CN102880700A - Three dimensional point model data denoising method - Google Patents

Abstract

The invention discloses a three dimensional point model data denoising method based on fuzzy C-means and bilateral filtering. The method comprises the following steps: setting a dataset P with N points, defining the threshold of the proximal point number in the bounding sphere of data point Pj, determining whether the number of proximal points in the bounding sphere of data point Pj is less than the given threshold, and denoising small scale noisy data in the new dataset with the point cloud bilateral filtering method, first, calculating the proximal points at the periphery of each data point in the new dataset, then, calculating the parameter of the bilateral filtering function for each proximal point and calculating the bilateral filtering weight factor, at last, updating the position of each data point. When every data point in the new dataset is updated, a data point set after denoising is obtained. The method has the advantages that noises are divided into large scale noises and small scale noises on the premise of achieving the goal of denoising the three dimensional point model data, different methods are respectively adopted to process different noises, the data denoising speed is higher, and the original characteristics of the denoised data are better preserved.

Description

A kind of three-dimensional point model data denoising method

Technical field

The present invention relates to a kind of three-dimensional point model data denoising method based on fuzzy C-mean algorithm and bilateral filtering, the method is applicable to medical science, Aero-Space, machine-building, archaeology etc. and removes noise by the three-dimensional point model data that 3-D scanning equipment obtains, and belongs to the computer graphics techniques field.

Background technology

Along with the promotion of medical science auxiliary diagnosis, space flight simulation, industrial design, construction work, archaeology, reverse-engineering, video display amusement etc. application demand, rely on and more and more receive publicity around the research of three-dimensional digital data.For having the abundant object of exquisite surface details and geological information, utilizing laser scanner to obtain its three-dimensional digital model is current the most reasonable and effective method.Along with spatial digitizer hardware device and process software ground are constantly updated, three-dimensional data take spatial digitizer as hardware foundation is obtained system and not only can be obtained geological information but also can obtain the superficial makings colouring information, the on the other hand continuous decline of hardware price is for popularizing of 3-D scanning technology brought opportunity.

The output of spatial digitizer is a large amount of discrete point cloud datas of model normally, and these sampled points have comprised the geometric jacquard patterning unit surface characteristic information of object, can reconstruct easily the digital model of object, and can edit, the operation such as drafting.Even yet high-quality scanning device, the sampling point information of output all can be subject to the pollution of noise, so must carry out noise reduction process before point model is done further operation.The purpose of denoising is to obtain Discrete Surfaces more in the smoothness of high order, prevents from as far as possible that model from producing to shrink and cross fairing.

At present, representative in the point model denoise algorithm is that Jones etc. has proposed a bilateral filtering algorithm that non-iterative feature keeps, although the method can keep by the size of control vertex adjacent area the feature of grid model, and can be applied to cloud data and non-manifold curved surface, but the drawback of this algorithm also is the selection of the adjacent area on summit, excessive when neighborhood, then need intensive to seek neighborhood during processing; When neighborhood too small, some slightly large noises of fairing effectively then, even strengthen these noises.

For the denoising of the point model data of magnanimity, how under the prerequisite that guarantees the master pattern surface characteristics, improving denoising speed is that the lot of domestic and international scholar is in the problem of making great efforts research.

Summary of the invention

Can not take into account the deficiency of holding point model feature information and denoising speed in order to overcome prior art, the present invention proposes a kind of three-dimensional point model data denoising method based on fuzzy C-mean algorithm and bilateral filtering, at first utilize fuzzy C-means clustering to distinguish large scale noise data and other data, then use the bilateral filtering denoising method for remaining small scale noise and non-noise spot, reach final denoising effect.

