WO2012007059A1

WO2012007059A1 - Method for lateral chromatic aberration detection and correction

Info

Publication number: WO2012007059A1
Application number: PCT/EP2010/060364
Authority: WO
Inventors: Jan Klijn; Sasa Cvetkovic
Original assignee: Robert Bosch Gmbh
Priority date: 2010-07-16
Filing date: 2010-07-16
Publication date: 2012-01-19
Also published as: DE112010005743B4; DE112010005743T5

Abstract

A method for detection (5) and correction (3) of a lateral chromatic aberration shift of at least one color component (R, G, B) in digital image data (1) comprising pixels (x,y) includes, for at least one pixel (x,y), providing a shift function (f(r)) describing a shift of the at least one color component (R, G, B) depending on at least a variable shift parameter (caRB) and a radial position (r) of the pixel (x,y) in the digital image data (1), determining (50,700,800) an optimum value of the variable shift parameter (caRB) suitable to minimize the average shift of the at least one color component (R, G, B) that remains after LCA correction and applying (3) a correction function (outRB(x,y)) based on the shift function (f(r)) including the optimum value of the variable shift parameter (caRB) to the image data (1) to generate corrected image data.

Description

Title

Method for Lateral Chromatic Aberration Detection and Correction

The current invention relates to a method for lateral chromatic aberration detection and correction, to a corresponding optical image processing device and to a corresponding computer program product.

Prior Art

Chromatic aberration (CA), achromatism or chromatic distortion is a defect of optical lenses in connection with focussing all colors to the same convergence point at a certain distance of the lens (axial or longitudinal chromatic aberration) and/or at a certain location in the focal plane (transverse or lateral chromatic aberration). Both types of aberration are caused by different refractive indices of lenses for different wavelengths (dispersion).

Chromatic aberration is evident from images in form of color fringes along boundaries separating dark and bright areas (edges). The visual effects of longitudinal and lateral chromatic aberration are different in that longitudinal chromatic aberration causes fringes at all places in the image, whereas lateral chromatic aberration affects objects further from the image center. Fringes caused by lateral chromatic aberration, in contrast to those due to longitudinal chromatic aberration, are typically absent in the image center (typically coinciding with the lens center) and progressively increase toward the image corners.

The performance of color separation, involved in digital image processing, is strongly affected by lateral chromatic aberration and the resulting misalignment of the different color planes. Chromatic aberration can be reduced or eliminated by using achromatic and apochromatic lenses comprising glasses with different dispersion. However, such lenses are heavy and expensive. A reduction of chromatic aberration by stopping down lenses is, especially in case of lateral chromatic aberration, not always practicable, desired or effective.

Methods for reducing lateral chromatic aberration, hereinafter referred to as LCA, are known from US 2008/0284869 A1 , US 7,221 ,793 B2, US 2008/0291447 A1 , US 6,747,702 B1 , US 7,227,574 B2, US 2009/0052769 A1 , US 7,577,292 B2,

US 2007/0242897 A1 , US 7,346,210 B2, US 7,142,238 B1 , JP 2002 320237 A, US 7,425,988 B2, JP 2000 299874 A, US 2008/0007630 A1 , US 7,466,495 B2, US 2008/0062409 A1 and US 7,356,198 B2. The prior art methods, however, have significant disadvantages. In some cases, pre-calculated LCA model parameters for a certain lens type in a form of a lookup table are used to correct LCA artifacts. These methods are thus limited to the specific lens described by the look-up table. Other prior art documents just disclose LCA correction without taking into account the specific characteristics of lenses and LCA present and thus may under- or overcorrect artifacts. In some cases, specific artifacts, referred to as "purple fringing", are targeted, regardless of their origin, in form of a post-processing step. Typically, the prior art methods are not suitable to correct the resolution losses caused by LCA. Thus, a need for improved ways for detecting and correcting lateral chromatic aberration exists.

