Method and apparatus for processing image signals of an image sensor of a vehicle

Abstract

The invention relates to a method for processing an image signal of an image sensor of a vehicle. The method comprises the step of reading an image signal of an image sensor. The image signal represents a color measurement value vector having a plurality of input measurement values. The method further comprises the step of determining weight values ​​according to the input measurement values ​​using a determination rule. The method further comprises the step of obtaining a model matrix from a plurality of stored model matrices according to the weight values. For possible input measurement values, each of the model matrices is generated using a training rule. The method comprises the step of generating a color reconstruction matrix according to a generation rule using the obtained model matrix and at least one determined weight value. The method further comprises the step of using the generated color reconstruction matrix for the color measurement value vector to generate an output color vector, the output color vector representing the processed color image signal.

Description

Technical Field

[0001] The invention relates to a device and a method for processing image signals of an image sensor. The subject matter of the invention is also a computer program. Background Art

[0002] For color reconstruction when processing the image sensor data, a linear color correction matrix may be used, for example, which can be applied to the acquired image data. Summary of the invention

[0003] Against this background, a method is proposed with the aid of the solution presented here, and furthermore a device using this method and finally a corresponding computer program are proposed. Advantageous embodiments and improvements of the device according to the invention can be achieved by means of the measures listed in the following description.

[0004] According to an embodiment, in particular a unified nonlinear color reconstruction can be performed. In this context, nonlinearity can, for example, mean the use of different display formats, additionally or alternatively, a nonlinear division of the color space, and a recalculation in a nonlinear manner with the aid of a multi-matrix system. In other words, a nonlinear combination of linear subtasks can be used for color reconstruction. For example, instead of a conventional linear color correction matrix, color reconstruction can be made into an optimization task or a minimization task, which can be optimized for a nonlinear color reconstruction mode. For this purpose, the following hardware resources can also be used: the hardware resources can be optimized for this improved nonlinear color reconstruction.

[0005] Advantageously, according to embodiments, in particular a color reconstruction can be achieved, which serves to improve the color reconstruction performance or the color reconstruction effect. For example, the nonlinear color reconstruction can be transformed with the aid of a plurality of color reconstruction matrices into an optimization task, which can be solved in an advantageous manner with the aid of conventional methods. Thus, in particular the sensor data can be converted into a separate color space with a specifically adapted color resolution, which is optimized for so-called computer vision. In this way, a high resolution can be achieved in the critical color components of the observed scene, in order to improve the evaluation of the signal with regard to different lighting conditions.

[0006] A method for processing an image signal of an image sensor of a vehicle is proposed, wherein the method comprises the following steps:

[0007] reading an image signal from an image sensor, wherein the image signal represents a color measurement value vector having a plurality of input measurement values;

[0008] In the case of using a determination rule, the weight values ​​are determined based on the input measurement values;

[0009] Ascertaining a model matrix from a plurality of stored model matrices as a function of the weight values, wherein each of the model matrices is generated for a possible input measurement value using a training rule;

[0010] generating a color reconstruction matrix according to a generation rule, using the determined model matrix and at least one of the determined weight values;

[0011] The generated color reconstruction matrix is ​​applied to the color measurement value vector to generate an output color vector representing the processed color image signal.

[0012] In this case, the above-mentioned method can be carried out individually for each color value measurement vector.

[0013] The method can be implemented, for example, in software or hardware or in a hybrid form of software and hardware (for example in a control unit or device), in particular as an integrated component of an image sensor with an integrated ISP. Such an implementation can be particularly usefully implemented in a dedicated or integrated ISP (image signal processor) module. The ISP module can be integrated into the image sensor module in order to convert the color measurement values ​​directly into output color values ​​after the measurement.

