DE102019006753A1 - Method and device for transmitting images in a vehicle - Google Patents

Abstract

The invention relates to a method and a device for transmitting images (I, I ') from an image source (S) to a display module (D) in a vehicle, an original image (I) being provided by the image source (S), which is transformed into a processed image (I ') and transmitted to the display module (D). At least one position gradient image is determined for each image (I, I ') and metadata (M, M') comprising at least one gradient histogram (H, N, H (i, j), H (i, j)) is determined therefrom. Using at least one difference measure (d, d, d (i, j), d (i, j)), which shows the difference between the source metadata (M) of the original image (I) and the control metadata (M ') of the processed image (I '), the perception equivalence of the images (I, I') is determined by comparison with a predetermined threshold value.

Description

The invention relates to a method according to the preamble of claim 1 and a device according to the preamble of claim 3.

Methods and devices for transmitting images, for example environmental images recorded by means of a camera for display on a display, are known. The processing of images along the transmission chain, for example the compression, a size adjustment and / or the superimposition (augmentation) of symbols over a surrounding image, are also known.

The document DE 39 13 620 A1 describes a method for image evaluation by means of an evaluation device by analyzing the directions of gray value gradients, with stored image patterns, e.g. B. textile samples or vehicles are determined in the image to be evaluated, the gray value gradients of the image being normalized, the histogram of the gradient directions being formed by the frequency of the gradient angle values in percent against the angle value, e.g. B. in a table, the cross-correlation is carried out with the histogram of the stored image pattern, the angular value of the image pattern is increased by 1 or any other value (modulo 360) and this new histogram is cross-correlated with the image, this measure 360 times or is repeated correspondingly less often and the image with the highest correlation of the cross correlation is determined.

The document DE 195 49 216 A1 describes a method for image evaluation of a parceled or tessellated image, the mean gray scale value L _m and the minimum and maximum gray scale values L _min and L _{max being} calculated for each parcel of a parceled picture. Their difference is called the span S _{p of} a parcel. Then an average span S _m and its standard deviation for the entire gray value image are calculated from the spans of all plots of the parceled image. This mean value S _m is used as a threshold value for defining conspicuous parcels: A parcel is conspicuous if its span S _{p is} larger than S _m . In addition, the distances between maxima in the histogram of the gray value gradient directions are used to define conspicuous areas. Distances of approx. 180 ° or 90 ° signal parallel edges or orthogonal edges. The "span method" is combined with the "direction histogram method" for the detection of artificial objects (e.g. vehicles, houses, streets).

The invention is based on the object of specifying an improved method and an improved device for transmitting images in a vehicle.

With regard to the method, the object is achieved by the features of claim 1. With regard to the device, the object is achieved by the features of claim 3.

Advantageous embodiments of the invention are the subject of the dependent claims.

In a method for transmitting images from an image source to a display module in a vehicle, an original image is provided by the image source, transformed into a processed image and transmitted to the display module.

According to the invention, at least one spatial gradient image is determined for each image. From the at least one location gradient image, metadata are in each case comprising (for each location gradient image) at least one gradient histogram. Using at least one difference measure, which measures the difference between the source metadata of the original image and the control metadata of the processed image, the perceptual equivalence of the original and the corresponding processed image is determined by comparing the at least one difference measure with a predetermined threshold value.

A perception equivalence between two images is present when they are perceived in the same way by an observer, in particular when an observer can extract the same relevant information, in particular essential for the control of the vehicle, from both images.

An advantage of the method is that an original and a processed image, although they are not mathematically or in terms of value identical, are recognized as perceptually equivalent if they convey essential information that is the same for the viewer. In contrast, two images can also be recognized as not perceptually equivalent if the processed image transports information that has been changed or missing in essential parts compared to the original image. A particular advantage of the method is that it can be implemented efficiently, in real time, with little or no additional hardware requirements and also for high-resolution image sequences or videos. In particular, available hardware resources can be used to implement the method. Another advantage is that the metadata required to determine perception equivalence are small compared to the image data.

Exemplary embodiments of the invention are explained in more detail below with reference to drawings.

