CN102509344B - Real-time drawing method based on nonuniform sampling three-dimensional object reflection - Google Patents

Abstract

The invention provides a real-time drawing method based on nonuniform sampling three-dimensional object reflection. The method comprises the following steps of acquiring a sampling templete of a reflected object image, generating a reflected object nonuniform sampling deformation pattern according to the sampling templete, and finally, drawing the reflection effect in the scene by the deformation pattern reconstructed nonuniform sampling image. The algorithm for automatically generating the sampling templete can draw the reflection phenomenon in real time. The reflection in the scene is more realistic when being drawn under the condition of low overall resolution (small reflection calculated amount), thereby achieving the goal of enhancing sense of reality.

Description

A kind of real-time drawing method based on the reflection of nonuniform sampling three dimensional object

Technical field

The present invention relates to a kind of real-time drawing method based on the reflection of nonuniform sampling three dimensional object.

Background technology

Illumination calculation is one of important technology of Realistic Rendering in computer graphics always, and reflex is the core content in illumination model especially.Reflect so importantly, because it not only shows aesthetic values in drawing result, and can disclose the geological information of object, as position relationship between the shape of object, object etc.People have carried out a large amount of research to the real-time rendering reflecting in scene, but still do not find out one gratifying method of while in effect and efficiency.Therefore, the real-time paint reflection of three-dimensional body is the focus that photo realism graphic is drawn always, is also difficult point simultaneously.

Paint reflection in scene means the problem that need to consider these aspects to each pixel: whether this light intersects with reflecting surface; If intersect, calculate intersection point and illumination information corresponding to this some place; For complex scene continue follow the tracks of this reflection ray, judge whether with scene in other reflecting body intersect.The algorithm of paint reflection mainly concentrates on the improvement of Environment and ray trace and the algorithm of mixing at present.These methods are normally raised the efficiency or one of effect, do not realize the requirement that the two strengthens simultaneously.

In recent years, experts and scholars have also obtained certain achievement in research at the aspect such as nonuniform sampling and paint reflection:

1) nonuniform sampling

Williams etc. have proposed mipmapping technology, and by making for accelerating drafting of image look-up table, but effect is not drawn in improvement.The sampling of the mipmap image of every one deck or uniformly, has more high-risely stored more pixels, even if the resolution of the regional area of this layer not have high requirement like this.Spatial level partitioning technology has improved rendering efficiency as the Kd-tree of Louis Bentley, but sampling rate lacks continuity.The image that the Kd-tree of the optimizations such as Silpa-Anan represents improves former method, can be used in compression of images and pattern match, but is not suitable for the drafting based on image.In the drafting based on image, can incorporate Kd-tree technology, now in Kd-tree each leaf be one can independent access single image, and these leaves can be obtained by mipmap technology.Nonuniform sampling image couples together two kinds of technology well continuously---and the image of the continuous nonuniform sampling of a width can carry out mipmap processing, and same, the every leaf of Kd-tree can be different images after treatment.

Nonuniform sampling can also be realized by breaking through visual range restriction in traditional images.Gascuel points out to become possibility because the development of graphic hardware makes the sampling pattern changing, and has avoided causing due to storage restriction in early days the problem of undersampling.Meanwhile, also have some researchs to attempt to break away from the constraint of traditional images single view, aspect camera model improvement, obtaining some achievements, as the camera model of blocking of the people's such as Yu general linear camera and Mei etc.Nonuniform sampling has been optimized these methods continuously, has increased the advantage of flexible sampling for improving the image of visual range and the image of non-pinhole camera generation.

Importance sampling method based on GPU is the method for a kind for the treatment of surface reflecting effect of proposing in 2007 such as Colbert, for the higher sampling density of the large region allocation of illumination contribution amount, otherwise distribute lower sampling density, adopt mipmap to filter the distortion of eliminating texture, but the method can only show local lighting effect simultaneously; On this basis, Xuan Yu etc. utilized Geometry Image structure LOD (Level of Detail) in 2008, use importance sampling method, in BVH (Bounding Volume Hierarchy) ergodic process, be that every sampling light distributes solid angle, carry out adaptive LOD hierarchy selection according to solid angle, improve rendering speed, but be only applicable to process the model that Geometry Image represents.

2) reflecting effect is drawn

Comparatively common reflecting material pledge object plotting method mainly contains three classes at present: Environment method, image space method and ray tracing method.

When environment for use pinup picture is played up reflecting body, only according to the direction of reflected ray, index Environment, the color value that obtains reflexing to.The method for object from reflecting body the scene away from, can obtain good effect; But for object nearby, can produce larger deviation.Meanwhile, Environment cannot show dynamic scene in addition.

