)=k_2N(t);

Owing to profile point that curvature value is bigger generally reflects the marked feature of target, according to k '_NT profile point all in profile are divided into characteristic point or non-characteristic point by (), set variable weight T_K, by judging that objective contour feature is how many, adaptive decision T_K, when | k '_N(t)|<T_K*max|k′_N(t) | time, characteristic function f (t)=0 otherwise, characteristic function f (t)=1.

Merge module: for rejecting the pseudo-random numbers generation owing to noise jamming produces, and the characteristic point and non-characteristic point that cannot form continuum is merged operation, thus obtaining effective characteristic area and non-characteristic area: a selected starting point O, profile starting point extends the point that merging is adjacent to both sides, using this starting point type as this region preset kind, extend each S × μ to both sides₀In time, stops, and wherein S is default minimum length,For the real-time curvature correction factor at O point place,Represent the radius of curvature of O point,Represent the mean radius of curvature of the O point obtained by above-mentioned window function, real-time curvature correction factor μ₀Different for the curvature according to difference, automatically revise development length, the distortion phenomenon after merging can be effectively reduced; Calculating number N+1 and the N-1 of dissimilarity in two side areas respectively, if the number of dissimilarity is less than the minimum number of the type dissimilarity set, then this region is identical with preset kind, otherwise, contrary with preset kind; Again with two halt O₊₁With an O_-1Restart to calculate as starting point, extend S × μ laterally_O+1Or S × μ_O-1In time, stops, wherein μ_O+1And μ_O-1Represent some O respectively₊₁With an O_-1The real-time curvature correction factor at place, O₊₁In two side areas, dissimilarity number is N₊₂, O_-1In two side areas, dissimilarity number is N_-2, according to above-mentioned decision condition, determine each section of types of profiles successively, the part of curtailment S, according to its ratio calculating dissimilarity number with S, counts corresponding characteristic area; Adjacent same type area is merged, obtains continuous print characteristic area and non-characteristic area;

Filtration module: multiplicative noise is owing to being relevant with picture signal, changing with the change of picture signal, adopt Wiener filtering to carry out first-level filtering and remove, now image information also includes the remnants property taken advantage of noise, by F wave filter F (x, y)=q × exp (-(x²+y²)/β²Carrying out two grades to filter, wherein q is by the coefficient of function normalization, it may be assumed that ∫ ∫ q × exp (-(x²+y²)/β²) dxdy=1, β be image template parameter;

After multiplicative noise filters, the arc length parameterized equation of noisy objective contour is expressed as G_N(t) '=G (t)+N₁(t); Assume that additive noise is white Gaussian noise: x_N(t) '=x (t)+g₁(t,σ²), y_N(t) '=y (t)+g₂(t,σ²), wherein x_N(t) ' and y_NT () ' represents after removal multiplicative noise each point coordinates, g on noisy profile respectively₁(t,σ²) and g₂(t,σ²) to be average respectively be zero, variance is σ²White Gaussian noise, for simulating additive noise in noisy objective contour;

Adopt functionNoisy profile is smoothed, called after K wave filter, divide through profile point classification and region, noisy profile G_NT () ' is expressed as the combination of dissimilar contour segmentation:WhereinRepresent the contour segmentation comprising characteristic area,Represent the contour segmentation comprising non-characteristic area, choose the parameter of K wave filter according to contour feature distribution, consider global characteristics and local characteristic factor simultaneously, at characteristic area, in order to retain detailed information, orderAt non-characteristic area, in order to improve the effect suppressing noise, order Wherein σ ' estimates the overall variance obtained, σ for priori₁Priori for selected characteristic area estimates variance, σ₀Priori for selected non-characteristic area estimates variance,For the average real-time curvature correction factor of selected characteristic area,Average real-time curvature correction factor for selected non-characteristic area;In order to reach good smooth effect, choose the half length as K wave filter 85% confidence interval of each type region minimum length S, thus the K wave filter of the length self adaptation different parameters according to two class regions.

The present invention by installing Target Identification Unit additional in diel, the ram efficiency of diel can be effectively improved, objective contour identification target is passed through in diel, can effectively filter out objective contour noise, thus targeted species is made correct identification in identification process.

Accompanying drawing explanation

The invention will be further described to utilize accompanying drawing, but the embodiment in accompanying drawing does not constitute any limitation of the invention, for those of ordinary skill in the art, under the premise not paying creative work, it is also possible to obtain other accompanying drawing according to the following drawings.

Fig. 1 is the structured flowchart of the diel that the outline identification filtering performance of the present invention is good.

Detailed description of the invention

The invention will be further described with the following Examples.

Fig. 1 is the structured flowchart of the present invention, comprising: MBM, segmentation module, merging module, filtration module.