• 1 schematisch eine Übertragungsstrecke zur gesicherten Übertragung von Bilddaten,
• 2 schematisch den Ablauf eines Verfahrens zur globalen Gradientenhistogrammbestimmung,
• 3 schematisch den Ablauf eines Verfahrens zur lokalen Gradientenhistogrammbestimmung sowie
• 4 schematisch den Ablauf eines Verfahrens zur Validierung einer Bilddatenübertragung.

Show:

• 1 schematically a transmission path for the secure transmission of image data,
• 2nd schematically the sequence of a method for global gradient histogram determination,
• 3rd schematically the sequence of a method for local gradient histogram determination and
• 4th schematically the sequence of a method for validating an image data transmission.

Corresponding parts are provided with the same reference symbols in all figures.

1 shows schematically a transmission path for secure transmission of images I. , I ' from an image source S up to a display module D , the image source S original pictures I. provides and the display module D processed images I ' represents.

The image source S can be designed as a photo or video camera. But it is also possible that the image source S is designed as a graphics processor, the original images I. generated electronically. Even image sources S in which natural images of the surroundings recorded with a camera are overlaid with electronically generated image data are possible. The transmission link is preferably for the transmission of images I. , I ' set up in a vehicle. It is also possible that the image source S a chronological sequence of original images I. as a video stream.

The original pictures I. are from a frontend FE recorded and sent to a head unit HU forwarded. The frontend FE is set up from the provided original images I. Source metadata M to extract the content of each original image I. characterize. Methods for extracting such source metadata are described below using the 2nd and 3rd described in more detail.

The frontend FE is set up for computer-aided processing of such a procedure. For example, the front end includes FE a microcontroller, a field programmable gate array (FPGA) or a digital signal processor (DSP). The frontend FE can also be part of an image source S , for example as part of a graphics processor. This can result in additional effort and costs for the extraction of the source metadata M avoided or reduced.

A head unit HU is a well-known and widespread data processing unit for use in automobiles, which can implement different functions. It is essential for the present invention that the head unit HU a fed original image I. into a processed image I ' transforms and outputs, with fed source metadata M output unchanged. For example, a head unit HU be set up for overlaying (augmentation) of additional information, for example for overlaying symbols over an image of the vehicle surroundings.

From the head unit HU becomes the processed image (for example provided with augmented symbols) I ' together with the unchanged source metadata M to a backend BE headed. The backend BE is set up from the processed images I ' corresponding control metadata M ' to extract and use the ones from the original image I. extracted source metadata M to compare, taking to extract the control metadata M ' the same procedure is used as for the extraction of the source metadata M . Such methods for extracting metadata M , M ' are subsequently based on the 2nd and 3rd described in more detail. A method of comparing source metadata M with the control metadata M 'is subsequently based on the 4th described in more detail.

The backend BE is set up for computer-aided processing of such a procedure. For example, the backend includes BE a microcontroller, a field programmable gate array (FPGA) or a digital signal processor (DSP). The backend BE can also partially or completely as part of the display module D , for example as part of the timing controller (TCON) of the display module D , be trained. This can result in additional efforts and costs for the extraction of the control metadata M 'and for the comparison of the metadata M , M ' avoided or reduced.

As a result of the comparison of the source and control metadata M , M ' gives the backend BE a flag or an error status F and transmits it together with the processed image I ' to the display module D .

The error status F can be binary coded and take the value (logically) true, when comparing the source and control metadata M , M ' results in a sufficiently high match and accordingly incorrectly assume the value (logically) when comparing the source and control metadata M , M ' there is an insufficient match. Alternatively, the error status F also specify a continuous measure of probability, where a probability value of 1 is a sure match of the image contents of the original image I. and the processed image I ' corresponds and a probability value of 0 a sure discrepancy between the image contents of the original image I. and the processed image I ' corresponds.

The display module D shows the processed image I ' on. Depending on the determined error status F additional information, for example warnings about a possibly limited reliability of the displayed image content, can be displayed. It is also possible to display a based on the determined error status F untrusted processed image I ' on the display module D to block.

2nd shows an embodiment of a method for global gradient histogram determination 2nd comprehensive steps 21 to 24th , with the metadata M , M ' from pictures I. , I ' be extracted.

