DE19853608C2 - Method for displaying an autostereoscopic image - Google Patents

Description

The invention relates to a method according to the preamble of claim 1 for Dar provision of autostereoscopic images.

Stereoscopic film and projection processes have been in use for years. Mostly Polarized light (horizontal / vertical, circular) used to image the right and left separate. With the advancement of LCD technology, it became possible to increase light transmission of crystals to be controlled electronically. This made the development of the shutter technique possible, in which the right and left alternate with the field frequency The lens becomes opaque and the right and left images are sequentially synchronized appear on the screen.

Autostereoscopic projections are made with the help of screens with streaked lens grids performed with several projection directions. The corresponding one Direction assigned the correct perspective image [R. Börner: Autosteroscopic 3-D imaging by Front and Rear Projection and on Flat Panel Displays, Displays Vol. 14. 1 (1993), pp. 39-46]. A smooth transition from one perspective to the next is hardly possible achieve, because the number of projection directions can not be increased arbitrarily. at an autostereoscopic display that is intended for one person only is used two perspectives that require a certain viewing direction [EP 0 836 332 A1; R. Börner: Autostereoskope rear projection and flat screens, television and Kinotechnik Vol. 48, No. 11 (1994). Pp. 594-600]. The representation of a stereoscope On the other hand, images like in a hologram are created using "head tracking" - Sensors achieved, on the one hand, a powerful real-time computer for calculating the suitable stereoscopic image perspectives and on the other hand an autostereoscopic screen control for tracking the optical beam path [EP 0 836 332 A2; S. Hentschke: Personal adaptive autostereoscopic monitor - an option for the TV? TV  and Kinotechnische Zeitschrift No. 5/1996, pp. 242-248; DE 39 21 061 A1]. Here will the exact head position and movement are detected, and the associated images are generated at the same time. In addition, there are also elaborate VR systems (virtual Reality) using "Head Mounted Displays" which takes getting used to.

In medical technology, magnetic resonance and computed tomography are the most important Areas of application for stereoscopic 3D visualizations. About certain perspectives you are looking for powerful special computers are used to calculate. Combined computer visi Visualizations and real-time transmission of endoscopes are becoming one of the most important most neurosurgical tools. Stereoscopes Endoscopes are already in use. A electronic motion control via an infrared head Tracking Sensor "will be easy to combine with a personal car stereoscopic screen system.

In addition to the shutter monitors, autostereoscopic adaptive monitors are known (DE 195 00 315 C1), two perspectives, one for the respective observer position right and left, separated on the monitor display for the right and left eye able to represent. However, this requires recalculation to be carried out in real time of object perspectives when the user is moving.

Devices for the display of autostereoscopic images are also in the form of Stereo hologram displays have become known (PCT WO 98/21 619), the several electronic cameras, a computer, an exposure device, or a printer Grid glass and a black and white raster pixel that can be mounted behind it Contain foil, with a large number of z. B. 128 perspectives is written, whereby the possibility is indicated, from two or more existing perspectives (through interpolation) to further perspectives win. However, it is tacitly assumed that the admitted Objects have known depth positions and do not overlap when shooting. How to deal with overlaps can neither this named proposal nor the further, relevant for methods of the type described in the introduction Technology can be taken (e.g. US 5 696 596, US 5 697 006, US 5 455 689, JP-37 813 B2). Such a natural three-dimensional representation of objects  no longer needs a head tracking system and can be used by several people at the same time can be used for autostereoscopic display.

The invention has for its object a method for representing autostereo to propose scopic images with the simple technical and inexpensive Mean an object, especially a portrait of a person, with a variety of Perspectives can be displayed true to color and depth. The three-dimensional Information of the image should be as shown in PCT WO 98/21 619 and optionally on one black and white, high-precision photo film layer, on a conventional film material or can be saved on a normal slide.

The characterizing features of claim 1 serve to achieve this object.