[0014] According to one embodiment, in the step of reading, an image signal can be read that represents a color measurement value vector related to a first color space. In this case, in the step of using, an output color vector can be generated, which output color vector relates to a second color space different from the first color space. In addition, in the step of determining, weight values ​​can be determined, which weight values ​​represent a nonlinear division of the first color space. According to the International Commission on Illumination (CIE = International Commission on Illumination, Commission Internationale de Illumination) ), the first color space may be, for example, an image sensor-specific color space of an RGB color pattern, an RCB color pattern, or an RCG color pattern, or a preprocessed RGB color space, or an xyz color space. The second color space may be, for example, an RGB color space, an xyz color space, a YUV color space, or a LUV color space, wherein these color spaces may have color position-dependent ( ) accuracy. The description of the color spaces selected here by way of example is to be understood as non-exhaustive; therefore, the solution proposed here can be implemented without restriction of generality when using color spaces different from those proposed here. This embodiment offers the advantage that, based on a given first color space or input color space, taking into account a desired second color space or output color space, a color reconstruction with a high reconstruction effect can be achieved taking into account particularly interesting areas of the input color space.

[0015] In the determination step, a weighting value and / or only one model matrix can also be determined for each input measured value. This embodiment offers the advantage that the color measured values ​​in the region of interest of the color space of the color measured value vector can be taken into account individually by corresponding weighting.

[0016] Furthermore, in the step of generating, a color reconstruction matrix can be generated by assigning the obtained model matrix to the associated determined weight values ​​while taking into account the input measurement values. The obtained model matrices can be reordered and mixed according to the weight values. During the assignment, the subordination of the model matrix to the weight values ​​can be determined while taking into account the input measurement values. This embodiment provides the following advantages: a color reconstruction matrix with a resulting high reconstruction performance can be generated in an accurate and resource-saving manner.

[0017] Furthermore, the method may include the following steps: generating a model matrix for possible input measurement values ​​using the training rules. This embodiment provides the advantage that the model matrix can be saved as configuration data and can be updated by training. As a result, computing power for implementing the method can be saved.

[0018] In this case, in the step of generating, each of the model matrices can be generated for each possible input value. In this case, the position of the input measured value in the color space of the color measured value vector can be taken into account by the corresponding model matrix, and additionally or alternatively, the value range of the input measured value can be taken into account. This embodiment offers the advantage that the model matrix can be provided for generating a color reconstruction matrix, which in particular also enables a reliable and accurate reconstruction of the region of interest in the color space of the color measured value vector.

[0019] Advantageously, another embodiment of the solution proposed here is that in the step of determining, a model matrix is ​​determined: the assigned weight values ​​of the model matrix are not equal to zero. This embodiment of the solution proposed here offers the following advantages: it can be implemented very easily digitally or in circuit technology. In particular, in the nearest neighbor implementation, only one weight value is not equal to zero; and in the linear interpolation in the three-dimensional color space, only 8 weights are not equal to zero. In this way and method, the entire input color space can always be mapped into the target color space by cleverly selecting 8 matrices. In the case of knowing that the selected matrices are fixed, further processing can be optimized in hardware technology.

[0020] The solution proposed here also provides an apparatus, which is configured to execute, control or implement the steps of the variant of the method proposed here in a corresponding device. Through these implementation variants of the apparatus form of the present invention, the task on which the present invention is based can also be solved quickly and efficiently.

[0021] For this purpose, the device can have at least one computing unit for processing signals or data, at least one storage unit for storing signals or data, at least one interface for reading sensor signals from sensors or for outputting data signals or control signals to actuators, and / or at least one communication interface for reading or outputting data embedded in a communication protocol. The computing unit can be, for example, a signal processor, a microcontroller or the like, wherein the storage unit can be a flash memory, an EEPROM (electrically erasable programmable read-only memory) or a magnetic storage unit. The communication interface can be designed to read or output data wirelessly and / or wired, wherein a communication interface that can read or output wired data can, for example, electrically or optically read the data from a corresponding data transmission line or output the data to a corresponding data transmission line.

[0022] In the present invention, a device can be understood as an electrical device that processes sensor signals and outputs control signals and / or data signals according to the sensor signals. The device can have an interface that can be constructed in hardware form and / or software form. For a hardware form of construction, the interface is, for example, part of a so-called system-specific integrated circuit, which contains various functions of the device. However, it is also possible that the interface is also an independent integrated circuit or is at least partially composed of discrete components. For a software form of construction, the interface can be a software module, which, for example, exists on a microcontroller together with other software modules.