In a first step 21 pixel-by-pixel image data from an image I. , I ' taken. A pixel of a color image taken at the vertical image position i and the horizontal image position j is described, for example, in the RGB color space by a red value R (i, j) , a green value G (i, j) and a blue value B (i, j) .

In a second step 22 will be at the pixel position for each pixel (i, j) from the color values R (i, j) , G (i, j) , B (i, j ') a corresponding gray value GV (i, j) determined, for example according to the transformation equation $$GV\left(i,j\right)=0.3\cdot R\left(i,j\right)+0.59\cdot G\left(i,j\right)+0.11\cdot B\left(i,j\right).$$

Transformation equations for the transfer of color values to gray values are also known for other color spaces, for example the CIE (Commission Internationale de l'eclairage) XYZ color space or the CIE Lab color space.

One advantage of converting a color image into a gray value image is that the information that is crucial for the perception of the image content is contained in a gray value image GV (i, j) is easier to differentiate from interference. Therefore, to examine the question of whether image content is perceptibly disturbed, the comparison of metadata is necessary M , M ' based on gray scale images GV (i, j) were found to be easier and more robust than a comparison of metadata M , M ' based on color images I. , I ' .

entlang mindestens einer Ortsrichtung r ermittelt oder geschätzt. Dadurch können im Wesentlichen homogene Intensitätsunterschiede zwischen einem ursprünglichen Bild I und einem verarbeiteten Bild I', die die Wahrnehmung des Bildinhalts im Allgemeinen nicht beeinträchtigen, von der Ermittlung der Metadaten M, M' ausgenommen werden. Damit wird die Spezifizität der Metadaten M, M' bezüglich eines tatsächlich wahrnehmbaren Unterschieds zwischen Bildern I, I' verbessert.In a subsequent third step 23 becomes the spatial gradient of the gray scale image $\frac{\partial GV}{\partial r}\left(i,j\right)$

determined or estimated along at least one local direction r. This allows essentially homogeneous differences in intensity between an original image I. and a processed image I ' that generally do not affect the perception of the image content by determining the metadata M , M ' be exempted. This changes the specificity of the metadata M , M ' for an actually noticeable difference between images I. , I ' improved.

zur Ermittlung von in horizontaler (x-) Richtung verlaufender Grauwertgradienten bekannt. Analog ist die Faltung eines Bildes I, I' mit einem vertikalen Sobel-Kern $$G_y=G_x^T=\left[\begin{array}{ccc}1& 2& -1\\ 0& 0& 0\\ -1& -2& -1\end{array}\right]$$

zur Ermittlung von in vertikaler (y-) Richtung verlaufender Grauwertgradienten bekannt. Auch weitere Verfahren zur Schätzung eines Ortsgradienten sind bekannt, beispielsweise eine Näherung des Laplace-Operators durch einen Laplace-Kern.Methods for estimating the spatial gradient are known. For example, the folding of an image I. , I ' with a horizontal Sobel core $$G_x=\left[\begin{array}{ccc}1& 0& -1\\ \mathrm{2nd}& 0& -\mathrm{2nd}\\ 1& 0& -1\end{array}\right]$$

for determining gray value gradients running in the horizontal (x-) direction. The folding of an image is analogous I. , I ' with a vertical Sobel core $$G_y=G_x^T=\left[\begin{array}{ccc}1& \mathrm{2nd}& -1\\ 0& 0& 0\\ -1& -\mathrm{2nd}& -1\end{array}\right]$$

for determining gray value gradients running in the vertical (y) direction. Other methods for estimating a spatial gradient are also known, for example an approximation of the Laplace operator by a Laplace kernel.

führt, hat sich für die Bestimmung von Metadaten M, M' die Bestimmung der Ortsgradienten in eine horizontale und in eine vertikale Ortsgradientenrichtung r als ausreichend erwiesen.While mathematically a single grayscale image GV (i, j) , depending on the investigated spatial gradient direction r, on any number of spatial gradient images $\frac{\partial GV}{\partial r}\left(i,j\right)$

has led to the determination of metadata M , M ' the determination of the spatial gradients in a horizontal and in a vertical spatial gradient direction r has proven to be sufficient.

