KR100441079B1 - apparatus and method for antialiasing - Google Patents

Abstract

An object of the present invention is to provide a method for generating a high quality image at an economical hardware cost. In a three-dimensional computer system comprising an API, a three-dimensional graphics accelerator, and a display unit, the three-dimensional graphics accelerator is provided through the API. Converts the transmitted 3D model data into a geometry processing unit that performs rotation, coordinate change, and size conversion, and converts the 3D data output from the geometry processing unit into 2D screen coordinates, and forms polygon data that forms a 3D model. A rasterizer for generating a fragment for the image, an image generator for performing an image mapping process for mask removal and a sensory image with respect to the generated fragment, and a surviving fragment among newly inputted fragments A frame buffer for storing the area mask, depth value, and color value of the segment; Vibe de standing fragment of that was stored in the frame buffer area mask of the fragment, and is composed, including RUF buffer for storing the object tag and the color values.

Description

Apparatus and method for antialiasing

TECHNICAL FIELD The present invention relates to three-dimensional computer graphics, and in particular, to a simple and efficient hardware assisted antialiasing algorithm and rendering structure.

3D computer graphics is the most important research field to build a multimedia environment. Recently, the 3D graphics field has expanded greatly in quality and quantity.

1 is a diagram illustrating a general three-dimensional computer graphics system.

Referring to FIG. 1, when the 3D application software is largely delivered to the 3D graphics accelerator 200 through an API (Application Program Interface) 100, the graphic system performs real-time hardware acceleration in the 3D graphics accelerator 200. Then it is sent to the display unit 300 consists of a process of displaying on the screen.

In this case, the 3D graphics accelerator 200 passes the 3D model data transmitted through the API 100 to the rendering processor 220 while undergoing a process such as rotation, coordinate change, and size conversion in the geometry processor 210. Is sent.

Then, the rasterizer unit 221 of the rendering processor converts the 3D data processed through the geometry processor 210 into 2D screen coordinates, and generates fragments for polygon data forming a 3D model. do.

Subsequently, the image generation unit 222 of the rendering processor performs an image mapping process for removing the hidden face and the sensory image on the generated fragment, and stores the result in the frame buffer 223.

When the above process is completed for all the fragments generated by the rasterizer unit 221, the color values stored in the frame buffer 223 are transferred to the display unit 300 and displayed on the screen.

However, when the polygon processed by the geometry processor 210 is output to the screen from the display 300 through the rendering processor 220, a common problem is that a sawtooth edge is formed in the image.

This effect is referred to as stepping or sawing, which is an example of what is known as aliasing. This problem is mostly caused by insufficient sampling when creating fragments.

Therefore, in order to solve the above aliasing problem, a supersampling method (Alan Watt. '3D Computer Graphics', Third Edition, Addison-Wesley, 2000.).

However, the supersampling method requires a large memory size and a memory bandwidth.

For example, n-supersampling using n points per pixel requires n times the memory size and memory bandwidth, respectively, than the 1-point sampling method.

Therefore, there is an accumulation buffer technique as a way to reduce the memory requirements. After rendering several times on one screen, each result is stored in an accumulation buffer, and then It is a technique to generate the final screen by storing the weight in the frame buffer with appropriate weight.

However, since the accumulation buffer technique is a multi-pass algorithm that repeats rendering several times on one screen, the accumulation buffer technique has a problem in that the rendering speed is slowed in proportion to the number of repetitions.

On the other hand, there is an A-buffer algorithm that supports blending of transparent objects as well as antialiasing for opaque objects, which represents the area occupied by subpixels using a coverage mask, and the average of these areas. Express fragments with color values (Loren Carpenter, "The A-buffer, an Antialiased Hidden Surface Method", InComputer Graphics Annual Conference Series (Proceedings of SIGGRAPH 84), volume 18, page 103-108, July 1984.) .

Accordingly, the A-buffer method has n subpixels per pixel, and the n-supersampling method requires one color value and a depth value for each subpixel, whereas the fragment representation through the area mask is a sampling point. Redundant color values can be removed to reduce memory size and memory bandwidth.

However, to show a relationship where polygon fragments of multiple objects overlap or intersect with each other, a list of fragments sorted by depth-per-pixel values is required, which makes the rendering pipeline design difficult. It works as a problem.

Also, although A-buffer algorithms can handle transparent objects well, they still require fragment lists for antialiasing of opaque objects, and in some cases even more hardware costs than supersampling.