[0023] A sensor system for a vehicle is also proposed, wherein the sensor system has the following features:

[0024] An image sensor for providing an image signal;

[0025] An embodiment of the above device, wherein the device and the image sensor are connected to each other in a data transferable manner.

[0026] In the sensor system, the above-mentioned embodiments of the device can be advantageously adopted or used in order to process the image signal of the image sensor.

[0027] Also advantageous is a computer program product or a computer program having a program code, which can be stored on a machine-readable carrier or storage medium such as a semiconductor memory, a hard disk memory or an optical memory, and which is used to execute, implement and / or control the steps of the method according to any of the embodiments described above, in particular when the program product or the program is implemented on a computer or a device. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] An exemplary embodiment of the solution proposed here is shown in the drawings and is explained in more detail in the following description. The drawings show:

[0029] Figure 1 A schematic diagram of a vehicle having a sensor system according to an exemplary embodiment is shown;

[0030] Figure 2 Show Figure 1 A schematic diagram of a device of a sensor system;

[0031] Figure 3 A flow chart of a method for processing according to one exemplary embodiment is shown.

[0032] In the following description of advantageous exemplary embodiments of the invention, identical or similar reference numerals are used for elements which are shown in the various figures and act similarly, wherein a repeated description of these elements is omitted. DETAILED DESCRIPTION

[0033] Figure 1 A schematic diagram of a vehicle 100 with a sensor system 110 according to an exemplary embodiment is shown. Vehicle 100 is a motor vehicle, for example a land vehicle, an aircraft or a watercraft, in particular a passenger car, a truck or another commercial vehicle. Light L having a spectrum shown symbolically here acts on vehicle 100. Figure 1 In the illustration of FIG. 1 , a windshield 102 and a sensor system 110 are shown in a vehicle 100 . In this case, light L passes through windshield 102 of vehicle 100 , wherein windshield 102 acts as a spectral filter.

[0034] The sensor system 110 has an image sensor 120 or an image sensor arrangement 120 and a device 130. The image sensor 120 is designed to provide an image signal 129. Specifically, the image sensor 120 is designed to provide the image signal 129 in response to and as a function of the light L. The image signal 129 represents a color measurement value vector having a plurality of input measurement values. (color measurement vector). The color measurement value vector relates to the color space defined by the image sensor 120. The device 130 is designed to process the image signal 129. The image sensor 120 and the device 130 are connected to each other in a data-transmittable manner or a signal-transmittable manner. According to the exemplary embodiment shown here, the sensor system 110 also has a storage device 140. The storage device 140 is connected to the device 130 in a data-transmittable manner or a signal-transmittable manner.

[0035] According to the embodiment shown here, the image sensor 120 has, by way of example and in the order starting from the side of the image sensor 120 facing the windshield 102, an optical device 121, a microlens 122, a color filter array 123 (CFA=Color Filter Array), a light sensor 124 and an analog-digital converter 125 (ADC=analog-digital converter). The optical device 121, the microlens 122 and the color filter array 123 are each used as a spectral filter for the light L. The color filter array 123 is similar to or corresponds to the color pattern applied to or before the light sensor 124. The light sensor 124 is used as an integrator with respect to the wavelength λ of the light L. The analog-digital converter 125 is used as a quantizer.

[0036] The device 130 is designed to perform image signal processing (ISP=Image Signal Processing) including color reconstruction. In this case, the device 130 is designed to generate an output color vector using the image signal 129 of the image sensor 120. The output color vector represents a processed image signal 139 (eg, either optimized for human vision or optimized for computer vision). The apparatus 130 will be discussed in more detail below.

[0037] In addition, Figure 11 and 106. The first block 104 represents, for example, a color value according to ISO 11664 in response to the processed image signal 139. The second block 106 represents a direct reference value, for example in the Luv color space (also ISO 11664) for a human observer and in response to the light L outside the windshield 102. A comparison can be made between the first block 104 and the second block 106.

[0038] Figure 2 Show Figure 1 Schematic diagram of the sensor system device 130. In addition, Figure 2 Show Figure 1 The device 130 has a determination device 231 , an ascertainment device 233 , a generation device 236 and an application device 237 . According to the exemplary embodiment shown here, the device 130 also has an alignment device 234 and a sorting device 235 .