und ein vertikales Ortsgradientenbild $$\frac{\partial GV}{\partial y}\left(i,j\right)=G_y\ast I$$

durch Faltung mit Sobel-Kernen ermittelt. Die Ermittlung der Ortsgradientenbilder für verarbeitete Bilder I' erfolgt analog durch Anwendung der gleichen Sobel-Kerne.In one embodiment, an original image I. a horizontal spatial gradient image $$\frac{\partial GV}{\partial x}\left(i,j\right)=G_x\ast \mathrm{I.}$$

and a vertical spatial gradient image $$\frac{\partial GV}{\partial y}\left(i,j\right)=G_y\ast \mathrm{I.}$$

determined by folding with Sobel cores. The determination of the spatial gradient images for processed images I ' is carried out analogously by using the same Sobel cores.

In the fourth step below 24th is a gradient histogram for each of the determined spatial gradient images H determines that for a plurality of intervals of a spatial gradient value assigns an absolute number or frequency of occurrence of spatial gradient values to each interval, all pixels of the image I. , I ' be taken into account. Such intervals are also known as bins. A histogram can thus be represented as a vector of absolute frequencies, the length of which is equal to the number of bins examined.

According to the exemplary embodiment with a horizontal and a vertical spatial gradient image, a horizontal gradient histogram is thus obtained H ₁ and a vertical gradient histogram H ₂ as source metadata M for an original picture I. certainly. Gradient histograms are used in an analogous manner H ₁ , H ₂ for a processed image I ' certainly.

An advantage of this embodiment is that a gradient histogram H ₁ , H ₂ a very compact representation of an image I. , I ' delivers. For example, gradient histograms have been used for some applications H ₁ , H ₂ with 8 bins each was found to be sufficient. Thus, when using a horizontal and a vertical gradient histogram H ₁ , H ₂ for pictures I. , I ' with a typical size of less than 2 ^31-1 pixels, a description of the metadata M 'M' with 2 x 8 x 4 = 64 bytes possible. The one for the transfer of this metadata M , M ' The bandwidth required is therefore negligible compared to that required for the transmission of the images I. , I ' required bandwidth. These are also gradient histograms H ₁ , H ₂ very easy to evaluate and with little computing effort.

3rd shows an embodiment of a method for local gradient histogram determination 3rd comprehensive steps 31 to 36 , with the metadata M , M ' from pictures I. , I ' be extracted.

und auf ein vertikales Ortsgradientenbild $\frac{\partial GV}{\partial y}\left(i,j\right).$

The steps 31 , 32 , 33 correspond to the steps 21 , 22 , 23 the based on the 2nd explained procedure for global gradient histogram determination 2nd and lead to location gradient images, for example to a horizontal location gradient image $\frac{\partial GV}{\partial x}\left(i,j\right)$

and on a vertical spatial gradient image $\frac{\partial GV}{\partial y}\left(i,j\right).$

Notwithstanding this in 2nd The methods shown are used for a local gradient histogram determination 3rd the determined spatial gradient images in step 34 tessellated in rectangular drawing files or patches. The patches can be arranged overlapping or disjoint, in particular adjacent, over the respective spatial gradient image and generally have the same dimensions. Each patch has a horizontal coordinate i and a vertical coordinate j assigned.

For example, an image I. , I ' with a horizontal dimension of 1920 pixels and a vertical dimension of 720 pixels in 10 x 10 patches with a horizontal dimension of 192 pixels and a vertical dimension of 72 pixels. Alternatively, such an image I. , I ' tessellated in 16 x 6 square patches with an edge length of 120 pixels. Each of these patches can have horizontal and vertical coordinates i, j , the coordinate values of the top left pixel of the patch are assigned.

In step 35 a horizontal and a vertical gradient histogram for each patch with the coordinates i, j H ₁ (i, j) , H ₂ (i, j) determined in principle in the same way, but limited to the respective patch, as in the step 24th the global gradient histogram determination 2nd based on 2nd has already been explained.