Accordingly, the present invention has been made to solve the above problems, and expresses the fragment using an area mask based on the A-buffer algorithm, and is opaque using the most recently used fragment color value. The purpose is to provide a rendering structure and a method that can effectively perform antialiasing on objects.

Another object of the present invention is to provide a method for generating a high quality image at economical hardware cost.

1 illustrates a typical three dimensional computer graphics system.

2 illustrates a three-dimensional computer graphics system in accordance with the present invention.

3 is a diagram showing a data structure of a sampling point and a fragment in the case of having eight sampling points;

4 is a schematic diagram of a rendering pipeline for antialiasing according to the present invention.

5 illustrates the interrelationship of fragments that may generally occur during the rendering process;

6 is a view showing an embodiment of determining the color value according to the arrival order and correlation of the fragment

7 is a diagram illustrating an average color error according to each sampling method of 3D model data.

* Explanation of symbols for main parts of the drawings

100: API 121: Depth Buffer

122: color buffer 200: three-dimensional graphics accelerator

210: geometry processor 220: rendering processor

221: rasterizer 222: image generator

223: frame buffer 224: RUF buffer

300: display unit

A feature of the antialiasing device according to the present invention for achieving the above object is a three-dimensional computer system consisting of an API, a three-dimensional graphics accelerator and a display, wherein the three-dimensional graphics accelerator is a three-dimensional delivered through the API The geometric processing unit that rotates the model data, changes the coordinates, and transforms the size, and converts the 3D data output from the geometric processing unit into two-dimensional screen coordinates, and a frag for polygon data forming a three-dimensional model. A rasterizer for generating a fragment, an image generator for performing an image mapping process for removing a phantom and a sensory image with respect to the generated fragment, and a region of a survived fragment among newly inputted fragments A frame buffer that stores masks, depth values, and color values; It is composed of an RUF buffer that stores the area mask, object tag, and color value of the survived fragment in the segment.

Here, the generated fragment includes an area mask indicating which part of the pixel is covered, the color value of the fragment, the depth value of each subpixel occupied by the fragment, and the fragment in the RUF buffer. It is preferable to include an object tag (ObjTag) used as a flag when merging the merge.

A feature of the antialiasing method according to the present invention for achieving the above object is to compare the depth value of each subpixel of the newly stored fragment and the fragment stored in the frame buffer, the depth value of the overlapping area is large Masking the fragment region and obtaining a survived fragment region, which is a fragment region having a small depth value, detecting a new region mask, and detecting a new region mask of the newly input fragment. Updating an area mask, a depth value, and a color value in a frame buffer, and storing an area mask, an object tag, and a color value of a survived fragment among the fragments stored in the frame buffer in an RUF buffer; Comparing an object tag stored in the RUF buffer with an object tag of a fragment previously stored in the RUF buffer And if the value is the same, performing OR bit operations on the area masks of the two fragments, calculating the color values and the area masks of the areas occupied by each other, and storing the area masks in the RUF buffer. As a result of the comparison, if the values are different, updating the RUF buffer with an area mask, an object tag, and a color value of a survived fragment among the fragments stored in the frame buffer, the color value stored in the frame buffer and Synthesizing and displaying the depth values.

The area mask updated in the frame buffer may be calculated by calculating an area mask stored in a frame buffer and an area mask of the newly detected surviving fragment through an OR bit operation.

The color value to be updated in the frame buffer silver

or

It is preferable to calculate using an equation.

In the step of storing in the RUF buffer area mask ( ) And color values ( ) Calculation Wow,

It is preferable to calculate.

Preferably, the fragment has a plurality of sampling points.

Other objects, features and advantages of the present invention will become apparent from the following detailed description of embodiments taken in conjunction with the accompanying drawings.

A preferred embodiment of an antialiasing method and apparatus according to the present invention will be described with reference to the accompanying drawings.

Basically, the fragment structure is based on the A-buffer algorithm, and the three-dimensional computer graphics system according to the present invention is shown in FIG.

As shown in FIG. 2, the 3D computer graphics system is largely composed of an API 100, a 3D graphics accelerator 200, and a display unit 300.

In addition, the 3D graphics accelerator 200 is a geometric processing unit 210 for performing rotation, coordinate change and size conversion, etc. 3D model data transmitted through the API 100, and calculated by the geometric processing unit 210 The rendering processor 220 generates a fragment from three-dimensional data and stores a color value and a depth value of the final pixel through image mapping.