The technical scheme that the present invention adopts for achieving the above object is may further comprise the steps:

(1) is provided with N data set P that puts, the definition data point p _jThe encirclement ball in the threshold value of contained neighbor point number,

(2) judge data point p _jWhether surround the interior neighbor point number of ball less than given threshold value, 1 ≤ j≤N; If data point p _jSurround the interior neighbor point number of ball less than given threshold value, with its removal; If data point p _jSurround the interior neighbor point number of ball more than or equal to given threshold value, to data point p _jThe point set D that the place surrounds in the ball uses FCM Algorithms to carry out cluster, obtains the fuzzy C-means clustering center collection C of data set P, and the data point cluster centre collection C that uses cluster centre to form replaces original data set P as new data point set; In set of data points P have a few all after treatment, obtain removing the new data set Q that comprises small scale noise and non-noise spot behind the large scale noise;

(3) utilize some cloud bilateral filtering method that new data set Q Small and Medium Sized noise data is carried out denoising, first to each data point among the new data set Q q _i Obtain its neighbor point on every side, again each neighbor point is obtained the parameter of bilateral filtering function, calculate the bilateral filtering weight factor, the position of final updating data point;

(4) each data point in new data set Q q _i After all upgrading, obtain removing the set of data points R behind the noise.

The invention has the beneficial effects as follows: under the prerequisite that realizes three-dimensional point model data removal noise, noise is divided into large scale noise and small scale noise, adopt respectively diverse ways to process, therefore, data de-noising speed is faster, the primitive character of the data of denoising keeps better, has taken into account the speed of denoising operation and the maintenance of point model primitive character.

Description of drawings

Fig. 1 is the inventive method structural representation;

Fig. 2 is that fuzzy C-mean algorithm is removed large scale noise schematic diagram;

Fig. 3 is that the present invention removes the particular flow sheet that the large scale noise obtains the data of small scale noise and non-noise spot;

Fig. 4 is that the present invention uses the particular flow sheet that the bilateral filtering method of protecting feature is removed the small scale noise;

Fig. 5 uses the inventive method denoising effect figure.

Embodiment

Fig. 1 has shown structural representation of the present invention, the noise of three-dimensional point model data is divided into the large scale noise in the present invention and the small scale noise should be processed by diverse ways respectively: remove the large scale noise with the fuzzy C means clustering method, noise spot and data point are carried out cluster, then with two-sided filter iteration point cloud method the small scale noise is carried out fairing, not only removed noise but also substantially kept the sharp features of model.

Referring to Fig. 1 and Fig. 2, be provided with the data set P of N point, P={ p ₁, p ₂..., p _N, N is the data point number.The definition data point p _jThe encirclement ball in the threshold value of contained neighbor point number, judge data point p _jWhether surround the interior neighbor point number of ball less than given threshold value.

Utilize FCM Algorithms to remove large scale noise among the data set P, obtain comprising the data of small scale noise and non-noise spot.As shown in Figure 3, concrete grammar comprises following content:

If data point p _jSurround the interior neighbor point number of ball less than given threshold value, it is regarded the large scale noise spot, with its removal.For example in Fig. 2, define data point p _jThe threshold value of surrounding contained neighbor point number in the ball is 2, and then Fig. 2 (a) mid point 1 and point 2 all are noise spots, because its neighbor point number is 1, less than given threshold value, therefore judges that this point is noise spot, with its removal.Point 3 is noise spots, but its neighbor point number equals 2, namely is not less than given threshold value, so this point is judged as data point, this point is shifted to the cluster centre point that surrounds in the ball with clustering method.If data point p _jSurround the interior neighbor point number of ball more than or equal to given threshold value, if data point p _jThe place surrounds the some set D={ in the ball d ₁, d ₂..., d _n, use FCM Algorithms to carry out cluster to D, adopt formula

Try to achieve the fuzzy C-means clustering center collection C of data set P, C={ c ₁, c ₂..., c _C, parameter wherein c _kCluster centre, k=1 ..., C, μ _JkIt is data point d _jRelative cluster centre c _kDegree of membership, the expression dBelong to kThe probability of class, mFuzzy coefficient, μ, cIt is parameter to be asked.The data point cluster centre set C that uses cluster centre to form replaces original data set P as new data point set.Such as Fig. 2 (b) mid point 1, point 2 with to put 3 all be noise, but its neighbor point all is not less than 2, namely be not less than given threshold value, therefore this point is judged as data point, this point is shifted to the cluster centre point that surrounds in the ball with clustering method, can find out the feature that has kept model surface, level and smooth phenomenon not occur.