In contrast to the global gradient histogram determination 2nd , instead of a single horizontal and vertical gradient histogram H ₁ , H ₂ in the process of local gradient histogram determination 3rd according to the number of patches, several horizontal and vertical gradient histograms H ₁ (i, j) , H ₂ (i, j) determined and assigned to the coordinates of the patch.

In step 36 these determined horizontal and vertical gradient histograms H ₁ (i, j) , H ₂ (i, j) saved.

An advantage of this embodiment is that with the plurality of gradient histograms H ₁ (i, j) , H ₂ (i, j) a more accurate assessment of the perceptibility of a difference between an original image I. and a processed image I ' is possible, the description of this metadata M 'M' required amount of data compared to the method of global gradient histogram determination 2nd increased but over that Bandwidth requirements for the transmission of the images I. , I ' is still low or even negligible.

In the embodiment of the global gradient histogram determination 2nd to 2nd become pictures I. 'I' is rated as similar if a sufficient number of gradient values occur with the same or a similar frequency, the local position of the gradient values being disregarded. In contrast, the embodiment requires the local gradient histogram determination 3rd to 3rd Similar frequencies of the gradient values in the same patches of the images I. , I ' and thus enables a more precise and differentiated image comparison. In particular, an image comparison with higher specificity and with higher sensitivity is made possible than in the embodiment according to 2nd .

4th shows schematically the sequence of a method for validation 4th the correspondence (in terms of their perceptibility) between an original and a processed image I. , I ' .

Analogous to the steps 21 and 31 becomes a processed image I ' in one step 401 read in pixel by pixel. Below are in one step 402 Control metadata M ' from the picture I ' determined. In step 402 can be a method of global gradient histogram determination 2nd or a method of local gradient histogram determination 3rd be implemented, this method corresponds to the method with which the source metadata M of the corresponding original image I. were determined. In further embodiments, these methods can also be other methods of feature extraction.

The source metadata M of the original picture I. be in one step 403 read in and in one step 404 saved.

In a subsequent step 405 become the control metadata M ' and the source metadata M compared. In one embodiment with global gradient histogram determination 2nd are therefore each for the original image I. and the processed image I ' determined horizontal and vertical gradient histograms H ₁ , H ₂ to compare with each other. In one embodiment with local gradient histogram determination 3rd are the corresponding patches of the original image (arranged in the same location) I. and the processed image I ' determined gradient histograms H ₁ (i, j) , H ₂ (i, j) to compare with each other.

Methods for comparing histograms, in particular measures of difference between relative or absolute frequency distributions, are known from the prior art. For example, the use of Kullback-Leibler divergence (also referred to as relative entropy) has proven to be advantageous for the comparison of gradient histograms H ₁ , H ₂ , H ₁ (i, j) , H ₂ (i, j) exposed. However, other comparison methods or difference measures are also possible, for example the earth movers distance, the helm metric, the pearson chi-square divergence or the chi-square distance. Each of these comparison methods leads to a numerical value which shows the difference between two gradient histograms compared with one another H ₁ , H ₂ , H ₁ (i, j) , H ₂ (i, j) describes. It is possible to use several comparison methods in parallel, so that for a pair of gradient histograms compared with one another H ₁ , H ₂ , H ₁ (i, j) , H ₂ (i, j) a plurality of difference measures are formed.

Thus, for the exemplary embodiment with a gradient determination using a horizontal and vertical Sobel core with global gradient histogram determination 2nd two numerical values and with local gradient histogram determination 3rd correspondingly twice as many numerical values as the number of patches determined. When using a larger number of different Sobel cores, the number of numerical values increases accordingly.

In other exemplary embodiments, location gradient images based on one or more color channels can also be determined in addition to or instead of the location gradient images based on the gray value image, the step 22 (with global gradient histogram determination 2nd ) or the step 32 (with local gradient histogram determination 3rd ) is skipped in each case. Accordingly, the number of gradient histograms to be compared tripled H ₁ , H ₂ , H ₁ (i, j) , H ₂ (i, j) , if all three color channels are analyzed instead of the gray value image. One advantage of such exemplary embodiments is the improved accuracy or specificity and sensitivity of the comparison, in particular of images I. , I ' with at least partially synthetically generated image content to which a predetermined color value is assigned.