In this case, the rendering processor 220 converts the 3D data that has undergone the process of the geometry processor 210 into the 2D screen coordinates through the rasterizer 221, and converts the polygon data that forms the 3D model into the 2D screen coordinates. After generating a fragment and performing the image mapping process for the mask removal and the sensory image on the generated fragment through the image generator 222, the result is stored in the frame buffer 223 and transferred. The fragment information is stored in the RUF buffer 224.

In this case, the frame buffer 223 includes a color buffer for storing the color value of the pixel and a depth buffer for storing the depth value of the pixel.

The RUF buffer 224 stores an object tag, a color value, and an area mask for the fragment portion that passes the mask removal during the rendering process, that is, the fragment portion most recently used to calculate the color value of the color buffer. do.

Subsequently, when the above process is completed for all fragments generated by the rasterizer 221, the display unit 300 outputs the color values stored in the frame buffer 223 to the screen.

3 shows a data structure of a sampling point and a fragment in the case of having eight sampling points.

As shown in FIG. 3, the fragment includes an area mask (m) indicating which part of the pixel is covered, a color value (C) of the fragment, and a depth value (Z _{1.) For each} subpixel occupied by the fragment. ..Z _m ), and an object tag (ObjTag) that is used as a flag when merging fragments in the RUF buffer.

The object tag is a unique value of the 3D model object, and is sequentially generated in the rendering hardware through linkage with modeling software. If two different fragments have the same object tag, it means that the two fragments were created in the same object polygon. If these fragments have the same object tag, the fragments are generated during RUF writing. Are merged.

In addition, the area mask is represented by m bits, and as shown in FIG. 3, the area mask is represented by 8 bits as shown in FIG. 3.

The color value C and the depth value Z are each represented by 32 bits, and the object tag is represented by 16 bits.

FIG. 4 is a schematic diagram of a rendering pipeline for antialiasing according to the present invention. As shown in FIG. 4, a three-stage pipeline process of mask removal inspection, framebuffer storage and color value determination, and RUF buffer recording is performed. Perform rendering through

First, for ease of explanation, if, ff, and RUF are subscripted for the new fragments in the rendering pipeline, the fragments stored in frames, and the property information of the fragments recorded in the RUF buffer. Express by using

For example, M _if , C _ff , and C _RUF refer to the region mask of the new fragments in the rendering pipeline, the color values written into the framebuffer, and the color values of the fragments written to the RUF buffer, respectively. .

In the first step, the mask removing inspection process 111 compares a depth value for each subpixel of a new fragment and a fragment stored in the frame buffer 223. The region of the fragment having a large depth value is removed from the overlapping region, and only the region of the fragment having a small depth value is stored. This is the same as the existing Z-buffer technique.

The fragment portion that has passed the mask removal inspection through the depth value comparison is referred to as "survived fragment (subscript sur)", and the region mask M _sur for the subpixel of the portion is expressed by the following equation. Obtained through 1.

In Equation 1, the subscript i indicates the position of the sampling point, that is, the position of each subpixel.

The second step is the frame buffer storage and color value determination process 112.

The survived fragment obtained during the mask removal process is a visible portion of the fragment newly entered into the rendering pipeline. Therefore, the area mask, depth value, and color value of this part should be added to the frame buffer 223.

The obtained new area mask is calculated by OR bit operator of the area mask M _ff stored in the frame buffer 223 and the area mask M _sur of the survived fragment, and the sub-section of the survived fragment. The frame buffer 223 is updated to a depth value of the pixel.

The color value determination process adds the color value of the survived fragment to the color buffer 122 and the color value of the portion that is removed by the surviving fragment, that is, the part that is masked, by the color buffer 122. Remove from).

5 is a diagram illustrating an interrelationship of fragments that may generally occur during a rendering process.

5 (a) and (b) shows a case in which one fragment covers all or part of another fragment, and FIG. 5 (c) and (d) shows a case in which the fragments penetrate each other.

As mentioned earlier, the RUF buffer 224 stores the information of the surviving fragments that most recently affected the color value of the frame buffer 223, and the fragment obtains the color value of the hidden part. To be used again.

5 (a) (b) (c), considering the order of fragments entering the rendering pipeline, the fragment stored in the RUF buffer is F _a , which is determined by fragment F _b with reference to this value. You can see the color value of the mask removed, F _ab .

In this case, the color value may be calculated using Equation 2.

In other words, a color value obtained by weighting the area mask of the survived portion of the fragment F _b is added to the current color buffer 122, and the portion of the fragment F _a is hidden with reference to the RUF buffer 224. , F _ab Remove as much as contributed to the color value.

In Equation 2, the function N () returns the number of bits of 1 in the area mask. T _m is the number of bits of the area mask. That is, the number of sampling points per pixel.