The invention extends the known method (PCT WO 98/21 619) An inexpensive black and white also for a large display (e.g. 64 cm × 48 cm) Film can be used. To simplify the three-dimensional preparation and on A graphic computer holds prepared three-dimensional to concentrate few objects Environments or backgrounds available for selection. A person to be portrayed photographed by an electronic camera system. The data becomes one Graphics computer directed this information to the specific 3D black and white format converted and fed directly to an exposure machine. After a few Minutes the picture can be viewed and, if necessary, changed and recorded being represented. The method preferably works with a background lighting tion, but also normal diffuse radiation from the front is possible.

For the computer-aided depth-accurate calculation of an image one needs before one Object up to over a hundred different views from different perspectives NEN. The simultaneous recording of such a large number of images represents a technical one Problem. The three-dimensional recording of a still environment can by a slowly and steadily moving camera should be made every few Millimeter takes a picture. The sequential fails with moving objects Recording. When photographing one or more people People have to start from movements, so that one is on the simultaneous recording  different perspectives can not do without. This problem will According to the invention preferably solved in that a small number of electronics cameras, for example two, four or eight, side by side in uniform Intervals are arranged and triggered synchronously. One person or one move The gender object is set against a white or the same colored background shown. From these few images, a graphics computer then calculates one quick correlation analysis all other necessary intermediate views. Then with Depth positions marked three-dimensional image pixels of an object can in a resting environment selected as desired can be mixed according to position.

The invention enables a true three-dimensional representation of a spatial Object. The 3D images themselves can be found on commercially available, inexpensive black white films with sizes larger than DIN A3 can be accommodated. The in the graphics compu The stored three-dimensional background images and objects carry all objects and can be positioned differently in a composite image. It can also be a group picture of many people in front of a known object be put together. Before you expose a film, you can have one Consider the scenario on a person-adaptive 3D monitor and changes in the Make representation.

For promotional purposes, objects with certain movement effects can also be recorded For example: a face or an arm can change when passing by lift, or a window may open from a certain direction, etc. It is quite conceivable that a very large picture can consist of several small ones Lets you assemble drawing files.

Even people who are denied a normal stereo view, for example because of one eye Seeing too weakly can perceive the 3D effect by changing their position vary slightly horizontally and recognize the different perspectives here.

The main area of application for the method according to the invention is 3D Photo studios using a system of electronic cameras, a graphics computer, a person-adaptive 3D monitor and equipped with a black and white exposure device  are. Of course, this technical equipment can also be used to produce inexpensive, large, two-dimensional color images. Instead of a lentil In this case, the window could use cheaper glass with a color mask become.

Another interesting area of application is radiology in medical technology: two-dimensional x-ray images that are difficult to interpret can be determined by three-dimensional le to be replaced. Every doctor who interprets X-rays today only needs additional a lenticular screen with a color mask when he uses a 3D X-ray station. The same applies to magnetic resonance imaging.

If you put several drawing files together to form a large picture, then in architecture opened up a particularly interesting area of application: instead of a model, a 3D large screen can be presented.

Further advantageous features of the invention emerge from the subclaims.

The invention will be described in connection with the accompanying drawings Embodiments explained in more detail. Show it:

Figure 1 schematically shows a 3D portrait with electronic cameras, a graphics computer and an exposure device and the subsequent playback.

Fig. 2 shows a stereo color hologram viewer for 3D photos, e.g. B. for nuclear spin and X-ray tomography;

Fig. 3 is a representation of object positions in camera images;

Fig. 4 is the representation of object positions in images of cameras arranged in a circle;

Fig. 5 is the representation of the different perspectives in a hologram image;

Fig. 6 is the representation of the interpolation of missing perspectives;

Fig. 7 is the representation of a relative coordinate system for the camera and for the viewer; and

Fig. 8 is a cylindrical lens grid in an exploded view together with a 3D photo that can be placed in the correct position.

Detailed description of the 3D portrait process A) Precise assembly and description of the film

To get a larger 3D image, e.g. B. 60 cm × 40 cm, inexpensive to keep it is important to use standard material for the black and white film can use. The problem arises that there is no additional dimension was impossible to record the film µm precisely on such a large area to apply the color and brightness masks. For example, Filmma material on a glass of this size and then exposed and be developed so that it fits exactly onto the front glass pane the lenticular grid and the sputtered color and brightness mask can be applied. That when using standard foils assembly problem occurring is also in the basic application "HOLDISP" unsolved. Therefore, a supplementary proposal will be given below wrote how to get standard footage, e.g. B. also for X-rays are used.  