Embodiment 1: the diel that a kind of outline identification filtering performance is good, including common diel and the Target Identification Unit that is arranged in diel, this diel has significantly high ram efficiency, target can be identified by Target Identification Unit according to objective contour, it is characterized in that, including MBM, segmentation module, merging module and filtration module;

MBM, for setting up the parametrization equation of objective contour: for given objective contour G (t), its arc length parameterized equation is expressed as G (t)=(x (t), y (t)), wherein x (t) and y (t) represents the coordinate of profile point respectively, t represents the parameter of contour curve equation and t ∈ [0,1];

The arc length parameterized equation of noisy profile is expressed as: G_N(t)=G (t)+N₁(t)+N₂(t) G (t), wherein additive noise part N₁(t)=N₁(x₁(t),y₁(t)), multiplicative noise part N₂(t)=N₂(x₂(t),y₂(t));

Segmentation module, the segmentation for profile: objective contour G (t) and noisy profile G_NCurvature corresponding to (t) respectively k (t) and k_N(t); Owing to being subject to effect of noise, noisy profile G_NThe curvature value k of (t) upper Partial Feature point_NT () can not accurately represent profile information, in order to obtain curvature accurately, selecting width is that { window function W (n) of 7,9}, to curvature k for D ∈_NT () carries out neighborhood averaging, obtain average curvature k_1NT (), sorts to the curvature value in window simultaneously, selected intermediate value curvature k_2NT (), by average curvature k_1N(t) and intermediate value curvature k_2NThe absolute value of (t) difference and selected threshold value T₁=0.24 compares, and determines noisy contour curvature k ' according to comparative result_N(t), it may be assumed that

As | k1N (t)-k2N (t) | > T₁Time, k '_N(t)=k_1N(t)

Otherwise, k '_N(t)=k_2N(t);

Owing to profile point that curvature value is bigger generally reflects the marked feature of target, according to k '_NT profile point all in profile are divided into characteristic point or non-characteristic point by (), set variable weight T_K, by judging that objective contour feature is how many, adaptive decision T_K, when | k '_N(t)|<T_K*max|k′_N(t) | time, characteristic function f (t)=0 otherwise, characteristic function f (t)=1;

The distribution of characteristic point obtained after classification and non-characteristic point is also discontinuous, it is impossible to it is carried out effective contour smoothing by selecting filter. In order to obtain good contour smoothing effect, it is necessary to the profile point of same type is merged process.

Merge module: for rejecting the pseudo-random numbers generation owing to noise jamming produces, and the characteristic point and non-characteristic point that cannot form continuum is merged operation, thus obtaining effective characteristic area and non-characteristic area: a selected starting point O, profile starting point extends the point that merging is adjacent to both sides, using this starting point type as this region preset kind, extend each S × μ to both sides₀In time, stops, and wherein S is default minimum length, in this embodiment,For the real-time curvature correction factor at O point place,Represent the radius of curvature of O point,Represent the mean radius of curvature of the O point obtained by above-mentioned window function, real-time curvature correction factor μ₀Different for the curvature according to difference, automatically revise development length, the local length needed that curvature is big is less, and the length of the local needs that curvature is little more greatly, so can effectively reduce the distortion phenomenon after merging; Calculating number N+1 and the N-1 of dissimilarity in two side areas respectively, if the number of dissimilarity is less than the minimum number of the type dissimilarity set, then this region is identical with preset kind, otherwise, contrary with preset kind; Again with two halt O₊₁With an O_-1Restart to calculate as starting point, extend S × μ laterally_O+1Or S × μ_O-1In time, stops, wherein μ_O+1And μ_O-1Represent some O respectively₊₁With an O_-1The real-time curvature correction factor at place, O₊₁In two side areas, dissimilarity number is N₊₂, O_-1In two side areas, dissimilarity number is N_-2, according to above-mentioned decision condition, determine each section of types of profiles successively, the part of curtailment S, according to its ratio calculating dissimilarity number with S, counts corresponding characteristic area; Adjacent same type area is merged, obtains continuous print characteristic area and non-characteristic area.

Filtration module: multiplicative noise is owing to being relevant with picture signal, changing with the change of picture signal, adopt Wiener filtering to carry out first-level filtering and remove, now image information also includes the remnants property taken advantage of noise, by F wave filter F (x, y)=q × exp (-(x²+y²)/β²Carrying out two grades to filter, wherein q is by the coefficient of function normalization, it may be assumed that ∫ ∫ q × exp (-(x²+y²)/β²) dxdy=1, β be image template parameter;

After multiplicative noise filters, the arc length parameterized equation of noisy objective contour is expressed as G_N(t) '=G (t)+N₁(t); Assume that additive noise is white Gaussian noise: x_N(t) '=x (t)+g₁(t,σ²), y_N(t) '=y (t)+g₂(t,σ²), wherein x_N(t) ' and y_NT () ' represents after removal multiplicative noise each point coordinates, g on noisy profile respectively₁(t,σ²) and g₂(t,σ²) to be average respectively be zero, variance is σ²White Gaussian noise, for simulating additive noise in noisy objective contour;

Adopt functionNoisy profile is smoothed, called after K wave filter, divide through profile point classification and region, noisy profile G_NT () ' is expressed as the combination of dissimilar contour segmentation:WhereinRepresent the contour segmentation comprising characteristic area,Represent the contour segmentation comprising non-characteristic area, choose the parameter of K wave filter according to contour feature distribution, consider global characteristics and local characteristic factor simultaneously, at characteristic area, in order to retain detailed information, orderAt non-characteristic area, pay close attention to the effect suppressing noise, orderWherein σ ' estimates the overall variance obtained, σ for priori₁Priori for selected characteristic area estimates variance, σ₀Priori for selected non-characteristic area estimates variance,For the average real-time curvature correction factor of selected characteristic area,Average real-time curvature correction factor for selected non-characteristic area;In order to reach good smooth effect, choose the half length as K wave filter 85% confidence interval of each type region minimum length S, thus the K wave filter of the length self adaptation different parameters according to two class regions.