When among the set of data points P have a few all through processing after the algorithm end, obtain removing the new data set Q that comprises small scale noise and non-noise spot behind the large scale noise.

The present invention can remove part large scale noise as seen from Figure 2, and sub-fraction large scale noise can move to data point, and the small scale noise can not be removed, and just these noise spots is carried out part smoothly.

Utilize some cloud bilateral filtering method that remaining small scale noise data among the new data set Q is carried out denoising.As shown in Figure 4, concrete grammar comprises following content:

For each data point among the new data set Q of previous step gained q _i ,Obtain around it m _q Individual Neighbor Points k _Ij , j=1 ..., m _q Each neighbor point is obtained the parameter of bilateral filtering function

, xFor the point q _i To neighbor point k _Ij Distance; Calculated characteristics keeps the parameter of weighting function

, yFor the point q _i Distance vector with neighbor point q _i - k _Ij And the point q _i The inner product of normal direction, the parameter of bilateral filtering function W _c ( x) and W _s ( y) all be the standard gaussian filtering, computing formula is

With

Wherein, parameter σ _CIt is data point q _i The distance of ordering to neighbours is to the factor of influence of this point, parameter σ _sTo be data point q _i Arrive the distance vector of neighbor point at this normal vector n _i On projection to data point q _i Factor of influence.Will W _c ( x) and W _s ( y) substitution

, calculate the bilateral filtering weight factor α, wherein N ( q _i ) be data point q _i Neighbours' point, W _c ( x) and W _s ( y) be the parameter of the bilateral filtering function in the step 3.2.Calculate through filtered new data point q _i := q _i + α n _i , n _i A little q _i Normal vector, αIt is the bilateral filtering weight factor that calculates in the step 3.3.

Each data point in data set Q q _i All pass through the formula in the step 3.4 q _i := q _i + α n _i After recomputating, method finishes, and obtains removing the set of data points R behind the noise.

Fig. 5 uses the inventive method denoising effect figure to Bunny-IH point model (10041 points).Wherein, Fig. 5 (a) is noise model, Fig. 5 (b) is with the effect after the denoising of fuzzy C-mean algorithm method, can find out that a part of large scale noise is removed, the noise of all the other a part of large scale noises and small scale is by the part fairing. and Fig. 5 (c) is the effect after the fairing of Bunny-IH model process bilateral filtering method, and Fig. 5 (d) shows that the present invention has kept the surface characteristics details of original model preferably.Can find out, when model comprised large and small yardstick noise simultaneously, the present invention was well more a lot of than direct employing bilateral filtering method, in the situation that keep model surface feature noise to remove, and can not produce fairing at the sharp features place yet.

Above-described example just is used for explanation the present invention, and is not construed as limiting the invention.Those skilled in the art can make various other various modifications and changes that do not break away from essence of the present invention according to these technology enlightenments disclosed by the invention, and these modifications and changes are still in protection scope of the present invention.

Claims (1)

1. three-dimensional point model data denoising method is characterized in that may further comprise the steps:

(1) is provided with N data set P that puts, the definition data point p _jThe encirclement ball in the threshold value of contained neighbor point number,

(2) judge data point p _jWhether surround the interior neighbor point number of ball less than given threshold value, 1 ≤ j≤N; If data point p _jSurround the interior neighbor point number of ball less than given threshold value, with its removal; If data point p _jSurround the interior neighbor point number of ball more than or equal to given threshold value, to data point p _jThe point set D that the place surrounds in the ball uses FCM Algorithms to carry out cluster, obtains the fuzzy C-means clustering center collection C of data set P, and the data point cluster centre collection C that uses cluster centre to form replaces original data set P as new data point set; In set of data points P have a few all after treatment, obtain removing the new data set Q that comprises small scale noise and non-noise spot behind the large scale noise;

(3) utilize some cloud bilateral filtering method that new data set Q Small and Medium Sized noise data is carried out denoising, first to each data point among the new data set Q q _i Obtain its neighbor point on every side, again each neighbor point is obtained the parameter of bilateral filtering function, calculate the bilateral filtering weight factor, the position of final updating data point;

(4) each data point in new data set Q q _i After all upgrading, obtain removing the set of data points R behind the noise.

Images