Fig. 8 illustrates the necessary steps. It can be assumed that a cylindrical lens grid is available with vertical cylindrical lenses ( 1 ) at a pitch distance of approx. 0.5 mm of the glass pane to which the lenses are applied ( 2 ) and with a color and brightness mask sputtered on with a micrometer. The color and brightness mask consists of horizontal, repeating strips of black, red, green, blue. Each individual color strip has a width of approx. 100 µm. Each red, green or blue color strip is subdivided into three further strips with a width of 25 µm, 50 µm and 25 µm. The middle, 50 µm wide strip contains no further brightness filtering to absorb the light, while the strips ( 5 ) and ( 6 ) are additionally blackened so that light is absorbed. The strip ( 5 ) absorbs 90% and the strip ( 6 ) 98%. The applied film ( 8 ) contains the actual image information with small square black or transparent areas in the size of 5 µm × 5 µm. This means that there is room for 20 black and white sub-strips on a 100 µm strip. The film ( 8 ) to be applied is now exposed in such a way that during the transitions from, for example, red to green or from more absorbent to less absorbent strips, a strip which remains black is applied symmetrically to the transitions on the film, while the other pixels are free for coding are. On the strips ( 5 ) and ( 6 ) 4 µ pixels can then be used one above the other for coding, while on the wide middle strip of 50 µm 9 superposed µ pixels can be used for the black and white coding. In contrast to the original method in "HOLDISP", a wide strip of 50 µm is used here, in which 9 sub-pixels can be used next to each other for coding. This has the advantage that less than 50% of the light is lost overall in the brightest image information, ie it is absorbed by the brightness filtering. In the middle strip, which does not absorb any brightness, the following brightness values can be coded 0, 25, 50, 75, 100, 125, 150, 175, 200, 225. In the strip ( 5 ) the following 5 brightness values can then be encoded first additively superimposed, are coded: 0, 5, 10, 15, 20. And finally in strip ( 6 ) the values 0, 1, 2, 3, 4, which can be superimposed additively. This means that a total of 250 different brightness levels can be coded.

The film to be applied may now be due to the covered transitions in have an offset of maximum plus / minus 2.5 µm in the vertical direction, without changing the quality of the image.

The black stripe provided in the hologram display (HOLDISP) enables the film to be positioned with µm accuracy in relation to the pane. A glass pane is now used which has small positioning pins ( 4 ) with a diameter of approximately 50 μm and a projection in the thickness of the film material of approximately 20 μm on the black strips provided. Before the exposure or writing, these positioning holes are also punched into the film, or film, to the exact micrometer. The imagesetter or printer must now also have these positioning pins at the same intervals, so that the film can be accurately written over a large area.

The positioning pins ( 4 ) and ( 11 ) are conical as shown in section C, so that the film material can be pressed more easily into the appropriate position. It could now happen that the printed film in the color stripes RGB between the glass and film remains air bubbles and is not in other places. And in this way Newton rings could be created later. Therefore, the proposed glass ( 2 ) is designed so that in the areas of the color strips RGB, ie in a width of about 300 microns, the glass has a recess of about 3 to 5 microns, so that between glass and film in the In the area of the color stripes an air cushion ( 10 ) is created over which the film is stretched.

There are several ways to code the brightness values in the 3 sub-strips of the color strip. If one uses an unfavorable coding, e.g. B. on top of each other pixels transparent / black, transparent / black etc., this can lead to disturbing interferences when using coherent light. Therefore, a coding is selected in the individual brightness sub-strips such that, as the brightness increases, the transparent sub-pixels are arranged one above the other so that the lowest brightness level in the middle of a strip is switched to transparent, and as the brightness increases around them directly above and then underneath switched transparent pixels are arranged so that there is no change between transparent and black. As shown in FIG. 8B, this leads to a wedge in the coding mask when the brightness falls in the horizontal direction.