In this embodiment, S=17, threshold value T₁=0.24, window function width D ∈ { 7,9}, to noise intensity I ∈, { the noisy image of 10dB, 20dB} has good smooth effect, and objective contour identification target is passed through in diel, effectively filtering out objective contour noise in identification process, after identification, ram efficiency improves 50%.

Embodiment 2: the diel that a kind of outline identification filtering performance is good, including common diel and the Target Identification Unit that is arranged in diel, this diel has significantly high ram efficiency, target can be identified by Target Identification Unit according to objective contour, it is characterized in that, including MBM, segmentation module, merging module and filtration module;

MBM, for setting up the parametrization equation of objective contour: for given objective contour G (t), its arc length parameterized equation is expressed as G (t)=(x (t), y (t)), wherein x (t) and y (t) represents the coordinate of profile point respectively, t represents the parameter of contour curve equation and t ∈ [0,1];

The arc length parameterized equation of noisy profile is expressed as: G_N(t)=G (t)+N₁(t)+N₂(t) G (t), wherein additive noise part N₁(t)=N₁(x₁(t),y₁(t)), multiplicative noise part N₂(t)=N₂(x₂(t),y₂(t));

Segmentation module, the segmentation for profile: objective contour G (t) and noisy profile G_NCurvature corresponding to (t) respectively k (t) and k_N(t); Owing to being subject to effect of noise, noisy profile G_NThe curvature value k of (t) upper Partial Feature point_NT () can not accurately represent profile information, in order to obtain curvature accurately, selecting width is that { window function W (n) of 10,12}, to curvature k for D ∈_NT () carries out neighborhood averaging, obtain average curvature k_1NT (), sorts to the curvature value in window simultaneously, selected intermediate value curvature k_2NT (), by average curvature k_1N(t) and intermediate value curvature k_2NThe absolute value of (t) difference and selected threshold value T₁=0.24 compares, and determines noisy contour curvature k ' according to comparative result_N(t), it may be assumed that

When | k_1N(t)-k_2N(t)|>T₁Time, k '_N(t)=k_1N(t)

Otherwise, k '_N(t)=k_2N(t);

Owing to profile point that curvature value is bigger generally reflects the marked feature of target, according to k '_NT profile point all in profile are divided into characteristic point or non-characteristic point by (), set variable weight T_K, by judging that objective contour feature is how many, adaptive decision T_K, when | k '_N(t)|<T_K*max|k′_N(t) | time, characteristic function f (t)=0 otherwise, characteristic function f (t)=1;

The distribution of characteristic point obtained after classification and non-characteristic point is also discontinuous, it is impossible to it is carried out effective contour smoothing by selecting filter. In order to obtain good contour smoothing effect, it is necessary to the profile point of same type is merged process.

Merge module: for rejecting the pseudo-random numbers generation owing to noise jamming produces, and the characteristic point and non-characteristic point that cannot form continuum is merged operation, thus obtaining effective characteristic area and non-characteristic area: a selected starting point O, profile starting point extends the point that merging is adjacent to both sides, using this starting point type as this region preset kind, extend each S × μ to both sides₀In time, stops, and wherein S is default minimum length, in this embodiment S=19,For the real-time curvature correction factor at O point place,Represent the radius of curvature of O point,Represent the mean radius of curvature of the O point obtained by above-mentioned window function, real-time curvature correction factor μ₀Different for the curvature according to difference, automatically revise development length, the local length needed that curvature is big is less, and the length of the local needs that curvature is little more greatly, so can effectively reduce the distortion phenomenon after merging;Calculating number N+1 and the N-1 of dissimilarity in two side areas respectively, if the number of dissimilarity is less than the minimum number of the type dissimilarity set, then this region is identical with preset kind, otherwise, contrary with preset kind; Again with two halt O₊₁With an O_-1Restart to calculate as starting point, extend S × μ laterally_O+1Or S × μ_O-1In time, stops, wherein μ_O+1And μ_O-1Represent some O respectively₊₁With an O_-1The real-time curvature correction factor at place, O₊₁In two side areas, dissimilarity number is N₊₂, O_-1In two side areas, dissimilarity number is N_-2, according to above-mentioned decision condition, determine each section of types of profiles successively, the part of curtailment S, according to its ratio calculating dissimilarity number with S, counts corresponding characteristic area; Adjacent same type area is merged, obtains continuous print characteristic area and non-characteristic area.

Filtration module: multiplicative noise is owing to being relevant with picture signal, changing with the change of picture signal, adopt Wiener filtering to carry out first-level filtering and remove, now image information also includes the remnants property taken advantage of noise, by F wave filter F (x, y)=q × exp (-(x²+y²)/β²Carrying out two grades to filter, wherein q is by the coefficient of function normalization, it may be assumed that ∫ ∫ q × exp (-(x²+y²)/β²) dxdy=1, β be image template parameter;

After multiplicative noise filters, the arc length parameterized equation of noisy objective contour is expressed as G_N(t) '=G (t)+N₁(t); Assume that additive noise is white Gaussian noise: x_N(t) '=x (t)+g₁(t,σ²), y_N(t) '=y (t)+g₂(t,σ²), wherein x_N(t) ' and y_NT () ' represents after removal multiplicative noise each point coordinates, g on noisy profile respectively₁(t,σ²) and g₂(t,σ²) to be average respectively be zero, variance is σ²White Gaussian noise, for simulating additive noise in noisy objective contour;

Adopt functionNoisy profile is smoothed, called after K wave filter, divide through profile point classification and region, noisy profile G_NT () ' is expressed as the combination of dissimilar contour segmentation:WhereinRepresent the contour segmentation comprising characteristic area,Represent the contour segmentation comprising non-characteristic area, choose the parameter of K wave filter according to contour feature distribution, consider global characteristics and local characteristic factor simultaneously, at characteristic area, in order to retain detailed information, orderAt non-characteristic area, pay close attention to the effect suppressing noise, orderWherein σ ' estimates the overall variance obtained, σ for priori₁Priori for selected characteristic area estimates variance, σ₀Priori for selected non-characteristic area estimates variance,For the average real-time curvature correction factor of selected characteristic area,Average real-time curvature correction factor for selected non-characteristic area; In order to reach good smooth effect, choose the half length as K wave filter 85% confidence interval of each type region minimum length S, thus the K wave filter of the length self adaptation different parameters according to two class regions.