Disclosure of the Invention According to the invention, a method for lateral chromatic aberration detection and correction, a corresponding optical image processing device and a corresponding computer program product according to the independent claims is provided. Preferred embodiments are subject of the dependent claims and of the description. Advantages of the Invention

In contrast to the prior art, the current invention includes providing a LCA model to estimate LCA magnification or shift parameters caR and caB, corresponding to a chromatic aberration shift of the red and blue color plane or component with sub-pixel accuracy, from an arbitrary input image. The method does thus not require use of a reference image known in advance. Based on the LCA model, a feed-back control loop using a residue method and/or a global minimum search for obtaining parameters to perform an optimum correction of the LCA shift between red (R), green (G) and blue (B) color planes is employed. Finally, LCA correction is performed with the LCA parameters determined to remove LCA deformations from the image.

Further advantages and embodiments of the invention will be evident from the description and the accompanying drawings. It should be noted that the features mentioned above and to be explained below are not limited to the indicated combinations but are likewise usable in other combinations or alone without departing from the scope of the current invention.

The invention is illustrated by embodiments in the drawings and will be described with reference to the drawings.

Figures

Figure 1 shows a schematic overview of a method according to a preferred

embodiment of the invention.

Figure 2 shows a shift vector described by an LCA model function according to a preferred embodiment of the invention.

Figure 3 shows the principles of determining a shift vector according to a

preferred embodiment of the invention. Figures 4A to 4C show an edge in an image with high LCA and derived signals for determination of a LCA measure according to a preferred embodiment of the invention.

Figures 5A to 5C show an edge in an image with low LCA and derived signals for determination of a LCA measure according to a preferred embodiment of the invention.

Figures 6A to 6C show acceptance masks based on a distance from an image center, on the size of a G edge, and on the size of R, G and B edges according to preferred embodiments of the invention.

Figure 7 schematically shows steps of optimizing a parameter for LCA reduction according to a preferred embodiment of the invention.

Figure 8 schematically shows steps of optimizing a parameter for LCA reduction according to a preferred embodiment of the invention.

Figure 9 schematically shows steps of a method according to a preferred

embodiment of the invention.

In Figure 1 , a schematic overview of a method 100 according to a preferred embodiment of the invention is given.

The method 100 initially involves providing image data (step 1 ), e.g. from an image sensor or from a data storage device, and an appropriate signal preprocessing step 2 of the provided image data.

In step 3, an initial LCA correction in form of an interpolation (for example, a bilinear interpolation) is performed based on a function derived from the LCA model and supplied with an estimated parameter or parameter set caR and caB. The correspondingly treated image data are subjected to color separation 4, resulting in R, G and B color data. The R, G and B color data are fed into a feed-back loop 50 involving an LCA measurement step 5 according to a preferred embodiment of the invention, preferentially resulting in inter-color difference values between the R, G and B color channel, ARG, ABG and ARB. These inter-color difference values, based on intra-color differences, are, as described below, suitable to describe lateral chromatic aberration. The obtained inter-color difference values are used in a calculation step 6 to find optimum correction parameters caR and caB (in common referred to as caRB) for the model used in LCA correction step 3. When, by use of feed-back loop 50, sufficient correction could be obtained, correspondingly corrected image data optimize the performance of the color separator 4 and the (conventional) further image processing pipeline 7.

In the following, the LCA model used in the invention will be described.

The current invention essentially involves re-sampling R and B data (i.e., color components) to correct for the lateral chromatic aberration shift. LCA

misalignment of color components can be modeled pixel-wise with a 3^rd order polynomial: f{r) = caRB₃ ^* i³ + caRB₂ ^* i² + caRB^ ^* r¹ + caRB₀, wherein r represents a radial distance of the current pixel from the center of the sensor (typically, but not always coincident with the center of the lens), caRB are parameters of the LCA model for the R (caR) and B (caS) color planes, and caRB₀ describes a lateral chromatic aberration at the center of the image. The caRB parameters can, in a first step, be estimated.