These positioning measures described here result in a three-dimensional image of high quality which offers 50 different stereo images with full photo quality for an observer in the intended spacing position. The prerequisite for this is that this image is evenly illuminated from behind, although one can of course also draw on experience with TFT flat screens.

B) DESCRIPTION OF THE PORTRAITING PROCEDURE

Determination of missing intermediate perspectives from two existing perspectives ven: A 3D image described here is normally made up of 128 individual images perspectives together. This number then corresponds to the number adjustable depth positions. This would have to be synchronized for moving objects 128 cameras can be recorded. With simple objects one could get along with less recorded perspectives when intermediate prospects can be calculated from the existing ones. However, this leads to a drastic reduction to just 4 or 8 recorded perspectives on significant problems, as usually undefined hidden regions absence. Also the reconstruction of depth positions from two or more  Images from different directions are correlated in most Cases not clearly possible when there are similar textures or small ones Repeat shapes. How to deal with this problem with 3D portrait pictures detailed background resolves, is described below.

The following individual measures ensure that a 3D portrait in detailed environment almost completely from a few 2-dimensional Images can be reconstructed from certain different directions can.

• 1. The object to be portrayed is in a precisely predefined limited space positioned.
• 2. Two, four or more cameras are placed exactly reproducible around the object to be portrayed against a known, previously recorded background or against a background of the same color (blue) with a calibration pattern, cf. Fig. 3 and Fig. 4.
• 3. Several objects to be portrayed in one picture are taken one after the other recorded.
• 4. The pixels on the recorded two-dimensional perspective images are made using a priori information about rough form and Spatial position of an object the most accurate depth positions possible individual pixels. The depth position values become the RGB Values saved as another word for the individual pixels of the file.
• 5. The perspectives missing between two camera positions are made from the two marginal images using the depth position indicator interpolated and in the corresponding file locations entered or added. Perhaps different color values of the same object point from different  Directions are interpolated linearly for the intermediate values.
• 6. A detailed three-dimensional image of the surroundings is either by Computer generated, with every pixel of every perspective using the depth posi tion is marked, or a natural environment is marked with a moving camera from the corresponding perspective one after the other recorded so that 128 or more perspectives are available which depth information can be determined for all pixels. In doing so depth positions not clearly identified and marked by a Special procedure specified later.
• 7. The 3D images of an object and its surroundings, consisting of 128 per spotting scopes in which each pixel is marked with a depth can now be superimposed true to the view by at each pixel the one with the lower depth position is selected. With partially trans Parent objects that must be labeled accordingly is that rear with a certain weight to be considered.
• 8.Should several objects or positions of 3D images be true to depth certain spatial positions can be represented on a picture, so the spatial positions of the individual objects shifted with a 3D mouse displayed on a monitor and saved in a fixed manner.
• 9. If an object is to be recorded all around, the partial circle shown in Figure 4 is continued by cameras to form a circle. All cameras then take two pictures, one with an object in the middle and one without an object in the middle. This means that the surroundings that do not belong to the object can later be removed from the image.
• 10. The depth position of a pixel in space becomes two neighboring ones Perspectives by minimizing an environmental norm when horizontal Moving from pixel to pixel found. The standard consists of the Sum of the absolute amounts of differences of pixels from the one image  with neighboring perspective pixels by a certain amount are shifted horizontally. Where the area around the point of one Image best coincides with the surroundings of the point from the other ren picture, there is the horizontal shift, which then corresponds with the depth position. Because the rough position of the object is known minimization can be started with an expected shift become. Variable weights are also taken into account when minimizing te environmental points.

Calculation of the depth positions from two images

It is assumed that a certain object point on two images of neighboring cameras has been clearly identified as a pixel. If the camera positions including their directions are then exactly known, the depth position of this point can be calculated from the horizontal offset of these pixels on the two images. The depth position is determined geometrically below. For this we use a relative xyz coordinate system and a local uv coordinate system as shown in Fig. 7. The relative coordinates are such that the origin lies in the point where the rays from the object points from this environment cross. The uv coordinates lie on the image plane and are oriented in negative xy directions. The uv image plane is parallel to the xy plane at z = Z _K. A visible object point with the coordinates (X ₁ , Y ₁ , Z ₁ ) then appears on the image at the location