Modeling based on image space and drafting (Image-based modeling and rendering, IBMR) can be used for processing reflecting effect equally.Light field Rendering can be used to calculate reflection, and its content is not need under the depth information of image or the condition of correlativity, by one group of scene photo taking in advance, sets up the light field data storehouse of this scene, realizes the roaming to whole scene.Even but the method is to small-scale scene, also needs to store a large amount of light information.Lischinski etc. have proposed a kind of scene method for expressing based on depth of seam division figure (layered depth images, LDIs).Scene solid is by 3 orthogonal LDIs institute records.Scene information under viewpoint is stored in the light field of low resolution LDIs, is used to provide smooth reflective information.Mirror-reflection is played up by the LDIs of ray tracing solid.Reduce memory space although do like this, but reduced interactivity.Hakura etc. have described a kind of parameterized Environment, are made up of the reference reflected image of one group of precomputation.These images carry out parametrization in the situation that meeting Minimum Mean Square Error most, just reflect more really reflecting effect when they are as Environment like this.Near reflecting body reference view can obtain good reflecting effect, draws the advantage of having utilized hardware supported Environment simultaneously.But shortcoming is very long pre-service (each viewpoint exceedes 20 minutes).

Biggest advantage of light track algorithm can calculate reflecting effect accurately, obtains high-quality drawing image.Ray tracing is lower than feed forward method efficiency, and main cause is that it need to spend sizable calculated amount and determines which pel has affected current output pixel point.A large amount of research work is devoted to how to accelerate biggest advantage of light track algorithm.Ward etc. have only realized with a general GPU who expands with floating-point vector point, reach the real-time ray-tracing rendering of little scene.Meanwhile, the speed of off-line ray tracing has also been accelerated in the development of hardware.And by parallel computer and the cluster of shared storage, can also realize the real-time ray tracing of large scale scene.Fixed function pipeline in original graphics accelerator, the pipeline able to programme that has nowadays also been supported summit and Pixel-level substitutes, so also there is the biggest advantage of light track algorithm based on programmable GPU.Although Programmable GPU is enough to process limited ray tracing, for present stage, still still main feedforward drawing engine of GPU.

Summary of the invention

Technology of the present invention is dealt with problems and is: compared with traditional three dimensional object paint reflection, what the three dimensional object based on nonuniform sampling reflected real-time rendering utilization is the sampling pattern of deforming template control reconfiguration low-resolution image, and then realize efficient paint reflection and calculate, improve the sense of reality of drawing effect.

The present invention proposes a kind of real-time drawing method based on the reflection of nonuniform sampling three dimensional object, it is characterized in that: specifically can be divided into following three steps:

(1) according to the reflecting body under current view point be reflected body, generate manually or automatically nonuniform sampling template: described sampling template is determined the distribution of the sampled point of piece image, described distribution is an irregular two-dimensional mesh trrellis diagram, described irregular referring to: after the two-dimensional grid distortion of rule, before the Area Ratio distortion of each little grid of correspondence image high-resolution areas, in the two-dimensional mesh trrellis diagram of rule, each little quadrilateral area is large; Before the Area Ratio distortion of the little grid in corresponding low resolution region, in the two-dimensional mesh trrellis diagram of rule, each little quadrilateral area is little; Sampling template is the set of net point after equally distributed two-dimensional mesh lattice point position is distorted, it has directly defined the change of each net point position, and how the pixel position of image will be mapped on the position of corresponding pixel points in continuous nonuniform sampling image; Non-deformation position (u, v) and width sampling template SM on a given original image, the deformation position (u that it is corresponding _d, v _d) position of searching SM upper (u, v) by the method for bilinear interpolation obtains;

(2) according to the sampling template generating and the high resolving power original image that is reflected body, can reconstruct obtain the image of the nonuniform sampling that is reflected body: the grid chart that is obtained a width two dimension distortion by step (1), the template of sampling SM, the binding of each grid vertex be the coordinate figure of corresponding regular grid; In conjunction with SM and given high-resolution original image, re-use mip pinup picture technology mipmapping, can reconstruct the deformation pattern of the low resolution of nonuniform sampling, on deformation pattern, each pixel need to be recorded depth value;

(3) reflecting body in scene and the nonuniform sampling depth map of deformation pattern reconstruct that is reflected the nonuniform sampling of body are carried out to reflecting effect drafting.

Compared with original algorithm, the main contributions of this algorithm is embodied in following 2 points: the resolution that is reflected volume image entirety is lower, but the focus area of visual attention location has higher resolution, can on the basis of amount of light of reducing entirety tracking, draw out the high-resolution reflecting effect of regional area like this; The automatic generation of sampling template, can realize the real-time rendering that in dynamic scene, three dimensional object reflects.All calculating that method of the present invention relates to are all carried out in GPU, can reach real-time performance.