In this embodiment, S=19, threshold value T₁=0.24, window function width D ∈ { 10,12}, to noise intensity I ∈, { the noisy image of 20dB, 30dB} has good smooth effect, and objective contour identification target is passed through in diel, effectively filtering out objective contour noise in identification process, after identification, ram efficiency improves 50%.

Embodiment 3: the diel that a kind of outline identification filtering performance is good, including common diel and the Target Identification Unit that is arranged in diel, this diel has significantly high ram efficiency, target can be identified by Target Identification Unit according to objective contour, it is characterized in that, including MBM, segmentation module, merging module and filtration module;

MBM, for setting up the parametrization equation of objective contour: for given objective contour G (t), its arc length parameterized equation is expressed as G (t)=(x (t), y (t)), wherein x (t) and y (t) represents the coordinate of profile point respectively, t represents the parameter of contour curve equation and t ∈ [0,1];

The arc length parameterized equation of noisy profile is expressed as: G_N(t)=G (t)+N₁(t)+N₂(t) G (t), wherein additive noise part N₁(t)=N₁(x₁(t),y₁(t)), multiplicative noise part N₂(t)=N₂(x₂(t),y₂(t));

Segmentation module, the segmentation for profile: objective contour G (t) and noisy profile G_NCurvature corresponding to (t) respectively k (t) and k_N(t); Owing to being subject to effect of noise, noisy profile G_NThe curvature value k of (t) upper Partial Feature point_NT () can not accurately represent profile information, in order to obtain curvature accurately, selecting width is that { window function W (n) of 13,14}, to curvature k for D ∈_NT () carries out neighborhood averaging, obtain average curvature k_1NT (), sorts to the curvature value in window simultaneously, selected intermediate value curvature k_2NT (), by average curvature k_1N(t) and intermediate value curvature k_2NThe absolute value of (t) difference and selected threshold value T₁=0.26 compares, and determines noisy contour curvature k ' according to comparative result_N(t), it may be assumed that

When | k_1N(t)-k_2N(t)|>T₁Time, k '_N(t)=k_1N(t)

Otherwise, k '_N(t)=k_2N(t);

Owing to profile point that curvature value is bigger generally reflects the marked feature of target, according to k '_NT profile point all in profile are divided into characteristic point or non-characteristic point by (), set variable weight T_K, by judging that objective contour feature is how many, adaptive decision T_K, when | k '_N(t)|<T_K*max|k′_N(t) | time, characteristic function f (t)=0 otherwise, characteristic function f (t)=1;

The distribution of characteristic point obtained after classification and non-characteristic point is also discontinuous, it is impossible to it is carried out effective contour smoothing by selecting filter. In order to obtain good contour smoothing effect, it is necessary to the profile point of same type is merged process.

Merge module: for rejecting the pseudo-random numbers generation owing to noise jamming produces, and the characteristic point and non-characteristic point that cannot form continuum is merged operation, thus obtaining effective characteristic area and non-characteristic area: a selected starting point O, profile starting point extends the point that merging is adjacent to both sides, using this starting point type as this region preset kind, extend each S × μ to both sides₀In time, stops, and wherein S is default minimum length, in this embodiment S=21,For the real-time curvature correction factor at O point place,Represent the radius of curvature of O point,Represent the mean radius of curvature of the O point obtained by above-mentioned window function, real-time curvature correction factor μ₀Different for the curvature according to difference, automatically revise development length, the local length needed that curvature is big is less, and the length of the local needs that curvature is little more greatly, so can effectively reduce the distortion phenomenon after merging; Calculating number N+1 and the N-1 of dissimilarity in two side areas respectively, if the number of dissimilarity is less than the minimum number of the type dissimilarity set, then this region is identical with preset kind, otherwise, contrary with preset kind; Again with two halt O₊₁With an O_-1Restart to calculate as starting point, extend S × μ laterally_O+1Or S × μ_O-1In time, stops, wherein μ_O+1And μ_O-1Represent some O respectively₊₁With an O_-1The real-time curvature correction factor at place, O₊₁In two side areas, dissimilarity number is N₊₂, O_-1In two side areas, dissimilarity number is N_-2, according to above-mentioned decision condition, determine each section of types of profiles successively, the part of curtailment S, according to its ratio calculating dissimilarity number with S, counts corresponding characteristic area;Adjacent same type area is merged, obtains continuous print characteristic area and non-characteristic area.