However, in FIG. 5 (d), some of the parts covered by the fragment F _c , for example F _ca , cannot be accurately calculated with reference to the RUF buffer 224.

This is because the current RUF buffer 224 only has information of F _b , and the frame buffer 223 has only information of F _a .

Therefore, in this case, the F _ca portion is calculated using Equation 3.

In other words, the F _ca part is corrected by removing color values given the area weights of the current color buffer.

The antialiasing method according to the present invention will be described in detail with reference to the embodiments with reference to the drawings.

6 is a view showing an embodiment of determining the color value according to the arrival order and correlation of the fragment.

6, fragments A, B, and C are all part of different objects, and the order in which the fragments arrive in the rendering pipeline is fragments A-> B-> C. In this case, the fragments overlap or cross each other.

First, when fragments A and B arrive, the color value, C _AB, is represented by Equation 4 below. In Equation 4, C _Φ means an initial state.

After performing this process, the frame buffer 223 stores the color value and the depth value of the fragment A, and the area mask, the color value, and the object tag of the fragment B are stored in the RUF buffer 224.

Subsequently, when fragment C arrives, two cases of Fig. 5 (a) (b) can be considered.

As in the case of FIG. 5A, if the depth value of fragment B is lower in all regions than the depth value of fragment A, fragment R of the portion corresponding to all pixels is included in RUF buffer 224. It will store the color value and depth value of.

Therefore, using the information of the fragment B stored in the RUF buffer 224 and the information of the input fragment C, the color value of the pixel can be obtained using Equation 2 as shown in Equation 5 below.

However, when fragments A and B do not intersect but only a portion of fragment A is removed as shown in FIG. 5 (b), the information according to the portion of F _a previously stored in the RUF buffer 224 is removed. Is cleared and only stores information based on F _b .

Therefore, since the information according to the F _aa portion of the fragment A is not known, when the color value is calculated using Equation 3, Will cause a color error.

The color error can be prevented in the final third step, the RUF buffer write process 113.

In the third step of the RUF buffer writing process, polygons forming an object generally exist in the same coplanar space, and neighboring polygons generate fragments that share pixels near a boundary line.

In addition, since these fragments are exclusive to each other, they can be merged and treated as one fragment.

As shown in Equation 2 and Equation 3, as the fragment portion stored in the RUF buffer 224 occupies a large area in the pixel, the portion that can provide color value information about the hidden portion becomes wider. This provides an opportunity to minimize errors in determining color values.

Thus, the following equation 6 is used to merge the two fragments in the RUF buffer 224.

In this way, the object tags of the two fragments are compared to merge the fragments stored in the survived portion and the RUF buffer 224.

As a result of the comparison, if the values are equal to each other, the fragments are generated from polygons of the same object, and thus OR bit operations are performed on the region masks of the two fragments, and the color values of the regions occupied by using the weight value Calculate and save.

As a result of the comparison, if the object tag is different, the RUF buffer 224 is updated with an area mask, an object tag, and a color value of the surviving portion.

When the rendering process for all fragments is finished, the color value stored in the frame buffer 223 is sent to the display unit 300 to output an image.

The effect of this method is clearly shown through the comparison experiment with the final image obtained through this method and the existing supersampling method.

Table 1 shows three-dimensional model data used in the experiment, Figure 7 is a view showing the average color error according to each sampling method of the three-dimensional model data shown in Table 1.

model name Vertex count Triangle number Number of fragments Number of objects Al 3618 7124 11975 35 Castle 6620 13114 17444 16 Dolphins 885 1692 4570 3 Pig 3522 7040 7499 3 Pose + vase 4028 3360 5425 5 Teapot 3644 6320 6807 One Venus 711 1418 5464 One

As shown in FIG. 7, it can be seen that the proposed method exhibits better image quality when there are a large number of sampling points, and that the proposed method having the same number of sampling points shows an image almost similar to that of eight supersampling.

Table 2 shows the memory size and average color error required per pixel of this method and supersampling with eight-sparse sampling points.

As can be seen from Table 2, the error increase rate of the proposed method is 1.3%, while the memory size decreases by about 31%. This represents a good way to minimize color errors and reduce the required memory size compared to supersampling with the same sampling point.

8ss The present invention Change ratio Average color error 1338250 1336154 -1.3% Memory size required per pixel 64 bytes 44 bytes + 31.3%

Table 3 shows the size of memory required per pixel in the supersampling method and the proposed method as the sampling point increases.