Brief description of the drawings

The sample deformation process figure of template of Fig. 1;

Fig. 2 algorithm overall flow figure;

Fig. 3 reflection spot record diagram;

Fig. 4 focus area control chart;

Fig. 5 focus area circular fit figure;

Fig. 6 nonuniform sampling Image Reconstruction figure;

Fig. 7 reflecting effect figure;

The sharpness of the upper figure below Chinese word of Fig. 8 (a) is close, Fig. 8 (b). in upper figure below, the contour sharpness of lotus is close, the sharpness of the upper figure below Chinese word of Fig. 8 (c) is close, Fig. 8 (d). the upper medium and small bird of figure below details exhibit comparable with it.

Embodiment

The idiographic flow of three dimensional object paint reflection algorithm of the present invention is shown in Fig. 2, and its implementation is specific as follows

Step 1: the sampling template of obtaining distortion

Observe under normal circumstances an object, the emphasis that human vision is paid close attention to all can rest on the one or more special region of object, is referred to as visual focus region here.Nonuniform sampling will allow compared with the focus area of vision and other regions of image possesses higher resolution.The deformation process of the template of sampling in the present invention is, first constructs the rule mesh trrellis diagram of a g*h, manually specifies a circle here, corresponding visual focus region during circle inner region reflects.The center of circle is that O is the initial point of the coordinate system of current circle, to the arbitrary net point P in circle, makes it move the distance of F (P) along OP direction; In circle, all the other points do similar move operation.Wherein to move formula as follows for x direction of principal axis:

F(x)＝x+size*(1-x ²) (1)

Wherein, x is the horizontal ordinate that P is ordered, the value of x is mapped to [1,1], size is coefficient, according to deformation restriction relation F (n*m/r) the < F of two adjacent net points, ((n+1) * m/r, can calculate size < r/ (2n+1).Here n is the grid number object arbitrary integer being less than between Dao Yuan edge, the center of circle, the length of side that m is single grid, and r is the radius of a circle of specifying.In like manner, y axle does same variation, is exactly important medium---sampling template in rendering algorithm process in the present invention, as shown in Figure 1 after distortion.

In order to reach the real-time paint reflection of three-dimensional body, key is to improve the manually computational problem of sampling template, the dynamic sampling template that automatically generates.This method introducing focal point control figure realizes the generation of automatic sampling template.The construction algorithm of focal point control figure below:

Reflecting body A in scene, is reflected body B, current view point V in scene;

A. the light ray that arrives the summit S of A from V, judges that whether the reflection ray of ray is crossing with B.If intersect, obtain joining S ' and record its position on impostor (representing the view plane of viewpoint in B dead ahead with impostor); Otherwise, judge next summit.Figure 3 shows that in experiment scene of the present invention, under certain viewpoint, the reflection spot location drawing that the spherical reflecting body of use obtains;

B. impostor is divided into the two-dimensional grid of specification, the each quadrilateral area on traversal grid records reflection spot number wherein.Use statistical principle to find out the distribution situation of the upper reflected ray intersection point of impostor, find out all quadrilaterals the mark that outnumber certain threshold value of intersection point, if Fig. 4 is exactly the focal point control figure using in the present invention.The densely distributed marked region of point is fitted in a circle as far as possible, finds center of circle O and the radius r of this circle, as Fig. 5;

C. the circle in back is carried out to deformation process, use formula (1) to calculate the variation of coordinate in length and breadth of each point in circle, thereby the grid chart that obtains a distortion is sampling template SM required in the present invention.

Step 2: the nonuniform sampling image of reconstruct low resolution

The sampling template SM being obtained by previous calculations and given high-resolution original image I, can reconstruct the deformation pattern of nonuniform sampling.Concrete steps are as follows:

I) structure size is the two-dimentional regular grid QM of α * β;

A. vertex v _ij=SM (i, j);

B. texture coordinate;

Ii) with resolution w _c* h _cto the painted (w of QM _c* h _crepresent the image resolution ratio of reconstruct);

C. according to sizing grids different in SM, utilize I to carry out mipmapping and search the each pixel (s that determines QM, t) (s, t represents respectively the coordinate figure in length and breadth of each pixel in QM) pixel value located, so just can reconstruct the deformation pattern (seeing Fig. 6) of low resolution;

Here the i.e. texture coordinate of this point of coordinate without distortion of grid vertex binding.

Step 3: paint reflection

Carry out paint reflection with depth map and can overcome motion parallax and object and move the problems such as the uncontinuity that causes, the present invention in the time of paint reflection, to use step 2 equally in the depth information of image of final distortion.