In the in 4th shown embodiment is a global horizontal difference measure d ₁ from the comparison of the global horizontal gradient histograms H ₁ of the pictures I. , I ' , a global vertical measure of difference d ₂ from the comparison of the global vertical gradient histograms H ₂ of the pictures I. , I ' , a plurality of local horizontal difference measures d ₁ (i, j) from the comparison of the local horizontal gradient histograms H ₁ (i, j) of the pictures I. , I ' as well as a plurality of local vertical difference measures d ₂ (i, j) from the comparison of the local vertical gradient histograms H ₂ (i, j) of the pictures I. , I 'determined.

gilt. Alternativ können auch für verschiedenen Gradientenrichtungen (horizontal, vertikal) und/oder für verschiedene Patches unterschiedliche Schwellwerte eingesetzt werden. The local difference measures d ₁ (i, j) , d ₂ (i, j) be in one step 406 evaluated. As an example, it can be determined whether any of the local difference measures d ₁ (i, j) , d ₂ (i, j) _locally reaches or exceeds a predetermined threshold value Θ. In other words: two pictures I. , I ' are considered not perceivably equivalent (not perceptually equivalent) if $$\underset{i,j,k}{{\displaystyle \text{Max}}}{d}_k\left(i,j\right)\ge {\mathrm{\Theta }}_{lOkal}$$

applies. Alternatively, different threshold values can also be used for different gradient directions (horizontal, vertical) and / or for different patches.

The threshold value θ _locally can be determined, for example, by means of a subject test depending on the application or user. Likewise, the number of patches in which the images I. , I ' be tesselated, determined by means of test subjects. With a comparatively small number of patches, for example with 10 patches, it is possible for a perceptual equivalence of the images I. , I ' to request that all patches _locally fall below the threshold value θ. With a comparatively high number of patches, for example with 1000 patches, it is possible, however, for a perceptual equivalence of the images I. , I ' only to require that the local difference measures d ₁ (i, j) , d ₂ (i, j) for a predetermined proportion of patches, for example for 90 percent of all patches, fall below the predetermined threshold.

An advantage of this embodiment is that the perceptual equivalence of the images I. , I ' is not adversely affected by the fact that the original image was taken, for example, by a camera I. , using the head unit HU locally limited symbols or icons are overlaid, which affect only a small proportion of patches.

The result of the step 406 can be formulated in binary terms ("threshold value is undershot for all patches" or "threshold value is not undershot for at least one patch"). But it is also possible that the result of the step 406 is formulated as a numerical value, which, for example, indicates the proportion of patches for which the threshold value is not reached.

Parallel to the step 406 can in one step 407 at least one criterion for a global comparison of difference measures d ₁ , d ₂ , d ₁ (i, j) , d ₂ (i, j) be evaluated. The accuracy (or the specificity and sensitivity) of a decision about the perceptual equivalence of the images can thereby be advantageously I. , I ' improve.

For example, in step 407 the sum of all local difference measures d ₁ (i, j) , d ₂ (i, j) formed and in a subsequent step 408 can be compared _globally with a threshold value θ: $${\displaystyle \sum _{i,j,k}{d}_k\left(i,j\right)}<{\theta }_{GlObal}.$$

Alternatively, in step 407 also the sum of squares of all local difference measures d ₁ (i, j) , d ₂ (i, j) formed and in a subsequent step 408 can be compared _globally with a threshold value θ: $${\displaystyle \sum _{i,j,k}{d}_k{\left(i,j\right)}^{\mathrm{2nd}}}<{\theta }_{GlObal}.$$

The sum of squares offers the advantage that individual statistical deviations in some of the local difference measures d ₁ (i, j) , d ₂ (i, j) ("Outliers") appear less strongly and are therefore easier to sort out. It is also possible to sum the squares only over part of the local difference d ₁ (i, j) , d ₂ (i, j) to form, for example, those patches in which the processed image I ' augmented elements are expected to be excluded from the sum of squares. The step 408 can deliver a binary result depending on the result of the comparison of the sum or the sum of the squares. For example, the step 408 provide the result (logically) true if the condition Σ _{i, j, k} d _k (i, j) <θ _global or the condition Σ _{i, j, k} d _k (i, j) ² <θ _global is fulfilled and in the other case the result is (logically) false (false).