The above misalignment function can, for reasons of robustness and easier parameter estimation, be simplified to a first order polynomial form, f{r) = caRB ^* r. The function f(r) describes a shift of the R (and B) pixels with respect to the G pixel (which is taken as a reference) and is, in Figure 2, depicted as a shift vector d, represented by {dRB_x,dRB_y). In Figure 2, the bilinear interpolation on a pixel grid 10 with axes x and y is shown. The top left image corner is defined as having coordinates (x,y) = (0,0). By convention, x and y are thus pixel coordinates, starting at 0 at the top left image corner and increasing to the right and towards down, respectively. In Figure 2, the real values of R or B pixels at any position (x,y) are not the ones that are measured on that position, but are displaced from position (x,y) to a new location 20 defined by the shift vector d. This new position 20 is most of the times not located on the existing pixel grid 10, so the intensity value at the position 20 has to be estimated from its neighbors. In case of bilinear interpolation, for instance, four pixels k, I, m and n around the real pixel position 20 are taken by observing the shift vector d. The integer part of the shift vector d, in the pixel grid 10, determines which four pixels k, I, m and n to use, and the fractional part of the shift vector d determines weights w_x and w_y for interpolation to obtain the intensity value at the position 20.

Advantageously, a bilinear re-sampling, as outlined above, is performed due to its simplicity. Bi-cubic or any other re-sampling is also possible, yet somewhat more expensive, since calculations with more pixels (for instance 16) and therefore more memory lines (3 or more instead of 1 ) are required.

The principles of determining an overall LCA vector, including the shift vector d and a position vector of a pixel (x,y) are visualized in Figure 3, wherein, as in Figure 2, a pixel grid 10 is shown and the top left image corner of an image is defined as having coordinates (x,y) = (0,0).

As seen from Figure 3, the final position of the shifted pixel is composed of the shift vector d, describing a shift of a pixel (x,y) by LCA, and by a position of the current pixel (x,y) given by vector r, which is represented with respect to an image center (Xc, Yc). The positions defined by vectors d and r may be given as {d_x,dy) and (r_x,r_y), respectively.

The final position of the pixel is calculated for each R and B pixel processed. The vector d depends on the chromatic aberration parameters caR (or caB, commonly referred to as caRB) as in d=caRB^*r, and vector r depends on the position, i.e. the radial distance of the pixel with respect to the optical image center (Xc,Yc). As can be noticed from Figure 3, the horizontal shift can be given as d_x = d ^* cosQ = d ^* r_x /r = caRB ^* r ^* r_x /r = caRB ^* r_x = caRB ^* (x - Xc). Similarly, d_y = d ^* s/^'ηθ = d ^* r_y/r = caRB ^* r ^* r_y /r = caRB ^* r_y = caRB ^* (y - Yc).

Hence, the shift vector can be represented in terms of the x and y position of the current pixel in the image, avoiding calculating its radial distance vector r. Thus, writing the above separately for the R and B color plane, one gets (notice that caR and caB mainly have opposite sign): dRx{x,y) = caR ^* (x - Xc); dRy{x,y) = caR ^* (y- Yc), and dBx{x,y) = caB ^* (x - Xc); dBy{x,y)= caB ^* (y- Yc). In case a 3^rd order LCA shift model is used, caRB are now vectors with 4 elements: caRB = [caRBz caRB₂ caRB^ caRB₀], and the corresponding shift vectors for a position (x,y) are: dRx{x,y) = caR₃ ^* (x - Xcf + caR₂ ^* (x - Xc)² + caR^ ^* (x - Xcf + caRo, dRy{x,y) = caR₃ ^* (y- Ycf + caR₂ ^* (y - Yc)² + ca ?i ^* (y - Ycf + caRo, dBx{x,y) = caB₃ ^* (x - Xcf + caB₂ ^* (x - Xc)² + caSi ^* (x - Xcf + caBo, dBy{x,y) = caB₃ ^* (y- Ycf + caB₂ ^* (y- Ycf + caSi ^* (y- Ycf + caB₀.