Deformation pattern depth image DI is crossing with light ray to be calculated as follows:

If the bounding box of ray and DI is non-intersect, return without intersection point.Otherwise the bounding box of DI carries out cutting to ray obtains light ray _c;

By ray _cin the plane of projection DI, obtain line segment ray _p;

To ray _pcarry out rasterizing processing, search ray _pintersection point with DI height territory.If exist intersection point just painted to it, if do not exist and be left intact.

The reflecting effect of the three dimensional object that use the inventive method is drawn as shown in Figure 7, for different reflection graphic picture in figure, in the case of the resolution in visual focus region is similar, in table 1, provided difference and be reflected the traditional images of body and the comparison of nonuniform sampling image size, its different reflecting effect is shown in Fig. 8.

The comparison of table 1 different situations hypograph size

Reflected image	Fig. 8 a	Fig. 8 b	Fig. 8 c	Fig. 8 d
					Non-homogeneous	518	578.1	442.4	682.2
Tradition	2334.7	1873.9	2211.8	2099.2

Table 1 is traditional images and nonuniform sampling image relatively, the form of image is all bmp, in table, the unit of data is K, if theirs possesses close resolution at focus area, the memory capacity of non-uniform image is more much smaller than even image, and the advantage in the drafting efficiency of the inventive method has showed soon so.

Fig. 8 is non-homogeneous and even image reflecting effect comparison diagram, in figure, one arranges the high-definition picture reflecting effect that represents uniform sampling above, represents the ground image in different resolution reflecting effect of nonuniform sampling below; From relatively can finding out of Fig. 8 (a)-(d), the even image of the non-uniform image of low resolution and high-resolution is after paint reflection, and on focus area, their drafting effect is very nearly the same.So method of the present invention does not reduce the sense of reality effect of paint reflection in improving drafting efficiency, has promoted efficiency in the situation that not sacrificing effect.

Claims (1)

1. the real-time drawing method based on the reflection of nonuniform sampling three dimensional object, is characterized in that: specifically can be divided into following three steps:

(1) according to the reflecting body under current view point be reflected body, generate manually or automatically nonuniform sampling template: described sampling template is determined the distribution of the sampled point of piece image, described distribution is an irregular two-dimensional mesh trrellis diagram, described irregular referring to: after the two-dimensional grid distortion of rule, before the Area Ratio distortion of each little grid of correspondence image high-resolution areas, in the two-dimensional mesh trrellis diagram of rule, each little quadrilateral area is large; Before the Area Ratio distortion of the little grid in corresponding low resolution region, in the two-dimensional mesh trrellis diagram of rule, each little quadrilateral area is little; Sampling template is the set of net point after equally distributed two-dimensional mesh lattice point position is distorted, it has directly defined the change of each net point position, and how the pixel position of image will be mapped on the position of corresponding pixel points in continuous nonuniform sampling image; Non-deformation position on a given original image (u, v) and width sampling template SM, the deformation position (u that it is corresponding _d, v _d) by the method for bilinear interpolation search SM upper (u, position v) obtains; The acquisition process of described sampling template is:

First construct the rule mesh trrellis diagram of a g*h, specify a circle, corresponding visual focus region during circle inner region reflects; The center of circle is that O is the initial point of the coordinate system of current circle, to the arbitrary net point P in circle, makes it move the distance of F (P) along OP direction; In circle, all the other points do similar move operation; Wherein to move formula as follows for x direction of principal axis:

F(x)=x+size*(1-x ²)

Wherein, x is the horizontal ordinate that P is ordered, the value of x be mapped to [ 1,1], size is coefficient, according to deformation restriction relation F (n*m/r) <F ((n+1) * m/r of two adjacent net points, calculate size<r/ (2n+1), here n is the grid number object arbitrary integer being less than between Dao Yuan edge, the center of circle, the length of side that m is single grid, and r is the radius of a circle of specifying; Y axle does same variation, obtains the template of sampling after distortion;

(2) according to the sampling template generating and the high resolving power original image that is reflected body, can reconstruct obtain the image of the nonuniform sampling that is reflected body: the grid chart that is obtained a width two dimension distortion by step (1), the template of sampling SM, the binding of each grid vertex be the coordinate figure of corresponding regular grid; In conjunction with SM and given high-resolution original image, re-use mip pinup picture technology mipmapping, can reconstruct the deformation pattern of the low resolution of nonuniform sampling, on deformation pattern, each pixel need to be recorded depth value;

(3) reflecting body in scene and the nonuniform sampling depth map of deformation pattern reconstruct that is reflected the nonuniform sampling of body are carried out to reflecting effect drafting.

Images