(U ₁ , V ₁ ) = (X ₁ , Y ₁ ) .Z _K / Z ₁ (1)

Now two cameras K ₁ , K _{2 are} given with the local coordinates u1, v1 and u2, v2, whose beam crossing points on the x-axis at the locations X _K1 = -D _K / 2 and X _K2 = + D _K / 2 lie. The image planes may again be arranged parallel to the xy plane, at a distance Z _K. Then the identified pixel appears on the images at the points

(U1, V1) = (X1 + -D _K / 2, Y ₁ ) .A _K / Z ₁ (2) or

(U2, V2) = (X1 - -D _K / 2, Y ₁ ) .A _K / Z ₁ (3).

Z _K = -A _{K is} the distance of the image plane from the crossing point. In practice, it must be taken into account later that the resolution of the images is only finite, e.g. B. is 1280 or 2560 pixels horizontally. It must therefore be taken into account that each discrete value, such as U1, U2, must be provided with certain tolerances due to the quantization.

Since the parameter values for the settings and setups of the cameras are known, the difference for the above equations results in the specification for the depth position.

Z ₁ = D _K .A _K / (U1 - U2) (4)

In order to achieve better utilization of the film in the camera, it is advantageous to arrange the cameras around the center of the intended object center, as shown in FIG. 4. In this case one would prefer polar coordinates, for which the following relationship applies:

x = r.cos (ϕ) .sin (δ), 0 <ϕ <2π

y = r.sin (ϕ) .sin (δ), 0 <δ <+ π

z = r.cos (δ) (5)

The inverse transformation is then

r = √ (x ² + y ² + z ² ) (6)

ϕ = arctan (y / x) (7)

δ = arctan [z / √ (x ² + y ² )] (8)

The circular cameras are aimed at the object center with the coordinates

M _O = (x _o , y _o , z _o ) = (0, 0, R _O ) (9)

The cameras are all at a distance R _O ; from the object center O _Z and from this point, viewed in the negative z direction, positioned at the angles Ψ _i , whereby two neighboring ones may include an angle difference Δ _K = Ψ _i - Ψ _i-1 (cf. FIG. 4).

The local coordinates of the i-th camera, which is aligned with the object center O _Z , are denoted by ui, vi. In the relative coordinate systems of the i-th camera K _i , the identified object point can be identified with the coordinates Ui ₀ , Vi ₀ associated relative solid angle can be calculated.

With regard to the i.te camera, the following solid angle results

ϕi ₀ = arctan (Vi ₀ / Ui ₀ ) (10)

δi ₀ = arctan [√ (Ui ₀ ² + Vi ₀ ² ) / A _K ] (11)

The projection of the solid angle on the plane y = 0 provides the angle to the z axis

δi _0z = arctan [Ui ₀ / A _K ] (12)

The intersection of the projections of two solid angles, viewed from two neighboring cameras K _i , K _i-1 , then leads to the depth position of the identified object point.

The point of intersection of the i-th camera in the xz plane is then given in the original xyz coordinate system by the following coordinates

Xi _K = R _OK .sin (Ψ _i ) (12)

Zi _K = R _OK . (1 - cos (Ψ _i )) (13).

The i th projected straight line through the associated camera crossing point then has the following equation in the xz plane:

x - Xi _K = (z - Zi _K ) .tan (δi _0z .Ψi _K ) (14):

The difference from two such adjacent equations provides the intersection z = Z ₀ , which at the same time indicates the depth position Z _{0 of} the identified object point.

Xi _K - XI _K - Zi _K .tan (δi _0z .Ψi _K ) + ZI _K .tan (δI _0z .ΨI _K ) = Z _O. [Tan (δi _0z .Ψi _K ) - tan (δI _0z .ΨI _K )]

for I = i - 1.

The division of the equation by the factor at Z _O on the right side of this equation leads to the depth position Z _{O of} the identified point O. If the same object point can be identified in all perspective images, the same depth positions must result from the neighboring perspectives - except for a permissible quantization uncertainty. As a check, the depth position of an identified object point should not only be determined from the two perspectives i and i - 1, but also from more perspectives such as I = i + 1, I = i - 2u, m. determine the deep positions of points in order to determine valid mean values from them if necessary. Those object points that could be clearly identified in at least two different perspective images can thus be provided with precise depth positions. This also applies to an arrangement in which the cameras are arranged around the entire object.