Filtration module: multiplicative noise is owing to being relevant with picture signal, changing with the change of picture signal, adopt Wiener filtering to carry out first-level filtering and remove, now image information also includes the remnants property taken advantage of noise, by F wave filter F (x, y)=q × exp (-(x²+y²)/β²Carrying out two grades to filter, wherein q is by the coefficient of function normalization, it may be assumed that ∫ ∫ q × exp (-(x²+y²)/β²) dxdy=1, β be image template parameter;

After multiplicative noise filters, the arc length parameterized equation of noisy objective contour is expressed as G_N(t) '=G (t)+N₁(t); Assume that additive noise is white Gaussian noise: x_N(t) '=x (t)+g₁(t,σ²), y_N(t) '=y (t)+g₂(t,σ²), wherein x_N(t) ' and y_NT () ' represents after removal multiplicative noise each point coordinates, g on noisy profile respectively₁(t,σ²) and g₂(t,σ²) to be average respectively be zero, variance is σ²White Gaussian noise, for simulating additive noise in noisy objective contour;

Adopt functionNoisy profile is smoothed, called after K wave filter, divide through profile point classification and region, noisy profile G_NT () ' is expressed as the combination of dissimilar contour segmentation:WhereinRepresent the contour segmentation comprising characteristic area,Represent the contour segmentation comprising non-characteristic area, choose the parameter of K wave filter according to contour feature distribution, consider global characteristics and local characteristic factor simultaneously, at characteristic area, in order to retain detailed information, orderAt non-characteristic area, pay close attention to the effect suppressing noise, orderWherein σ ' estimates the overall variance obtained, σ for priori₁Priori for selected characteristic area estimates variance, σ₀Priori for selected non-characteristic area estimates variance,For the average real-time curvature correction factor of selected characteristic area,Average real-time curvature correction factor for selected non-characteristic area; In order to reach good smooth effect, choose the half length as K wave filter 85% confidence interval of each type region minimum length S, thus the K wave filter of the length self adaptation different parameters according to two class regions.

In this embodiment, S=21, threshold value T₁=0.26, window function width D ∈ { 13,14}, to noise intensity I ∈ { 30dB, the noisy image of 40dB} has good smooth effect, amount of calculation and detailed information reservation situation all interior in zone of acceptability and acquirement preferably balances, and objective contour identification target is passed through in diel, effectively filtering out objective contour noise in identification process, after identification, ram efficiency improves 50%.

Embodiment 4: the diel that a kind of outline identification filtering performance is good, including common diel and the Target Identification Unit that is arranged in diel, this diel has significantly high ram efficiency, target can be identified by Target Identification Unit according to objective contour, it is characterized in that, including MBM, segmentation module, merging module and filtration module;

MBM, for setting up the parametrization equation of objective contour: for given objective contour G (t), its arc length parameterized equation is expressed as G (t)=(x (t), y (t)), wherein x (t) and y (t) represents the coordinate of profile point respectively, t represents the parameter of contour curve equation and t ∈ [0,1];

The arc length parameterized equation of noisy profile is expressed as: G_N(t)=G (t)+N₁(t)+N₂(t) G (t), wherein additive noise part N₁(t)=N₁(x₁(t),y₁(t)), multiplicative noise part N₂(t)=N₂(x₂(t),y₂(t));

Segmentation module, the segmentation for profile: objective contour G (t) and noisy profile G_NCurvature corresponding to (t) respectively k (t) and k_N(t);Owing to being subject to effect of noise, noisy profile G_NThe curvature value k of (t) upper Partial Feature point_NT () can not accurately represent profile information, in order to obtain curvature accurately, selecting width is that { window function W (n) of 15,17}, to curvature k for D ∈_NT () carries out neighborhood averaging, obtain average curvature k_1NT (), sorts to the curvature value in window simultaneously, selected intermediate value curvature k_2NT (), by average curvature k_1N(t) and intermediate value curvature k_2NThe absolute value of (t) difference and selected threshold value T₁=0.28 compares, and determines noisy contour curvature k ' according to comparative result_N(t), it may be assumed that

When | k_1N(t)-k_2N(t)|>T₁Time, k '_N(t)=k_1N(t)

Otherwise, k '_N(t)=k_2N(t);

Owing to profile point that curvature value is bigger generally reflects the marked feature of target, according to k '_NT profile point all in profile are divided into characteristic point or non-characteristic point by (), set variable weight T_K, by judging that objective contour feature is how many, adaptive decision T_K, when | k '_N(t)|<T_K*max|k′_N(t) | time, characteristic function f (t)=0 otherwise, characteristic function f (t)=1;

The distribution of characteristic point obtained after classification and non-characteristic point is also discontinuous, it is impossible to it is carried out effective contour smoothing by selecting filter. In order to obtain good contour smoothing effect, it is necessary to the profile point of same type is merged process.

Merge module: for rejecting the pseudo-random numbers generation owing to noise jamming produces, and the characteristic point and non-characteristic point that cannot form continuum is merged operation, thus obtaining effective characteristic area and non-characteristic area: a selected starting point O, profile starting point extends the point that merging is adjacent to both sides, using this starting point type as this region preset kind, extend each S × μ to both sides₀In time, stops, and wherein S is default minimum length, in this embodiment S=23,For the real-time curvature correction factor at O point place,Represent the radius of curvature of O point,Represent the mean radius of curvature of the O point obtained by above-mentioned window function, real-time curvature correction factor μ₀Different for the curvature according to difference, automatically revise development length, the local length needed that curvature is big is less, and the length of the local needs that curvature is little more greatly, so can effectively reduce the distortion phenomenon after merging; Calculating number N+1 and the N-1 of dissimilarity in two side areas respectively, if the number of dissimilarity is less than the minimum number of the type dissimilarity set, then this region is identical with preset kind, otherwise, contrary with preset kind; Again with two halt O₊₁With an O_-1Restart to calculate as starting point, extend S × μ laterally_O+1Or S × μ_O-1In time, stops, wherein μ_O+1And μ_O-1Represent some O respectively₊₁With an O_-1The real-time curvature correction factor at place, O₊₁In two side areas, dissimilarity number is N₊₂, O_-1In two side areas, dissimilarity number is N_-2, according to above-mentioned decision condition, determine each section of types of profiles successively, the part of curtailment S, according to its ratio calculating dissimilarity number with S, counts corresponding characteristic area; Adjacent same type area is merged, obtains continuous print characteristic area and non-characteristic area.

