Romaniuk S. O., Romaniuk O. N., Piddubetska M. P.

Vinnytsia National Technical University

BASIC ANALYSIS OF TEXTURE FILTERING METHODS

In this article the main methods texture filtering are treated: point sampling, bilinear filtering, trilinear, anisotropic filtering. The basic principles, the classification, the main advantages and disadvantages of these methods of filtration are describes.

Key words: texture, texel, point sampling, bilinear filtering, trilinear filtering, anisotropic filtering.

One of the key issues of modern computer graphics is increasing realistic of images that are synthesized. This issue is very important in solving many applied tasks. These include tasks such as: visualization system for scientific research, medical and technical diagnostics, building visual models of processes and phenomena, using computer graphics to create films and computer games. New technologies for 3D-visualization are continuously developed; algorithms for increasing memory bandwidth are improved and upgraded. An important part of a 3D-visualization is texturing.

Texture is a two-dimensional raster image that is overlay on the surface of the object. Texture filtering – is a mechanism that provides texture blending to the polygons which are differ in size. Color of pixel projection of a three-dimensional object on the screen should clearly define to color of texel of proper texture. But this is true only in simple cases when projecting angles close to normal, with the appropriate distance from the image plane. Pixel and texel only in the mathematical sense referred to as point: physically they are very specific size and look like a circle whose diameter depends on the screen resolution of the monitor and on texture. As the projection angle is beyond the certain limits, it happens that one pixel projected two or more texels and then the form of the projection is close to oval (Fig. 1). If the object is very close to the plane of projection - one texel is projecting into several pixels [1].

Figure 1 - The elliptical shape of the light spot

The process of blending textures on the surface leads to the appearance of artifacts and methods of their elimination. This paper reviews the main methods of texture filtering: point sampling, linear filtering, bilinear filtering, trilinear and anisotropic filtering (Fig. 2).

Figure 2 – The classification of methods for texture filtering

Analysis of the main methods of filtering

Point Sampling – is the easiest way to determine the color of a pixel in texture -based image. To implement this method you must select texel, which is closer than the others located to the center of the light spot (Fig. 3) [2].

Figure 3 – Point sampling method

The main advantage of this method of filtration is low demands on memory bandwidth, for determining pixel colors you must select only one pixel of texture memory.

The main disadvantage of this method – is if the polygon is located close to the screen (or from the point of observation), the number of pixels is greater than the number of texel leading to general deterioration of image quality If such a texture overlaid over the object that goes into the depth of the scene, perspective effect is greatly distorted. This method was widely used until the hardware accelerators were invented. After the distribution of hardware accelerators it was possible to use more qualitative methods.

Bilinear filtering - the method that uses interpolation techniques. To determine the texels to be involved for interpolation, commonly is used basic form of the light spot - a circle. In fact, the circle is approximated by four texels (Fig. 4). This filtering method is significantly more productive than the method of point filtering, as taken into account the shape of the light spot and used interpolation. The resulting color of pixel is determined by blending operation. First blended colors of two pairs of texels on the X-axis, then two obtained colors are blended along the Y-axis. More texels are need for the interpolation than actually available if the polygon is approaching too close to the screen or to the observation point. As a result blurred images are generated with very high quality, but this is just a side effect of this type of filtering.

Figure 4 - The method of bilinear filtering

The main disadvantage of bilinear filtering is that approximation performed correctly only for polygons that are disposed parallel to the screen or to the observation point. If the polygon turned at an angle to the screen (which is in 99% of cases) the approximation is incorrect. In this case, it is used a circle approximation, while should be used approximation of the ellipse. With bilinear filtering you have to read four texels from texture memory to determine the color of each pixel that is displayed on the screen, which means that the demands on memory bandwidth increased four times in comparison with a sample pointing filtration.

Trilinear filtering is a method of filtering that combines mip-texturing and bilinear filtering. Mip‑texturing is a method of reducing the amount of calculations required for accurate overlay texture image on the polygon. [3]

Trilinear filtering has the advantage of hardware simplicity and efficiency at the expense of visual quality. Instead of calculating the shape of the projected track of light beam, this method uses the square of filter in texture space. Blending two bilinear filter from adjacent map-level trilinear filtering forms an approximate to a circle filtering area to form the area of random size.