Has an object region of uniform brightness in a certain environment and color, points of this region cannot appear on neighboring images be clearly identified. In these cases, an entire environment can be used to find the depth position. This will follow the explained in more detail. The size of an environment to be considered be adjustable by means of a parameter a. For weighting the reverse an exponential function is used, just like for a optical filtering applies.

The color values on the pixels of the images from the cameras are not stored continuously, but are quantized and digitized via the local coordinates u, v with numbers of pixels in the horizontal direction usually from 1280 and in the vertical direction from 1024. That means the relative horizontal coordinate ui of the i-th camera has integer values between -640 and +629 and the vertical coordinate vi is between -512 and +511. The horizontal coordinate points may be characterized by the integer index values m and the vertical coordinates v, by the index values n, m then denotes column positions of a matrix and n row positions. If the horizontal coordinates vary between -u _max and + u _max and the vertical coordinates between -v _max and + v _max , then the quantization can be expressed by the integer function.

m = int (640.u / u _max ) (16) and

n = int (612.v / v _max ) (17).

The colors R, G, B are also quantized in 256 steps, i. H. she have integer values between 0 and 255. A quantized color value in the i-th camera may be called Fi. To the colors R, G, B on to take into account the pixels n, m, the pixel values are marked by the function Fi (f, m, n). Where f = 0 for R (red), f = 1 for G (green)  and f = 2 for B (blue).

As a rule, it will be sufficient to take into account an environment of ± 10 points in both directions. Then the identification function between two images of adjacent cameras is i and I.

The integer λ value for which this sum is a minimum delivers the difference in the u-coordinate, which then leads to the depth position. The The area that λ passes through is limited by the portrayal room determined. The above difference formation of different pixels is for small values of n correct and would have to be minor for larger n values Getting corrected. Different points in front of and behind the object center Chen images can fall on different v-coordinates. The second Term in the difference is then due to an integer index value ρ complete. This value ρ depends on n, i.e. H. the height in the picture, and the Camera difference i - I: ρ = ρ (n, i - I). On a precise calculation of the belonging integer values can be omitted here, because it it is advantageous to store this value in a table and, if necessary, to call. The term is FI (f, m - µ - λ, n - ν - ρ).

The essential image information from which the depth positions are reconstructed are located in the edges of figures, i.e. H. in color and light changes in the image. Therefore it may be advantageous to take the pictures before Filter determination of depth positions. Such a filter operator, which is not would filter the required image content away, would be an edge filter, the vertical edges raises and even brightness regions to a constant value  puts. This can be achieved that the above-mentioned is more pronounced minimum λ. For a more detailed description edge filtering need not be dealt with here, since it is one of the Standard methods used in digital image processing.

The background image, which later blends in depth with a portrait image may contain numerous details at different depths u. U. from neighboring images can not be clearly reconstructed. But you can assume that in this case all perspectives from one camera that have moved horizontally and in a straight line. It's brewing then no intermediate perspectives can be reconstructed. Yet the depth positions must be determined and added for all pixels, so that it can be mixed with other images in depth. There but there is a whole sequence of neighboring images, it is possible Detect a movement of contours across multiple perspectives animals. From the determined over a certain number of neighboring images Speed can then go directly to the depth position of this contour getting closed. There is still a region left in which there are none There are contours and consistent colors appear and in the too no movement is recognizable, one must also make the assumption that a surface that is surrounded by a moving contour, itself moved just as quickly as the contour, d. H. themselves in the same depth posi tion.

In this way it becomes possible that every RGB pixel in every person spotting can be marked with a depth. Darüberhin from it will be possible via software tools, also depth positions from Change objects. If one cuts through the dreidi thus captured dimensional landscape, this way it can be tested whether the results nisse agree with the desired or actual.  