Here, caR₀ and caS₀ are present to accommodate a possibility that the value of LCA is not equal to zero in the optical center of the image. On the other hand, if the optical center of the lens (and hence a reference, zero point for the LCA) does not coincide with the image center, then one has to introduce an offset:

Xc =Total number of pixels (horizontal) 1 2 + offset (horizontal), and

Yc = Total number of pixels (vertical) 1 2 + offset (vertical).

Knowing all model parameters and all shift vectors, one needs to obtain, as the last step, an output pixel value which represents a real value that should have been on a position (x,y) if there was no LCA. In other words, a value is to be found which is suitable to correct or counteract the LCA. For this purpose, various interpolation techniques can be used, for instance a bi-linear interpolation.

For calculating a real pixel value that corresponds to a current pixel position, one needs to obtain the surrounding four pixels k, I, m and n (see Figure 2), given by a shift vector d(dRBx,dRBy), for instance equal to (2.2,-2.45). Furthermore, one needs to obtain a sub-pixel position given by weights w_x and w_y. If a position of the current pixel is (x,y), the coordinates of the neighboring pixels are given as: k: x+floor dRBx{x,y)), y + floor dRBy{x,y)), I: x+floor dRBx{x,y)) + 1 , y + floor dRBy{x,y)), m: x+floor(dRBx(x,y)), y + floor(dRBy(x,y))+ , n: x+floor{dRBx{x,y)) + 1 , y + floor{dRBy{x,y^ , and the corresponding weights are given by w_x = dRBx{x,y) - floor{dRBx{x,y)), w_y = dRBy{x,y) - floor{dRBy{x,y)), wherein floor{.) is a function giving an integer part of the shift. It is equal to a first integer value smaller than its argument. In the previous example, the following holds:

(dRBx,dRBy) = (2.2,-2.45), floor(dRSx) = 2 and floor(dRSy) = -3.

Finally, a bilinear interpolation function (as in Figure 2) gives an output R or B value: outRB(x,y) = ( -w_x) -Wy)^*k+w ( -Wy)l+( -w_x)*Wy^*m+w_x ^*w_y ^*n, which represents a real R (or B) value that should have been on a position (x,y), if there was no LCA.

In case lateral chromatic aberration is present in an image, the R and B color channels or components are misaligned with the G color channel or component. As an example, Figure 4A depicts such an edge where R and B color edges are shifted from the G color edge which is used as a reference edge. This shift of the color planes mainly has an opposite direction (but a similar or equal quantity) for R and B color channels.

The purpose of LCA measurement is thus to estimate or quantify the amount of LCA based on that color shift between edges in different color channels. In Figure 4B, intra-color difference signals A(R), A(G) and Δ(Β) are shown, representing, respectively, a high-pass version of the corresponding signals from Figure 4A. The corresponding inter-color differences |Δ(/?) - Δ(Θ)|, |Δ(β) - Δ(Θ)| and |Δ(/?) - Δ(β)| (Figure 4C) indicate the LCA shift. In cases of a lens corrected for a chromatic aberration, these inter-color differences are small (cf. e.g. Figures 5A to 5C) while, if LCA is present, they are large.

Since color edges can have different amplitude and sign, to enable good matching between different colors as well as non-dependence on the image brightness, intra-channel color differences have to be normalized. An

advantageous form of normalization gives a formula for intra-channel difference:

AC = *- ^HP(C>

LP(C) where k is a scaling constant, HP(C) represents a high-pass version and LP(C) is a low-pass version of a certain color channel C, where C = R,G,B. These calculations can be applied in horizontal, vertical or diagonal directions. HP(C) is, as stated, a high-pass filtered color signal, where filters that can be used are for instance [1 -1 ] or [1 0 -1 ]. LP(C) is a low-pass filtered color signal, using for instance a [1 1 ], [1 0 1 ], [1 2 1 ] or [3 8 10 8 3] filter. These filters are applied in the direction of calculation, for instance in horizontal, vertical or diagonal direction. Since edges in the diagonal direction are detected with both horizontal and vertical filters, diagonal calculations can be skipped.