Filtration module: multiplicative noise is owing to being relevant with picture signal, changing with the change of picture signal, adopt Wiener filtering to carry out first-level filtering and remove, now image information also includes the remnants property taken advantage of noise, by F wave filter F (x, y)=q × exp (-(x²+y²)/β²Carrying out two grades to filter, wherein q is by the coefficient of function normalization, it may be assumed that ∫ ∫ q × exp (-(x²+y²)/β²) dxdy=1, β be image template parameter;

After multiplicative noise filters, the arc length parameterized equation of noisy objective contour is expressed as G_N(t) '=G (t)+N₁(t);Assume that additive noise is white Gaussian noise: x_N(t) '=x (t)+g₁(t,σ²), y_N(t) '=y (t)+g₂(t,σ²), wherein x_N(t) ' and y_NT () ' represents after removal multiplicative noise each point coordinates, g on noisy profile respectively₁(t,σ²) and g₂(t,σ²) to be average respectively be zero, variance is σ²White Gaussian noise, for simulating additive noise in noisy objective contour;

Adopt functionNoisy profile is smoothed, called after K wave filter, divide through profile point classification and region, noisy profile G_NT () ' is expressed as the combination of dissimilar contour segmentation:WhereinRepresent the contour segmentation comprising characteristic area,Represent the contour segmentation comprising non-characteristic area, choose the parameter of K wave filter according to contour feature distribution, consider global characteristics and local characteristic factor simultaneously, at characteristic area, in order to retain detailed information, orderAt non-characteristic area, pay close attention to the effect suppressing noise, orderWherein σ ' estimates the overall variance obtained, σ for priori₁Priori for selected characteristic area estimates variance, σ₀Priori for selected non-characteristic area estimates variance,For the average real-time curvature correction factor of selected characteristic area,Average real-time curvature correction factor for selected non-characteristic area; In order to reach good smooth effect, choose the half length as K wave filter 85% confidence interval of each type region minimum length S, thus the K wave filter of the length self adaptation different parameters according to two class regions.

In this embodiment, S=23, threshold value T₁=0.28, window function width D ∈ { 15,17}, to noise intensity I ∈, { 40dB, although the noisy image of 50dB} adds part amount of calculation, but has excellent smooth effect to the image in this interval, and detailed information reservation situation is better, objective contour identification target is passed through in diel, can effectively filter out objective contour noise in identification process, and after identification, ram efficiency improves 50%.

Embodiment 5: the diel that a kind of outline identification filtering performance is good, including common diel and the Target Identification Unit that is arranged in diel, this diel has significantly high ram efficiency, target can be identified by Target Identification Unit according to objective contour, it is characterized in that, including MBM, segmentation module, merging module and filtration module;

MBM, for setting up the parametrization equation of objective contour: for given objective contour G (t), its arc length parameterized equation is expressed as G (t)=(x (t), y (t)), wherein x (t) and y (t) represents the coordinate of profile point respectively, t represents the parameter of contour curve equation and t ∈ [0,1];

The arc length parameterized equation of noisy profile is expressed as: G_N(t)=G (t)+N₁(t)+N₂(t) G (t), wherein additive noise part N₁(t)=N₁(x₁(t),y₁(t)), multiplicative noise part N₂(t)=N₂(x₂(t),y₂(t));

Segmentation module, the segmentation for profile: objective contour G (t) and noisy profile G_NCurvature corresponding to (t) respectively k (t) and k_N(t); Owing to being subject to effect of noise, noisy profile G_NThe curvature value k of (t) upper Partial Feature point_NT () can not accurately represent profile information, in order to obtain curvature accurately, selecting width is that { window function W (n) of 17,19}, to curvature k for D ∈_NT () carries out neighborhood averaging, obtain average curvature k_1NT (), sorts to the curvature value in window simultaneously, selected intermediate value curvature k_2NT (), by average curvature k_1N(t) and intermediate value curvature k_2NThe absolute value of (t) difference and selected threshold value T₁=0.26 compares, and determines noisy contour curvature k ' according to comparative result_N(t), it may be assumed that

When | k_1N(t)-k_2N(t)|>T₁Time, k '_N(t)=k_1N(t)

Otherwise, k '_N(t)=k_2N(t);

Owing to profile point that curvature value is bigger generally reflects the marked feature of target, according to k '_NT profile point all in profile are divided into characteristic point or non-characteristic point by (), set variable weight T_K, by judging that objective contour feature is how many, adaptive decision T_K, when | k '_N(t)|<T_K*max|k′_N(t) | time, characteristic function f (t)=0 otherwise, characteristic function f (t)=1;

The distribution of characteristic point obtained after classification and non-characteristic point is also discontinuous, it is impossible to it is carried out effective contour smoothing by selecting filter. In order to obtain good contour smoothing effect, it is necessary to the profile point of same type is merged process.