As shown in Table 3, it can be seen that as the sampling points increase, the memory saving effect of the proposed method increases.

Number of sampling points Supersampling The present invention Reduction ratio Memory size required (bytes) 4 32 27 16% 8 64 44 31% 16 128 78 39% 64 512 282 45%

Table 4 shows the results of calculating the memory bandwidth required to generate any particular image for each experimental data set.

As can be seen from Table 4, it can be seen that the larger the number of subpixels of a fragment, the greater the memory bandwidth saving effect. This is because, unlike supersampling, the overhead for object tags, area masks, etc., added to represent fragments in the proposed scheme decreases as the number of subpixels of the fragment increases.

model name 8ss The present invention Reduction ratio Average number of subpixels per fragment Memory Bandwidth (Bytes) Al 518512 456654 11.9% 2.9 Castle 614394 574144 6.5% 2.8 Dolphins 229144 194176 15.3% 3.2 Pig 301489 279204 7.4% 2.7 Pose + vase 173760 163487 5.9% 2.6 Teapot 281472 260347 7.5% 2.7 Venus 338332 268950 23.5% 3.9

As described above, the anti-aliasing method and apparatus according to the present invention can effectively save memory size and memory bandwidth while providing a high-quality antialiased image at a level comparable to that of supersampling. Moreover, as the sampling point increases, the effect increases relatively.

In addition, the apparatus according to the present invention can be implemented by the addition of relatively simple hardware, and can also be easily integrated into the existing system because it uses the z-buffer algorithm, which is a representative mask removal technique used in the existing rendering system. .

Those skilled in the art will appreciate that various changes and modifications can be made without departing from the spirit of the present invention.

Therefore, the technical scope of the present invention should not be limited to the contents described in the embodiments, but should be defined by the claims.

Claims (8)

In a three-dimensional computer system consisting of an API, a three-dimensional graphics accelerator and a display unit, The 3D graphics accelerator is a geometry processing unit for rotating, coordinate change and size conversion of the 3D model data transmitted through the API; A rasterizer unit for converting 3D data output from the geometry processor into 2D screen coordinates and generating fragments for polygon data forming a 3D model; An image generator which performs an image mapping process for removing the mask surface and the sensory image on the generated fragments; A frame buffer that stores the area mask, depth value, and color value of the survived fragment among the newly entered fragments; And an RUF buffer configured to store an area mask, an object tag, and a color value of a survived fragment among the fragments stored in the frame buffer. The method of claim 1, The frame buffer is configured to include a color buffer for storing the color value of the pixel and a depth buffer for storing the depth value of the pixel. The method of claim 1, The generated fragment includes an area mask indicating which part of the pixel is covered, a color value of the fragment, a depth value for each subpixel occupied by the fragment, and a fragment in the RUF buffer. Anti-aliasing device comprising an object tag (ObjTag) used as a flag when merging. Compares the newly entered fragment with the depth value of each subpixel of the fragment stored in the frame buffer, and removes the fragment area with the larger depth value from the overlapping area and removes the fragment area with the smaller depth value. Detecting a new region mask by obtaining a vibes fragment region; The region mask, depth value, and color value of the survived fragment among the newly input fragments are updated in the frame buffer, and the region mask of the survived fragment among the fragments stored in the frame buffer. Storing the object tag and color values in the RUF buffer, Comparing an object tag stored in the RUF buffer with an object tag of a fragment previously stored in the RUF buffer; As a result of the comparison, if the values are the same, performing OR bit operations on the area masks of the two fragments, calculating the color values and the area masks of the areas occupied by each other, and storing the area masks in the RUF buffer; If the value is different, updating the RUF buffer with an area mask, an object tag, and a color value of a survived fragment among the fragments stored in the frame buffer; And synthesizing and displaying the color value and the depth value stored in the frame buffer. The method of claim 4, wherein The area mask updated in the frame buffer is calculated by calculating an area mask stored in a frame buffer and an area mask of the newly detected survived fragment through an OR bit operation. The method of claim 4, wherein The color value to be updated in the frame buffer silver or Antialiasing method, characterized in that the calculation using a formula. Where: C: color value, M: area mask, Tm: number of bits of area mask, subscript if: newly entered fragment, subscript ff: fragment stored in frame, and subscript RUF: RUF buffer Indicates. The method of claim 4, wherein In the step of storing in the RUF buffer area mask ( ) And color values ( ) Calculation Wow, Antialiasing method characterized in that it is calculated as. The method of claim 4, wherein And said fragment has a plurality of sampling points.