Another possibility for LCA measurement is using kC

AC =^■

LP(C) where variables have the same meaning as in the previous explanation. The meaning of this approach lies in the image formation model which states that image is composed of the illumination layer (low-pass signal) multiplied with the detail layer (high-pass signal). By dividing the image with the low-pass information, a high frequency signal Δ(Ο) remains which can be further used in the same way as in the previous description.

The second chromatic aberration measurement step is to calculate inter-color differences ARG = \A{R) - A{G)\, ABG = \A{B) - A{G)\ and ARB = \A{R) - A{B)\.

The better LCA is corrected (by a corresponding correction method or an optimized lens), the smaller are the values of these differences, since color differences A(R), A(B),A (G) then overlap better (see Figures 5A to 5C). The original ARG formula can be rewritten as:

A(R)-A(G)

\HP(R) HP(G)

\ LP{R) LP{G)

Herewith, the same result as with the second A(C) formula is obtained, meaning that both formulas lead to identical results. A second form, however, simplifies the calculation.

In the method shown in Figure 1 , in the LCA measurement step 5 sums of ARG, ABG and RB are calculated per pixel in the image field/frame and the total sum values ARG, ABG and ARB are output, which can be used for calculation of the parameter(s) of LCA model (or, more precisely, for estimating the adequacy of these parameters for reducing LCA). The sum gives an estimate on the quality of a correction previously performed.

The original assumption is that ARG, ABG and ARB signals are discriminative enough to only measure the effect of LCA in the image. However, they do not measure only wanted lateral aberration color differences, but they may also unwontedly include various other effects like longitudinal color aberration, purple fringing caused by blooming and saturated image signals, noise, non-perfect normalization factors, etc. Therefore, even in cases of no LCA, the output of these measurements will not be zero. To improve the performance of this algorithm, several additional considerations and modifications are therefore needed.

Firstly, LCA shift according to a simple model is linearly dependent on the radial distance of the pixel from the image center: f(r) = caRB*r. For small radial distances, this shift is very small and can not be estimated well, since mainly noise will be measured. To prevent this, it is more advantageous not to perform any measurement close to the image center (where LCA is low), for r < /¾, for instance /¾ = 0.2^*D, where D is the length of a half diagonal of the image. In other words, as shown in Figure 6A, an acceptance mask M based on a distance /¾ from an image center of an image 60 is defined. The distance /¾ can also depend on the maximum expected LCA shift and correspond to a value where for instance the absolute value of the LCA color shift equals half a pixel. As a simplification, in order to decrease complexity of calculating the acceptance mask M, a rectangle can be used (like in the acceptance mask M of Figure 6A). In this case, the acceptance condition is that the current pixel coordinates (x,y) satisfy the criteria

\x - Xc\ > /¾ OR \y- Yc\ > /¾_, where OR represents a logical or function, |.| represents an absolute value and (Xc , Yc) are coordinates of the center of the image.

Furthermore, it is considered more advantageous to measure (only) in the neighborhood of the large edge transients and to skip small edges which will add a noise-like component to the measurement. For example, one can measure at places where output of Δ(Θ) signal is large, or Δ(Θ) > TH, TH being a threshold, as shown in Figure 6B. However, in this case desirable measurement areas of (R) and Δ(β) could also be excluded, especially if inter-color shift is large (Figure 6B, only the shaded part is measured). The region of interest can therefore broadened by performing measurement operations also on the neighboring pixels (additional pixels left and right for a horizontal filter and up and down for a vertical one). However, it is easier and potentially advantageous to modify the acceptance condition to all pixels that satisfy

{A{R) > TH) OR (Δ(Θ) > TH) OR ( Δ(β) > TH), as shown in Fig 6C, or even A{R) + A{G) + Δ(β) > TH.