[0039] The device 130 is configured to read an image signal 129 from an image sensor. The read image signal 129 represents in particular a color measurement value vector related to a first color space. The determination device 231 is configured to determine a weight value {α} as a function of the input measurement value represented by the image signal 129 using a determination rule. Advantageously, the determination device 231 is configured to determine a weight value {α} for each of the model matrices stored in the storage device 140 for each input measurement value. The determination device 231 is in particular configured to determine a weight value {α} that represents a nonlinear division of the first color space. In addition, the determination device 231 is configured to output the determined weight value {α} and a selection signal 232.

[0040] The ascertainment device 233 is designed to ascertain a model matrix {M} from a plurality of stored model matrices as a function of the weighting values ​​{α}.

[0041] Here, the determination device 233 is configured to determine the model matrix {M} while taking into account the selection signal 232. In addition, the determination device 233 is configured to read the model matrix from the storage device 140. When using the training rule, each of the stored model matrices is generated for possible input measurement values. The determination device 233 is also configured to output the determined model matrix {M}. The model matrix {M} is part of the mathematical model, and the model matrix is ​​optimized for the input values ​​or matched to the corresponding positions of the respective input values. The model matrix {M} is decomposed according to the value range or different sampling points. Advantageously, the determination device 233 is configured so that for a given color measurement vector 129, it is not necessary to select most of the model matrices 140, and thus the amount of data after the determination device 233 is kept very low.

[0042] The alignment device 234 and the sorting device 235 are configured to sort the determined model matrix {M} by working together and to align the sorted determined model matrix {M} with the correct or associated determined weight values ​​{α}. In other words, the alignment device 234 and the sorting device 235 are configured to compare the determined model matrix {M} with the determined weight values ​​{α} and to establish an assignment between them.

[0043] The generation device 236 is configured to generate a color reconstruction matrix M according to a generation rule using the determined model matrix {M} and the determined weight value {α}. The generation device 236 is also configured to output the color reconstruction matrix M. The generation device 236 is configured to use an assigned determined weight value {α} for each of the determined model matrices {M}. Thus, the weights of the determined model matrix {M} are obtained according to the measurement values ​​of the image sensor.

[0044] The size of the color reconstruction matrix M depends on the color measurement value vector and the output color vector According to one embodiment, the alignment device 234 and the sorting device 235 may be implemented as a part of the generating device 236 .

[0045] The application device 237 is designed to apply the generated color reconstruction matrix M to the color measurement value vector To produce the output color vector The output color vector represents the processed image signal 139. In this case, the application device 237 is designed in particular to generate the output color vector The output color vector is related to the color measurement value vector In addition, the application device 237 or the apparatus 130 is configured to output or prepare for output the processed image signal 139. In layman's terms, the apparatus 130 traverses once for each pixel of the entire image signal, so for a 2-megapixel image, it traverses 2 million times.

[0046] Figure 3 1 shows a flow chart of a method 300 for processing according to an exemplary embodiment. The method 300 can be implemented for processing an image signal of an image sensor of a vehicle. In this case, the method 300 can be implemented for processing in conjunction with and / or using an apparatus according to one of the above figures. In addition, the method 300 can be implemented for processing in conjunction with Figure 1 The sensor system in the device is used for processing.

[0047] In the method 300 for processing, in the step of reading 310, an image signal is read from an image sensor of a sensor system. The image signal represents a color measurement value vector with a plurality of input measurement values. Subsequently, in the step of determining 320, weight values ​​are determined as a function of the input measurement values ​​using a determination rule. In the step of determining 330, a model matrix is ​​determined from a plurality of stored model matrices as a function of the weight values. In this case, each of the model matrices is generated for the possible input measurement values ​​using a training rule.

[0048] Subsequently, in the step of generating 340, a color reconstruction matrix is ​​generated according to a generation rule, using the model matrix obtained in the step of determining 330 and the at least one weight value determined in the step of determining 320. In the step of using 350, the color reconstruction matrix generated in the step of generating 340 is used for the color measurement value vector to generate an output color vector representing the processed image signal.