Interpolation of intermediate perspectives

Object points or object regions that could be clearly identified in the depth position from two camera positions can be represented on intermediate perspectives by the interpolation described below. As shown in FIG. 6, a point of the object is clearly identified on the two cameras K1 and K2, whose local coordinates u1 and v1 or u2 and v2 span a plane that is parallel to the image plane. In addition, it is assumed without restriction of generality that the grid in the coordinates u1 or u2 corresponds to the grid or the pitch distance of the lenticular screen. So all 1280 pixels can be found on the local coordinates u1 and u2. If the plane spanned by u1 and u2 is not parallel to the plane of the lenticular disk, a linear aspect ratio must be taken into account. It is now assumed that an image of perspective 1 with the correct screening is recorded in camera 1 . The same is assumed for the perspective 6 of the object, which is recorded in the correct grid in the camera 2 . Then the RGB values including the determined depth positions can be entered in perspectives 1 and 6, which generally contain 1280 pixels horizontally and 128 perspectives per pixel. The objects in space are now to be mapped onto the image plane, which is located in the focal point of the cylindrical lenses and is located on the plane z = Z _B. If an identified object point O1 is known in the perspectives P1 and P6 of the cameras K1 and K2 with the associated depth position and the color values, then the screening shown in the plane z = 0 can be carried out for the perspectives P2, P3, P4 and P5. Rays can be drawn to the object point u1 and the penetration points through the image plane Z _B can be calculated precisely. The pitch grid created on the image plane Z _B denotes the affiliation to certain pixels. It can now happen that the rays from neighboring perspectives to the object point fall into the same pixel column. Then these must also be entered in the associated perspectives of the same pixel column. If there are already values with the same depth position at this position, an average is formed from this. If an object point with a larger depth position is already entered at one point, it will be overwritten. If an object point is entered with a smaller depth position, no entry is made. If the same identified object point O1 in camera 1 has a slightly different color value than in camera 2 , these color values are linearly interpolated from perspective 1 to 6 . If the grid is fine enough for the perspectives and the image objects are in the correct position, no pixel is left out in the intermediate positions when going through the different perspectives. However, if there is an object point in front of the pane, it can happen that the rays pass through certain pixels and some are omitted in between. In this case it is necessary that the omitted pixel points get entries with the same values, the same depth positions and the same perspectives as they are entered in the directly adjacent pixel columns. If all identified object points have now been run through, 128 different RGB values with depth positions are entered in all pixels of the image plane (1280 × 1024).

Just as the horizontal and vertical coordinates of the image are quantized it also makes sense to quantize the depth positions. You choose For example, a 12 bit accurate for marking the depth information Quantization, depth positions between 0 and 4095; where 0 is the position of the camera or the one to be viewed later Eye and the largest position the selected maximum depth position, for example 10 m. When entering the object points in the image plane, d. H. in the different perspectives of all pixels one starts from one Initial entry, with the greatest possible in every perspective of every pixel che depth position is entered, that of any other occurring Object of a lower depth position can be overwritten. On this way you get all the desired entries for a portrait picture  in the perspectives. A full-length portrait is provided hen, one must differentiate between different levels of representation convert the and non-parallel image planes and camera planes.

On extensions of this process, such as for partially trans Parent objects occurs, should not be discussed here. It would then require that at a pixel with the same perspective several entries and color value positions would have to be provided.

A system with a fixed distance of two cameras has the advantage that you can easily can record moving scenes, in which always the two Perspectives are recorded at the same time, which the eyes later too see. The closely spaced perspectives can then be under belong to different time slots, but see the eyes of the viewer time-synchronous recordings. With slight lateral movements of the Be Trachters can also detect slight movements of an object be taken. For example, a portrait, the picture is from the left considered, have a serious facial expression and when looking at a friendly one from the right. Also random movements like with for example, winking will not be a nuisance. However, it would disrupt when the viewer's left eye from a certain position a portrait's closed eye would see and the right one open. Of course here is the correct valid distance of the viewer from to adhere to the surface of the raster image.