For reasons of longitudinal chromatic aberration and in general due to a different response of R and B color channels to light, it is better to use different noise removing thresholds for G and for R and B. Hence,

(A(R) > TH_G) OR (A(G) > TH_RB) OR (Δ(β) > TH_RB).

In summary, as shown in Figures 6B and 6C, acceptance masks can be defined depending on the size of a G edge and on the size of R, G and B edges.

In certain cases it may be problematic to find a minimum of the ARG, ABG and ARB functions. For instance, when aliasing or high frequency features are present in an image, these functions can falsely measure non-wanted features. To prevent this, it is possible to exclude areas with higher frequency features. However, this operation is costly, since one has first to detect whether a feature of interest is an isolated edge (with no other edges in its neighborhood). For the detection of isolated edges, firstly an edge map with one pixel wide edges may be generated and after that all edges without another edge in their neighborhood, which could be confused with an LCA shifted edge, can be excluded.

Finally, purple fringing caused by blooming and saturated image signals

(streaking) can be confused for a LCA effect. This problem can be solved by excluding all measurements in the neighborhood of very bright pixels. For example, all pixels in (-δ,δ) neighborhood of the bright pixel in both horizontal and vertical direction can be excluded.

An essential step of the current invention may consist in estimating the caRB parameters by minimizing ARG, ABG and ARB functions in a feed-back loop comprising a number of optimizing iterations, as visualized in Figure 7. Method 700 of Figure 7 will be exemplified with caR but can, as explained below, also start with a caB value or may be performed in parallel for caR and caB. As mentioned before, iterations are started with an initial value, e.g. with an estimated default value or a value of zero for the LCA parameter caR. This step corresponds to step 702 of Figure 7 when executed for the first time. In step 702 of Figure 7, the initial caR value can be changed with a predefined value δ representing a modification value essentially used to modify the caR values in the successive iteration(s). In the method shown in Figure 7, this value is, in the first iteration, multiplied with either 1 or -1 , depending of a setting of a variable initially provided in a variable setting step 701. Essentially, this variable defines the direction of optimization by either increasing or decreasing a value of caR by providing a value of δ with either a positive or a negative sign. In summary, the operations in step 702 can be written as caR+sign^*6. Alternatively, in step 702, as mentioned, a caR value can in the first iteration simply be predefined.

Based on the value ca/?+sign^*5 (or, in the first iteration, a predefined value) provided in step 702, a LCA correction is performed in step 703 (corresponding to step 3 of Figure 1 ). Subsequently, data are subjected to color separation in step 704 (step 4 of Figure 1 ). Color separation results in R, G and B values for every pixel of the image.

By executing an LCA measurement step 705 (step 5 of Figure 1 ), Δ(/?), Δ(β), Δ(Θ), ARG, ABG and ARB values are calculated. The parameters caR and caB are mutually independent and independently influence the ARG and ABG values, while ARB is dependency influenced by any change of caR or caB.

In step 706, a reference ARG value REF_ARG is set, e.g. to a value of ARG from the previous field/frame (iteration). For reasons of robustness, REF_ARG can advantageously be set to a first order recursively filtered value of ARG:

- 1 )), where n is the iteration number.

Steps 707 and 708 can be performed to take into account an acceptance mask as explained above. In step 707, a check is made whether the current pixel satisfies a defined set of conditions (see above), and, if it does so, it is added to an acceptance mask M. For all pixels in an image belonging to the acceptance mask M, all ARG and RB values are summed in step 708 to give a measure of the resulting lateral chromatic aberration.

As shown e.g. in Figure 4, the initial value of for e.g. ARG is typically high. The LCA defect is thus corrected by reiteration of method 700 including a caR parameter modified by a δ value and by re-sampling the R color plane in the LCA correction in step 705 (step 5 of Figure 1 ), following the model of LCA defect given by the polynomial f(r) and the current value of the caR parameter.

After each iteration, a comparison step 709 is performed. If, in step 709, a newly calculated value of ARG is found to be larger than the reference value REF_ARG set in step 706, the parameter caR was changed in a wrong direction (i.e. by using a wrong sign) and the method 700 proceeds to step 71 1.