[0049] According to one embodiment, the method 300 for processing further comprises a step of generating 360: generating a model matrix for possible input measurement values ​​using the training rule. The step of generating 360 generates each of the model matrices for a separate possible input measurement value.

[0050] With reference to the above-mentioned drawings and for the purpose of explaining the embodiments, the background and principles of the embodiments are generally, conclusively, in other expressions and more particularly explained below.

[0051] First, let's explain the definition of human color perception. Humans perceive color in a way that is described mathematically by CIE's XYZ color space or the CIE standard chromaticity system. This is based on the physiological fact that humans have three types of light sensors: red, green, and blue. The specifications of these perception units are used to produce a weighted integral over the light power spectrum as follows:

[0052]

[0053]

[0054]

[0055] function Defined in standard ISO 11664-1 by sampling values ​​with a spacing of 5 nanometers.

[0056] Human color perception can be summarized as follows:

[0057]

[0058] Therefore, with the help of the effective reconstruction function Φ(λ), the sensitivity curve given in the standard can be reconstructed.

[0059]

[0060]

[0061] Assuming that at the same sampling point λ S By sampling the spectrum at , the above equation can be simplified to:

[0062]

[0063] This applies similarly to Y and Z.

[0064] Use the following features of the refactoring function:

[0065] Φ(0)=1 (8)

[0066]

[0067] ∫Φ(x)dx=1 (10)

[0068] And simplify:

[0069]

[0070]

[0071]

[0072]

[0073] Now, the definition of color approximation is explained by color filter array sampling or CFA sampling (CFA = Color Filter Array; color filter array). Every color image sensor tries to imitate human behavior. Therefore, at least three partially independent color perception channels are installed in each color filter array. With the knowledge of the color migration curve of the CFA, the CFA response value can be extracted with the help of the mathematical expression mentioned above. The result is:

[0074]

[0075]

[0076]

[0077] …

[0078] Take the RCG color sensor as an example:

[0079] CFA1=R (18)

[0080] CFA2=C (19)

[0081] CFA3=G (20)

[0082] The color impression observed with the aid of a camera can be summarized as follows:

[0083]

[0084] To explain the color correction matrix in more detail, the following should be noted. Because it is usually assumed that the camera colors are close to the colors observed by humans, it is assumed that the color reconstruction can be done by a linear function (assuming that the initial color error is low):

[0085]

[0086]

[0087] is an offset correction or bias correction and is usually not necessary.

[0088] Assume that there are at least N spectra among the given number of and The pairs of π are available, then the color reconstruction problem can be expressed as follows:

[0089]

[0090] In the following, the above contents are expressed as a minimization task. Assume that M 3×3 is unknown, but the number If is known, the best matching M can be found by rewriting the problem as a known minimization problem. First,

[0091]

[0092] Breaking down into three equations:

[0093]

[0094]

[0095]

[0096] Each of these equations can also be described as follows:

[0097]

[0098]

[0099] …(31)

[0100] A shorter way to write it is:

[0101]

[0102]

[0103]

[0104] The same holds true for the submatrix or subordinate matrix M:

[0105]

[0106]

[0107]

[0108] And finally for the measurements:

[0109]

[0110] Equations (26) to (28) can be rewritten as:

[0111]

[0112]

[0113]

[0114] The three equations can be recombined into one equation by correlating the vectors and by repeating over the matrix C:

[0115]

[0116]

[0117] To minimize the equation in terms of the least squared error in x, known least squares methods and minimization solving methods (eg, conjugate gradient method) may be used.

[0118]

[0119]

[0120] If we expand these terms, we get a further simplification and we can see how the matrix A is calculated from the color measurements.

[0121] The use of multiple color correction matrices is discussed below. Assume that the color filter array has N C channels, each sensor generates a digital number DN∈[DN min ...DN max ]. In the case of current HDR sensors, there is: DN min =0 and DN max ≈2 24 For each individual measurement There is a matrix This enables a reconstruction of the human observation. A compromise is metamerism of the following type: a human observer distinguishes two colors, but a machine observes only one of these two colors (eg due to quantization effects).