Trilinear filtering as bilinear uses blocks of four texels, the color of the pixel that should be deduced on the screen is determined by interpolating of two mip-textures. Mip-map levels are beforehand calculated smaller versions of the original texture, which means, a qualitative approximation of texels located in the spot light is obtained.

The last stage is analysis boundaries of the two blocks, corrected any mistakes and inconsistencies at the boundaries of these two blocks. In bilinear filtering it is quite common to see lines that occur at the boundaries of blocks that disappear when in trilinear filtering. In addition, trilinear filtering is better neutralized distortions and irregularities in motion and when the angle of observation changes.

Nowadays support of trilinear filtering has become a standard feature in graphics chip, because it provides the formation of three-dimensional images with high quality. This method has disadvantages: there are some defects at a greater distance, because this method was originally developed to reduce the distortion between mip-map levels. Images formed very efficiently only at the direct corners of observation. In real visualization the geometric shape of the object can be affected.

The form of textured objects in the bilinear and trilinear filtering may be distorted because both of this filtering are isotropic - the image is filtered in some form - in the form of a square. The majority of the formed objects do not fit into this certain and immutable form: for their high-quality processing it is necessary to use a different type of filtering – anisotropic filtering.

The term "anisotropy" consists of several Latin words: "an" for not, "iso" for same, and "tropic" from tropism, relating to direction; anisotropic filtering does not filter the same in every direction. The name of this technology reflects its technical implementation. Anisotropic filtering typically operates at least 8th texels, using the undefined form model. As a result, this filtering removes noise and distortion of objects, and the whole image is a more qualitative.

For today the best results in formation of three-dimensional scenes provides anisotropic filtering. Texels in this type of filter are selected not from the symmetrical region, but taken from the perspective corrected field (anisotropic form). The shape of this area is not advanced, but commonly used rectangular or trapezoidal area or areas in the form of a parallelogram - it all depends on the angle of the surface location (Fig. 5).

Figure 5 - The method of anisotropic filtering for texture image

The level of filtering is determined by the number of Texel processed in the calculation of final pixel. At the level 1x anisotropic filtering uses eight texels. On modern graphics cards filtration level can be put in the drivers. The maximum level is 16x, but producers do not use all the intermediate steps. The most common level is 2x (16 Texel), 4x (32 Texel), 8x (64 Texel) and 16x (128 Texel). When the level of filtration is increases computing workload also increases.

Like bilinear and trilinear filtering, anisotropic filtering eliminates the effects of aliasing, but this method also reduces the degree of blur and keeps the image details at extreme viewing angles (Fig. 6).

Figure 6 – The Comparison of trilinear and anisotropic filtering

Anisotropic filtering method has some disadvantages: the realization of anisotropic filtering is quite complicated and the rate of formation of dynamic scenes significantly decreases [4].

Conclusions

In this work classification and comparative analysis of the basic methods of filtering of texture images was carried out. Bilinear filtering requires less time for the formation of three-dimensional scenes, but has a major disadvantage: visible transitions between the boundaries of mip-levels. Best results are obtained in trilinear and anisotropic filtering, where the transitions between mip-levels are calculated correctly. In the near future further development and using of anisotropic filtering is expected, which requires further study this area in computer graphics.

Literature:

1. Гусятин В.М., Чаговец Я. В., Кожушко Д. Г. Метод анизотропной фильтрации текстур при синтезе зображений обратным трассированием / В.М. Гусятин, Я. В. Чаговец, Д. Г. Кожушко // Наукові праці ДонНТУ, серія «Інформатика, кібернетика та обчислювальна техніка». – 2009.

2. Роджерс Д. Алгоритмические основы машинной графики/ Д. Роджерс: Пер. с англ. – М.: Мир, 1989.

3. McCormack Joel, Farkas K. I., Perry R., Jouppi N. P.. Simple and Table Feline: Fast Elliptical Lines for Anisotropic Texture Mapping. / Joel McCormack, K. I. Farkas, R. Perry, N. P. Jouppi. Western Research Laboratory. – 1999.

4. Foley J. D., van Dam A., Feiner S. K., Hughes J. F., Computer graphics (principles and practice)./J. D. Foley, A. van Dam, S. K. Feiner, J. F. Hughes //Addison-Wesley Publishing Company, Inc. – 1996.