Consequently, in step 71 1 the sign is changed and, in the next cycle, caR is lowered by multiplication of δ with the changed sign. If, in contrast, a new value of ARG is found to be smaller than REF_ARG in step 709, the parameter caR was changed in a correct direction and, by continuing the procedure (iterating) using the unchanged sign, finally a caR parameter resulting in a minimum value of ARG will be found. At every iteration (and when in step 709 the measured ARG value was found to be smaller than the reference value), in step 710 the current caR value is memorized by setting it as an (currently) optimal caR value.

The above method steps are equally used for the parameter caB and an error signal ABG, and the method 700 is also performed on a B color plane, where an optimal parameter caB is found. The search for the best caR and caB parameters can be performed in parallel fashion, since they are independent of each other. The ARB error signal also gives valuable information about the parameters of the model of LCA. It can be used in addition to the errors ARG and ABG in the same algorithm to insure that the basic method is not, from a certain point, influenced by convergence problems. For instance, if one observes that the value of ARB starts to increase, it can be concluded that the caR (or caB) parameter is changed in the wrong direction.

In addition and/or alternatively to comparing with a reference value in a feed-back loop approach for finding the optimal value of caR and caB (resulting in minimum errors &RG and ΔΒβ), a global search strategy 800 as shown in Figure 8 can also be performed. In Figure 8, method steps identcal or essentially identical to those of method 700 (Figure 7) are indicated with values incremented by 100. Step 820 of method 800 includes browsing trough the whole possible range of the caR parameter and find a value of caR parameter that results in a minimum value of RG. The same is valid for the caB parameter. However, in certain situations, such a global search is not directly allowed since it enables direct visible effects of LCA correction with various caRB parameters in the image.

Since often a large range of values of caRB parameters is scanned, larger visible LCA errors can be introduced to be able to measure &RG and BG for these values as well. Also, in case of a video signal, the caRB parameter is possibly changed every field/frame, introducing temporal changes of an LCA artifact. In many cases this is not allowed, since these effects are visible in the image.

One solution to this problem is, in modification of the method shown in figure 1 , building a parallel image processing pipeline 910 as shown in Figure 9, comprising steps 3', 4' and 5' (cf. Figure 1 ) and a step 800' (in principle corresponding to step 800 of Figure 8). Image processing pipeline 900 is, however, used only for measurement and its effects are not visible in the image.

The main advantage of this approach is a possibility to find the most optimal values of caRB parameters which result in a real minimum value of &RG and ABG measurements without the danger of getting trapped in a local minimum value as might be possible with the feed-back loop approach 700.

Advantageously, the two approaches 700 and 800 are combined as shown in Figure 9. A global minimum search 800 in a parallel pipeline 900 can be used to perform a coarse search resulting in a coarse minimum value 910, probably also slower and/or using a less complicated color separator. In a feed-back loop 50, a value of caRB will then be optimized around this coarse working point.

Finally, to minimize global minimum search hardware used in method 800 or parallel pipeline 900, it is possible to perform global search in a slower rate (for instance once per second) and mask a current field/frame where visible LCA errors can occur by repeating the previous field/frame.

Claims

1 . A method for detection (5) and correction (3) of a lateral chromatic aberration shift of at least one color component (R,G,B) in digital image data (1 ) comprising pixels (x,y), the method including for at least one pixel (x,y): a) providing a shift function (f(r)) describing a shift of the at least one color component (R,G,B) depending on at least a variable shift parameter (caRB) and a position (r) of the pixel (x,y) in the digital image data (1 ), b) determining (50,700,800) an optimum value of the variable shift

parameter (caRB) suitable to minimize the shift of the at least one color component (R,G,B), and c) applying (3) a correction function (outRB(x,y)) based on the shift function (f(r)) including the optimum value of the variable shift parameter (caRB) to the image data (1 ) to generate corrected image data.