[0122]

[0123] This solution requires a certain amount of storage space. The current solution is: Sampling and reconstruction are performed, for example as described above (Equation 5):

[0124]

[0125] By location Use equidistant scanning at each location, and assume that there are N C =3 color channels, the total number of color correction matrices is:

[0126]

[0127]

[0128] Regarding the reconstruction function Phi or Φ, the following properties can be preserved:

[0129]

[0130] Furthermore, Φ is decomposable and identical because the sampling function itself is decomposable and identical in every dimension:

[0131]

[0132] For example, for RCCG = Φ R (·)·Φ C (·)·Φ G (·) (52)

[0133] Linear interpolation can be used for Φ.

[0134]

[0135]

[0136] Alternatively, one can use the Φ=rect() function and obtain the nearest neighbor interpolation, which may be computationally more efficient in terms of hardware consumption.

[0137]

[0138]

[0139] In terms of implementation, color reconstruction can be simplified to:

[0140]

[0141] In the case of a CFA with three channels and three sampling points of the color correction matrix, the implementation is:

[0142] S(1)=0 (58)

[0143]

[0144] S(3)=2D=DN max (60)

[0145] C(1)=R(pix) (61)

[0146] C(2)=C(pix) (62)

[0147] C(3)=G(pix) (63)

[0148] This means that there are many , however, most of them have α values ​​of 0:

[0149]

[0150]

[0151]

[0152]

[0153] Depending on the choice of Φ, any other cases (different values ​​of α not equal to zero) are possible.

[0154] It should be noted that and Therefore, in the color reconstruction process, it must be calculated for each introduced pixel

[0155] Depending on the computing power and hardware resources, either one or eight matrices are used to perform the color reconstruction in the above example. The other matrices are zero, i.e. Since for these matrices there exists α p =0.

[0156] In terms of finding matrix values, it should also be noted that there are a known number of pairs Furthermore, it should be assumed that the sampling positions are known for M, which makes it possible to The value of α can be calculated according to formula 57.

[0157]

[0158]

[0159]

[0160] Only the shorter write-up is introduced in the last step.

[0161] Introducing N known In the case of measurement pair, we get:

[0162]

[0163] When a matrix of weighted color measurements is introduced, we obtain:

[0164]

[0165] And then refactoring the formula into After the components are obtained:

[0166]

[0167]

[0168]

[0169] When using the matrix transformation rule, we get:

[0170]

[0171]

[0172]

[0173] Combined into a system of equations:

[0174]

[0175] For N C= 3 channels and for each CFA channel there is In the case of sampling points, the following matrix is ​​obtained to train N chromatograms:

[0176] p max =27 (80)

[0177]

[0178] A 3N×243 ·x 243×1 =b 3N×1 (82)

[0179] Assuming the least squares method is used for minimization, the specific numbers obtained in this example are:

[0180]

[0181] The first step is a large matrix multiplication. This is especially true when N is large. For example, for N = 5·10 4 The spectral values ​​are optimized, which results in a memory requirement of approximately 2·300 Mbytes for this calculation. It should be noted that this calculation can be performed efficiently on a typical PC with MATLAB or Phython installed, since it fits in RAM.

[0182] The second step is to perform a matrix optimization on the resulting matrix for the optimization problem, which has a size of about 243×243·32 Bit≈1 Mbyte. This can also be easily processed by a standard PC.

[0183] According to (produced The selection of the initial color filter array (measured) results in the This may lead to: either introducing more sampling points M p , and thus generate more matrices for selection when reconstructing colors; or change the position of the sampling points of the model matrix, and thus recalculate each time When, according to the sampling position M p The reselection and / or matching selection of Φ uses more storage space.

[0184] Typically, optimization problems are computed with floating point precision. However, since the final result can most likely be computed with fixed point iterations, the entire optimization process can also be used directly as fixed point iterations. Therefore, if the evaluated values ​​of the matrix are used, the loss of precision can be avoided and it can be seen how much fixed point precision is needed to achieve the desired result.

[0185] In color reconstruction hardware with the nonlinear characteristics described above, the hardware requirements may be partially too high. For the matrix M of 8 nearest neighbors p For trilinear interpolation between , there should be 8 parallel color correction matrices. However, in the nearest neighbor calculation, by definition, only one color correction matrix should be calculated for each incoming pixel. The latter is considered to be a standard problem that should be able to be solved very efficiently by conventional hardware logic generation methods.