Claims (18)

1. A method for displaying an autostereoscopic image using a plurality of electronic cameras (A), a computer (B), an exposure device or printer (C), a scanning glass ( 1 ) and a black and white mountable behind the scanning glass ( 1 ), Film (D) having raster pixels, a large number of perspectives being written side by side in each raster pixel, characterized in that the perspectives of a single object (O ₁ , O ₂ ) of the image to be displayed are recorded separately with the cameras and then these perspectives and corresponding but also separately recorded perspectives of a further object and / or separately recorded or generated perspectives of a background image are combined to form an autostereoscopic image by 3D-appropriate superimposition in the computer (B). 2. The method according to claim 1, characterized in that each perspective Depth position is assigned and in the 3D-appropriate overlay of perspectives with different depth positions those perspectives for producing the autostereoscopic image can be used, showing the lower depth positions assigned. 3. The method according to claim 2, characterized in that the perspectives too Transparency features are assigned and in the composite autostereoskopi per picture behind the perspectives with the lowest depth positions Spectives with a color component are taken into account. 4. The method according to claim 2 or 3, characterized in that the depth positions can be calculated from changes in object contours of neighboring perspectives.   5. The method according to any one of claims 1 to 4, characterized in that the Representation of the autostereoscopic image on an adaptive 3D monitor separately recorded, three-dimensional object or a background and copied with a 3D cursor is positioned as desired in this. 6. The method according to any one of claims 1 to 5, characterized in that the object (O1, O2) is positioned in the recording in a precisely defined, known space ( 31 , 32 ). 7. The method according to any one of claims 1 to 6, characterized in that for Production of the image (6) uses far fewer cameras (K1 to K5) than corresponds to the multitude of perspectives, and that the multitude of perspectives through Displacement of the cameras (K1 to K5) relative to the object (O1, O2) is obtained. 8. The method according to any one of claims 1 to 5, characterized in that the various cameras (K1 to K5) around the object to be photographed (O1, O2) be distributed in a circle. 9. The method according to any one of claims 7 or 8, characterized in that the Cameras (K1 to K5) are moved in sections and that after each movement section with all cameras (K1 to K5) one image of the object at a time (O1, O2) is made. 10. The method according to any one of claims 7 to 9, characterized in that the Cameras (K1 to K5) to record intermediate perspectives in order to record this Rotate the object (O1, O2). 11. The method according to any one of claims 7 to 10, characterized in that exactly two cameras (K1, K2) are used which have a certain horizontal distance (Δ _k ) such that the two recorded perspectives correspond to those that a viewer with the right and left eye when viewing the autostereoscopic image at standard distance. 12. The method according to any one of claims 1 to 6, characterized in that much less stationary cameras (K1 to K5) are used to produce the autostereoscopic image than corresponds to the plurality of perspectives, and that the plurality of perspectives is obtained in that Perspectives that cannot be delivered by the cameras (K1 to K5) are generated by interpolation of the perspectives supplied by the cameras (K1 to K5) with the graphics computer ( 4 ). 13. The method according to claim 11 or 12, characterized in that the cameras (K1 to K5) are arranged on a circular path around the object (O1, O2) and the intermediate perspectives by interpolation in the graphics computer ( 4 ) using a minimization algorithm for cameras arranged in a circle (K1 to K5). 14. The method according to any one of claims 1 to 13, characterized in that the raster pixels contain information about color and / or brightness arranged one above the other in digital form, this information depending on a color and / or brightness mask used when viewing the autostereoscopic image ( 2 , 3 ) are selected. 15. The method according to any one of claims 1 to 14, characterized in that each raster pixel of the image in a single color and / or brightness mask ( 2 , 3 ) associated area has a central section with nine superimposed, black and white codable pixels and each has a section above and below it, each with four superimposed, black and white codable pixels. 16. The method according to claim 15, characterized in that the autostereoscopic Image in the area of transitions between the sections, each of which is not codable, has black pixels intended to compensate for assembly tolerances. 17. The method according to any one of claims 15 or 16, characterized in that the Pixels in the sections are encoded so that the white encodable pixels each in the middle of a section and come to lie directly one above the other.   18. The method according to any one of claims 1 to 17, characterized in that the image carrier is provided with specific for µm-accurate positioning, with corresponding adjustment pin ( 11 ) in the recording and / or viewing device interacting Justierlö holes.