Merge module: for rejecting the pseudo-random numbers generation owing to noise jamming produces, and the characteristic point and non-characteristic point that cannot form continuum is merged operation, thus obtaining effective characteristic area and non-characteristic area: a selected starting point O, profile starting point extends the point that merging is adjacent to both sides, using this starting point type as this region preset kind, extend each S × μ to both sides₀In time, stops, and wherein S is default minimum length, in this embodiment S=25,For the real-time curvature correction factor at O point place,Represent the radius of curvature of O point,Represent the mean radius of curvature of the O point obtained by above-mentioned window function, real-time curvature correction factor μ₀Different for the curvature according to difference, automatically revise development length, the local length needed that curvature is big is less, and the length of the local needs that curvature is little more greatly, so can effectively reduce the distortion phenomenon after merging; Calculating number N+1 and the N-1 of dissimilarity in two side areas respectively, if the number of dissimilarity is less than the minimum number of the type dissimilarity set, then this region is identical with preset kind, otherwise, contrary with preset kind; Again with two halt O₊₁With an O_-1Restart to calculate as starting point, extend S × μ laterally_O+1Or S × μ_O-1In time, stops, wherein μ_O+1And μ_O-1Represent some O respectively₊₁With an O_-1The real-time curvature correction factor at place, O₊₁In two side areas, dissimilarity number is N₊₂, O_-1In two side areas, dissimilarity number is N_-2, according to above-mentioned decision condition, determine each section of types of profiles successively, the part of curtailment S, according to its ratio calculating dissimilarity number with S, counts corresponding characteristic area; Adjacent same type area is merged, obtains continuous print characteristic area and non-characteristic area.

Filtration module: multiplicative noise is owing to being relevant with picture signal, changing with the change of picture signal, adopt Wiener filtering to carry out first-level filtering and remove, now image information also includes the remnants property taken advantage of noise, by F wave filter F (x, y)=q × exp (-(x²+y²)/β²Carrying out two grades to filter, wherein q is by the coefficient of function normalization, it may be assumed that ∫ ∫ q × exp (-(x²+y²)/β²) dxdy=1, β be image template parameter;

After multiplicative noise filters, the arc length parameterized equation of noisy objective contour is expressed as G_N(t) '=G (t)+N₁(t); Assume that additive noise is white Gaussian noise: x_N(t) '=x (t)+g₁(t,σ²), y_N(t) '=y (t)+g₂(t,σ²), wherein x_N(t) ' and y_NT () ' represents after removal multiplicative noise each point coordinates, g on noisy profile respectively₁(t,σ²) and g₂(t,σ²) to be average respectively be zero, variance is σ²White Gaussian noise, for simulating additive noise in noisy objective contour;

Adopt functionNoisy profile is smoothed, called after K wave filter, divide through profile point classification and region, noisy profile G_NT () ' is expressed as the combination of dissimilar contour segmentation:WhereinRepresent the contour segmentation comprising characteristic area,Represent the contour segmentation comprising non-characteristic area, choose the parameter of K wave filter according to contour feature distribution, consider global characteristics and local characteristic factor simultaneously, at characteristic area, in order to retain detailed information, orderAt non-characteristic area, pay close attention to the effect suppressing noise, orderWherein σ ' estimates the overall variance obtained, σ for priori₁Priori for selected characteristic area estimates variance, σ₀Priori for selected non-characteristic area estimates variance,For the average real-time curvature correction factor of selected characteristic area,Average real-time curvature correction factor for selected non-characteristic area;In order to reach good smooth effect, choose the half length as K wave filter 85% confidence interval of each type region minimum length S, thus the K wave filter of the length self adaptation different parameters according to two class regions.

In this embodiment, S=25, threshold value T₁=0.26, window function width D ∈ { 17,19}, to noise intensity I ∈ { 50dB, the noisy image of 60dB} has preferably smooth effect, and detailed information retains situation better, and objective contour identification target is passed through in diel, effectively filtering out objective contour noise in identification process, after identification, ram efficiency improves 50%.

Finally should be noted that; above example is only in order to illustrate technical scheme; but not limiting the scope of the invention; although having made to explain to the present invention with reference to preferred embodiment; it will be understood by those within the art that; technical scheme can be modified or equivalent replacement, without deviating from the spirit and scope of technical solution of the present invention.

Data simulation

This diel have the beneficial effect that the multiformity for noise type and the unicity of current denoising method, adopt a kind of novel repeatedly filter, and propose new contour segmentation, merging means and filter function; Amount of calculation is relative and uncomplicated, considers the factor of global characteristics and local feature and smooth effective except making an uproar simultaneously; Consider profile diversity between dissimilar region, between suppression noise and reservation details, obtain good balance; Curvature according to difference is different, and development length is automatic adaptive change correspondingly, effectively reduces the distortion phenomenon after merging.