[0186] If trilinear reconstruction is not possible, the nearest neighbor method is recommended, and p More than 27 sampling points are introduced to compensate for the linear transformation.

[0187] However, for each color value received and to be reconstructed, the alpha value of that color should be calculated. In the case of trilinear reconstruction, alpha is in the range [0, 1], while for the nearest neighbor, alpha values ​​are either 0 or 1 only.

[0188] The computational device used to calculate the value of α can be optimized in terms of speed, power, and number of gates. max = 128 matrices M p As well as in the case of nearest neighbor reconstruction, sufficiently good results can be obtained, and with the help of trilinear interpolation even excellent result quality can be achieved.

[0189] According to an embodiment, the method can also be transferred to other color spaces and to the reconstruction of the color palette. It is assumed that human color perception is presented in the XYZ color space, but when other color spaces (especially color spaces with independent separate chromaticity planes) are used, the human chromaticity impression can also be defined only by the chromaticity plane. For the CIE Luv color space, there is:

[0190]

[0191] This can be used in the implementation above to directly obtain the target color space in the following way:

[0192]

[0193] This reduces the matrix and enables a faster search for the correct sampling point and number of sampling points. The matrix state can be improved by adding more sampling points or by changing the positions of the sampling points.

[0194] If an embodiment includes an “and / or” relationship between a first feature and a second feature, it should be interpreted as follows: the embodiment may have both the first feature and the second feature according to one embodiment; and may have either only the first feature or only the second feature according to other embodiments.

Claims

1. A method (300) for processing an image signal (129) of an image sensor (120) of a vehicle (100), wherein: The method (300) comprises the following steps: reading (310) an image signal (129) of the image sensor (120), wherein the image signal (129) represents a color measurement value vector having a plurality of input measurement values; In case of using a determination rule, determining (320) a weight value according to the input measurement value; Determining (330) a model matrix {M} from a plurality of stored model matrices according to the weight values, wherein each of the model matrices {M} is generated for an input measurement value using a training rule; According to the generation rule, a color reconstruction matrix is ​​generated (340) using the determined model matrix {M} and the associated determined weight values; applying (350) the generated color reconstruction matrix to the color measurement value vector to generate an output color vector representing the processed color image signal (139), Therein, in the step of generating (340), the color reconstruction matrix is ​​generated by assigning the determined model matrix {M} to the corresponding determined weight values ​​while taking into account the input measurement values, wherein the determined model matrix {M} is reordered and mixed according to the weight values.

2. The method (300) according to claim 1, characterized in that: In the step of reading (310), an image signal (129) is read, the image signal representing a color measurement value vector, the color measurement value vector relating to a first color space, wherein in the step of using (350), an output color vector is generated, the output color vector relating to a second color space different from the first color space, wherein in the step of determining (320), a weight value is determined, the weight value representing a non-linear division of the first color space.

3. The method (300) according to claim 1 or 2, characterized in that: In the step of determining (320), only one weight value is determined for each input measurement value, and / or, in the step of determining (330), only one model matrix {M} is determined.

4. The method (300) according to claim 1 or 2, characterized in that: In the step of determining (330), each of the model matrices {M} is generated for a separate input measurement value range.

5. The method (300) according to claim 1 or 2, characterized in that: In the step of determining (330), a model matrix {M} is determined in which the assigned weight values ​​of the model matrix {M} are not equal to zero.

6. A device (130), comprising a determining device (231), an obtaining device (233), a generating device (236) and an application device (237), wherein the determining device (231), the obtaining device (233), the generating device (236) and the application device (237) are used to implement the steps of the method (300) according to any one of claims 1 to 5.

7. A sensor system (110) for a vehicle (100), wherein: The sensor system (110) comprises: An image sensor (120) configured to provide an image signal (129); The device (130) according to claim 6, wherein the device (130) and the image sensor (120) are connected to each other in a data transferable manner.

8. A machine-readable storage medium on which a computer program is stored, the computer program being configured to execute and / or control the steps of the method (300) according to any one of claims 1 to 5.

Images