2. A method according to claim 1 , wherein step b) includes: i) setting the variable shift parameter (caRB) to an initial value, ii) applying (3,703) a correction function (outRB(x,y)) based on the shift function (f(r)) including a current value of the shift parameter (caRB) to the digital image data (1 ), iii) measuring (5,705) the shift of the at least one color component (R,G,B) after applying (3,703) the correction function (outRB(x,y)) and setting the variable shift parameter (caRB) to a modified value based on the results of the measurement (5), and iv) repeating steps ii) and iii) until the results of the measurement (5) in step iii) satisfy predifined criteria and using the current value of the shift parameter (caRB) in defining the optimum value of the variable shift parameter (caRB) if the results of the measurement (5) in step iii) satisfy the predifined criteria.

A method according to claim 2, wherein measuring (5,705) the shift of the at least one color component (R,G,B) after applying (3,703) the correction function (outRB(x,y)) in step iii) is limited to certain regions of the digital image data and/or depending on a measure of an edge of at least one color component (R,G,B).

A method according to claim 1 to 3, wherein step b) includes: aa) defining a number of possible values of the variable shift parameter (caRB) from a range of values, bb) applying (820) a number of correction functions (outRB(x,y)) based on the shift function (f(r)), each of the number of correction functions (outRB(x,y)) including one of the possible values of the variable shift parameter (caRB), to the digital image data (1 ), and cc) determining which of the number of correction functions (outRB(x,y)) yields a minimum shift of the at least one color component (R,G,B) in the digital image data (1 ) and using the value of the shift parameter (caRB) used in that correction function (outRB(x,y)) in defining the optimum value of the variable shift parameter (caRB).

A method according to claim 4 when depending from claims 2 or 3, wherein steps i) to iv) and steps aa) to cc) are performed in parallel and/or wherein using steps aa) to cc) a coarse optimum value of the shift parameter (caRB) is obtained which is optimized in steps i) to iv) to result in the optimized shift parameter (caRB).

A method according to one of the preceding claims, wherein the correction function (f(r)) is a 3^rd order polynomial of the form: f{r) = caRB₃ * r³+caRB₂ *r² +caRB-_\ * r¹ + caRB₀, or a 1 st order polynomial of the form: f{r) = caRB*r, wherein (r) represents a position of the at least one pixel (x,y) in form of a radial distance from a center (of an image and/or an image sensor, (caRB) are parameters for the red (caR) and blue (caS) color components or planes and caRB₀ is a parameter relating to a lateral chromatic aberration at the center (Xc, Yc) of the image and/or image sensor.

A method according to claims 2 to 6, wherein measuring (5,705) the shift of the at least one color component (R,G,B) includes calculating inter-color differences {ARG,ABG,ARB) from intra-color differences (Δ(/?),Δ(Θ),Δ(β)) between the at least one color component (R,G,B) and at least a further color component (R,G,B) of the digital image data.

A method according to claims 2 to 7, the method being embedded into an image processing pipeline including preprocessing (2) and further processing (7) of the digital image data (1 ).

Image processing device for detection (5) and correction (3) of a lateral chromatic aberration shift of at least one color component (R,G,B) in digital image data (1 ) comprising pixels (x,y) supplied to the image processing device, adapted to perform, for at least one pixel, a method according to any one of the preceding claims, the image processing device including: means for determining (50,700,800) an optimum value of a variable shift parameter (caRB) of a shift function (f(r)) describing a shift of the at least one color component (R,G,B) depending on at least a variable shift parameter (caRB) and a position (r) of the pixel (x,y) in the digital image data (1 ), suitable to minimize the shift of the at least one color component (R,G,B), and means for applying (3) a correction function (outRB(x,y)) based on the shift function (f(r)) including the optimum value of the variable shift parameter (caRB) to the image data (1 ) to generate corrected image data.

10. Computer program product adapted to be executed in an image processing device according to claim 9 and adapted, when executed, to perform a method according to any one of claims 1 to 8.