In one embodiment, it is also possible to step 406 and / or crotch 408 to make a plurality of comparisons, based on different measures of difference d ₁ , d ₂ , d ₁ (i, j) , d ₂ (i, j) are related and use different threshold values. In this embodiment, the step delivers 406 and / or the step 408 a variety of binary results. This can improve the quality of the comparison between the images I. , I ' , in particular the specificity and / or the sensitivity of an equivalence decision, can be improved. That is, bad (incorrectly transmitted) processed images I ' be sorted out while processing good (correctly transmitted) images I ' including the correctly augmented processed images I ' be accepted.

Furthermore, in addition to binary output (error-free / error) through the step 406 and / or the crotch 408 also issue an error message or warning based on a parent system US that the image transmission from the frontend FE to the backend BE controls, can take precautionary measures. For example, a requirement for higher quality in the Image processing, such as the degree of image compression.

In one step 409 be the of the steps 406 and 408 delivered binary results logically linked, for example by means of a logical AND link. In this case, the output of the step 409 if (logically) true if every one of the steps 406 and 408 tested inequalities is satisfied. In other words: the output of the step 409 then shows a perceptual equivalence of the images I. , I ' if all of the gradient histograms examined H ₁ , H ₂ , H ₁ (i j) H ₂ (i , j) based on the difference dimensions used d ₁ , d ₂ , d ₁ (i, j) , d ₂ (i, j) match within the predetermined threshold values.

However, embodiments are also possible in which the step 409 the binary outputs of the steps 406 and 408 can be linked using any Boolean (logical) expression, for example based on the application used by the head unit HU processing carried out and / or is determined on the basis of the user.

Dealing with one by the step 406 and / or the crotch 408 The error message or warning issued is the responsibility of the user or the higher-level system US . The parent system US may have additional protection information that is generated outside of this system or outside of the 1 shown transmission path from the image source S to the display module D Are available. It can therefore be helpful if this parent system US from validation 4th Receives information of high resolution and / or accuracy, for example instead of or in addition to binary information in accordance with the comparisons of the difference measurements d ₁ , d ₂ , d ₁ (i, j) , d ₂ (i, j) with the predetermined threshold values also categorical (enumerated) and / or numerical information (integers or floating point numbers). Such information can be in an optional step 410 from the differences d ₁ , d ₂ , d ₁ (i, j) , d ₂ (i, j) won and to the parent system US be transmitted. Based on this and other information that the parent system US this can make a holistic decision so that a higher security standard can be achieved. In addition, such information, in particular numerical information, can be used for downstream machine learning processes (machine learning), so that the processing of the original images I. in processed images I ' is further improved during operation.

The local gradient histogram determination 3rd according to 3rd sets the function of a frontend FE according to the 1 um and can in a conventional integrated circuit (IC), for example in a camera chip of an image source designed as a camera S , to get integrated. The storage of the metadata M , M ' causes no significant hardware costs due to the comparatively small size of less than 2000 bytes, for example.

The graying and the determination of the gradients, for example using Sobel cores, is quick and easy and can be integrated well in a pixel pipeline. A three-line memory is sufficient to calculate the Sobel gradients. The establishment of the gradient histograms H ₁ , H ₂ , H ₁ (i, j) , H ₂ (i, j) can also be integrated in a pixel pipeline and only requires very simple logic. So that's the frontend FE to realize with low hardware costs. The transmission technology known from the prior art can continue and also for the transmission of the metadata M , M ' be used.

The validation 4th according to 4th sets the function of the backend BE according to 1 um and can in an available IC, for example in one of the display module D assigned timing controller (TCON). The received source metadata M of the original picture I. can be stored in an available memory due to the small size compared to the image data.

The extraction of the control metadata M 'from the processed image I ' is identical to the logic of the frontend FE and therefore consumes only minimal hardware resources. The complexity for determining difference measures d ₁ , d ₂ , d ₁ (i, j) , d ₂ (i, j) in step 405 is limited because only a few numbers have to be calculated. For example, for 100 patches with 10 bins each 1000 Values in horizontal local gradient histograms H ₁ (i, j) as well as 1000 values in vertical local gradient histograms H ₂ (i, j) , only a total of 2000 values are calculated. This is compared to the number of numerical values used to describe a typical picture I. , I ' are required (typically a few million integer values), negligible.