By emulating, this device is adopted to compare under noise intensity N, the discrimination such as following table to target:

Noise intensity	0.03	0.06	0.09	0.12
					This device embodiment 1	94%	93%	92%	90%
This device embodiment 2	93%	94%	92%	90%
					This device embodiment 3	92%	91%	93%	90%
This device embodiment 4	91%	90%	91%	92%
					This device embodiment 5	90%	90%	92%	93%

Claims (2)

1. the diel that an outline identification filtering performance is good, including common diel and the Target Identification Unit that is arranged in diel, this diel has significantly high ram efficiency, target can be identified by Target Identification Unit according to objective contour, it is characterized in that, including MBM, segmentation module, merging module and filtration module; Wherein,

MBM, for setting up the parametrization equation of objective contour: for given objective contour G (t), its arc length parameterized equation is expressed as G (t)=(x (t), y (t)), wherein x (t) and y (t) represents the coordinate of profile point respectively, t represents the parameter of contour curve equation and t ∈ [0,1];

The arc length parameterized equation of noisy profile is expressed as: G_N(t)=G (t)+N₁(t)+N₂(t) G (t), wherein additive noise part N₁(t)=N₁(x₁(t),y₁(t)), multiplicative noise part N₂(t)=N₂(x₂(t),y₂(t));

Segmentation module, the segmentation for profile: objective contour G (t) and noisy profile G_NCurvature corresponding to (t) respectively k (t) and k_N(t); Window function W (n) selecting width to be D, D ∈ 7,9}, to curvature k_NT () carries out neighborhood averaging, obtain average curvature k_1NT (), sorts to the curvature value in window simultaneously, selected intermediate value curvature k_2NT (), by average curvature k_1N(t) and intermediate value curvature k_2NThe absolute value of (t) difference and selected threshold value T₁Compare, determine noisy contour curvature k ' according to comparative result_N(t), T₁=0.2, it may be assumed that

When | k_1N(t)-k_2N(t)|>T₁Time, k '_N(t)=k_1N(t)

Otherwise, k '_N(t)=k_2N(t);

Owing to profile point that curvature value is bigger generally reflects the marked feature of target, according to k '_NT profile point all in profile are divided into characteristic point or non-characteristic point by (), set variable weight T_K, by judging that objective contour feature is how many, adaptive decision T_K, when | k '_N(t)|<T_K*max|k′_N(t) | time, characteristic function f (t)=0

Otherwise, characteristic function f (t)=1.

2. diel according to claim 1, it is further characterized in that, merge module: for rejecting the pseudo-random numbers generation owing to noise jamming produces, and the characteristic point and non-characteristic point that cannot form continuum is merged operation, thus obtaining effective characteristic area and non-characteristic area: a selected starting point O, profile starting point extends the point that merging is adjacent to both sides, using this starting point type as this region preset kind, extends each S × μ to both sides₀In time, stops, and wherein S is default minimum length, if S=15,For the real-time curvature correction factor at O point place,Represent the radius of curvature of O point,Represent the mean radius of curvature of the O point obtained by above-mentioned window function, real-time curvature correction factor μ₀Different for the curvature according to difference, automatically revise development length, the distortion phenomenon after merging can be effectively reduced; Calculating number N+1 and the N-1 of dissimilarity in two side areas respectively, if the number of dissimilarity is less than the minimum number of the type dissimilarity set, then this region is identical with preset kind, otherwise, contrary with preset kind; Again with two halt O₊₁With an O_-1Restart to calculate as starting point, extend S × μ laterally₀₊₁Or S × μ_0-1In time, stops, wherein μ₀₊₁And μ_0-1Represent some O respectively₊₁With an O_-1The real-time curvature correction factor at place, O₊₁In two side areas, dissimilarity number is N₊₂, O_-1In two side areas, dissimilarity number is N_-2, according to above-mentioned decision condition, determine each section of types of profiles successively, the part of curtailment S, according to its ratio calculating dissimilarity number with S, counts corresponding characteristic area; Adjacent same type area is merged, obtains continuous print characteristic area and non-characteristic area;

Filtration module: multiplicative noise is owing to being relevant with picture signal, changing with the change of picture signal, adopt Wiener filtering to carry out first-level filtering and remove, now image information also includes the property the taken advantage of noise of remnants, by F wave filter F (x, y)=q × exp (-(x²+y²)/β²Carrying out two grades to filter, wherein q is by the coefficient of function normalization, it may be assumed that ∫ ∫ q × exp (-(x²+y²)/β²) dxdy=1, β be image template parameter;

After multiplicative noise filters, the arc length parameterized equation of noisy objective contour is expressed as G_N(t) '=G (t)+N₁(t); Assume that additive noise is white Gaussian noise: x_N(t) '=x (t)+g₁(t,σ²), y_N(t) '=y (t)+g₂(t,σ²), wherein x_N(t) ' and y_NT () ' represents after removal multiplicative noise each point coordinates, g on noisy profile respectively₁(t,σ²) and g₂(t,σ²) to be average respectively be zero, variance is σ²White Gaussian noise, for simulating additive noise in noisy objective contour;

Adopt functionNoisy profile is smoothed, called after K wave filter, divide through profile point classification and region, noisy profile G_NT () ' is expressed as the combination of dissimilar contour segmentation:WhereinRepresent the contour segmentation comprising characteristic area,Represent the contour segmentation comprising non-characteristic area, choose the parameter of K wave filter according to contour feature distribution, consider global characteristics and local characteristic factor simultaneously, at characteristic area, in order to retain detailed information, orderAt non-characteristic area, in order to improve the effect suppressing noise, order Wherein σ ' estimates the overall variance obtained, σ for priori₁Priori for selected characteristic area estimates variance, σ₀Priori for selected non-characteristic area estimates variance,For the average real-time curvature correction factor of selected characteristic area,Average real-time curvature correction factor for selected non-characteristic area;In order to reach good smooth effect, choose the half length as K wave filter 85% confidence interval of each type region minimum length S, thus the K wave filter of the length self adaptation different parameters according to two class regions.