The effort for the implementation of the steps is similar 406 to 408 to evaluate the difference dimensions d ₁ , d ₂ , d ₁ (i, j) , d ₂ (i, j) low to negligible compared to the effort required to prepare and display an image I. , I ' typical size. So both the backend requires BE as well as the frontend FE only very little additional hardware resources.

Reference list

BE

Backend

D

Display module

d ₁

global horizontal difference measure

d ₂

global vertical difference measure

d ₁ (i, j)

local horizontal difference measure

d ₂ (i, j)

local vertical difference measure

F

Error status

FE

Frontend

H ₁

horizontal gradient histogram

H ₂

vertical gradient histogram

H ₁ (i, j)

horizontal gradient histogram

H ₂ (i, j)

vertical gradient histogram

HU

Head unit

I.

original picture

I '

processed image

M

Source metadata

M '

Control metadata

S

Image source

US

parent system

2nd

global gradient histogram determination

21 to 24

step

3rd

local gradient histogram determination

31 to 36

step

4th

Validation

401 to 410

step

QUOTES INCLUDE IN THE DESCRIPTION

This list of documents listed by the applicant has been generated automatically and is only included for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• DE 3913620 A1 [0003]
• DE 19549216 A1 [0004]

Claims (3)

Method for transferring images (I, I ') from an image source (S) to a display module (D) in a vehicle, the image source (S) providing an original image (I) which is converted into a processed image (I ') is transformed and transmitted to the display module (D), characterized in that at least one location gradient image is determined for each image (I, I') and metadata (M, M ') comprising at least one gradient histogram (H ₁ , H ₂ , H ₁ (i, j), H ₂ (i, j)) are determined and based on at least one difference measure (d ₁ , d ₂ , d ₁ (i, j), d ₂ (i, j)), the measures the difference between the source metadata (M) of the original image (I) and the control metadata (M ') of the processed image (I'), the perceptual equivalence of the images (I, I ') by comparison with a predetermined threshold is determined. Procedure according to Claim 1 , characterized in that for each image (I, I ') a horizontal location gradient image is determined by means of a horizontal Sobel core and a vertical location gradient image by means of a vertical Sobel core and a global gradient histogram (H ₁ , H _2.) for each location gradient image ) and / or in each case a plurality of local gradient histograms (H ₁ (i, j), H ₂ (i, j)) are determined via tessellated patches correspondingly arranged in the original image (I) and in the processed image (I ') and at least one difference measure (d ₁ , d ₂ , d ₁ (i, j), d ₂ (i, j)) is designed as a Kullback-Leibler divergence. Device for transmitting images (I, I ') to Claim 1 or 2nd comprising an image source (S), a front end (FE), a head unit (HU), a back end (BE) and a display module (D) arranged along a transmission chain, characterized in that - the front end (FE) for determining at least one Location gradient image from an original image (I) and for determining at least one gradient histogram (H ₁ , H ₂ , H ₁ (i, j), H ₂ (i, j)) from a location gradient image and for forming source metadata (M) is set up from this and - the head unit (HU) is set up for transforming an original image (I) into a processed image (I ') and for transmitting source metadata (M) and - the backend (BE) for determining at least one Location gradient image from a processed image (I ') and to determine at least one gradient histogram (H ₁ , H ₂ , H ₁ (i, j), H ₂ (i, j)) from a location gradient image and to form control metadata (M ') is set up from this and - the backend (BE) also for determination g at least one difference measure (d ₁ , d ₂ , d ₁ (i, j), d ₂ (i, j)), which is the difference between the source metadata (M) of the original image (I) and the control metadata (M ') of the processed image (I'), and for determining a perceptual equivalence of the images (I, I ') by comparing the at least one difference measure (d ₁ , d ₂ , d ₁ (i, j), d ₂ ( i, j)) is set up with a predetermined threshold value.

Images