EP1974541A2

EP1974541A2 - Method for processing a digital image, in particular for processing contour regions, and corresponding device

Info

Publication number: EP1974541A2
Application number: EP07718101A
Authority: EP
Inventors: Fritz Lebowsky; Yong Huang
Original assignee: STMicroelectronics SA; STMicroelectronics Asia Pacific Pte Ltd
Current assignee: STMicroelectronics SA; STMicroelectronics Asia Pacific Pte Ltd
Priority date: 2006-01-17
Filing date: 2007-01-17
Publication date: 2008-10-01
Also published as: US20110129146A1; FR2896369A1; US8457436B2; WO2007083019A2; WO2007083019A3

Abstract

The invention concerns a device for processing a digital image comprising at least one contour region, including sharpness processing of the contour region. The sharpness processing includes converting data of the level of pixels of the contour zone into initial main data (step 2), ranging between a minimum value, for example 0 and a main value based on the amplitude of the contour, a sharpness sub-processing performed on said initial main data so as to obtain final main data (steps 3 to 10), and converting the final main data into final data of levels (step 11).

Description

Method of processing a digital image, in particular the treatment of the contour areas, and corresponding device

The invention relates to the processing of digital images, in particular the treatment of the contour areas of the image so as to improve the sharpness thereof.

The invention applies advantageously but not exclusively to the processing of the blur effects of a given image, for example by an implementation within a software tool for processing blur.

The invention applies advantageously but not exclusively to applications requiring a modification of the digital image, implementing for example a method of linear processing of a sampled digital signal, such as a modification of the size of the image by Linear interpolation.

For example, in the case of digital video applications, it is often necessary to resize it according to the display standard used. These resizes alter the sharpness of the outline areas, and thus the quality of the displayed image. The invention can be applied in particular for the display of digital images on large screens, such as plasma or LCD screens. For example, to display an image initially in VGA format ("Videographics Adapter", in English language) of dimensions 640 x 480, on a screen accepting an HDTV type broadcasting format, of dimensions 1920 x 1080, a vertical and horizontal magnification image is needed.

Conventionally, to limit the visual artifacts altering the contours of the processed image, in particular because of the approximations resulting from the use of linear method, a post-filtering (by example for performing sharpness processing as described in US Patent 6,847,738 in the name of Philips) can be realized, particularly at the contour areas.

However, post-filtering can attenuate one type of artifact (for example the appearance of halos, using a linear compensation filter) but accentuate another one.

In addition, the use of an additional filter adds constraints on the opening of the signal sampling window, which can not be as narrow as desired, which accentuates the degradation of the processed image.

An object of the invention is to improve the sharpness of the contours, in particular after it has been digitally processed.

For this purpose, according to one aspect of the invention there is provided a method of processing a digital image which comprises at least one contour area, comprising a sharpness treatment of the contour area.

According to a general feature of this aspect of the invention, the sharpness processing includes converting the pixel level information of the outline area to initial key information between a minimum value, for example 0, and a main value. depending on the amplitude of the contour, a sub-processing of sharpness performed on these initial main information so as to obtain final main information, and a conversion of the final main information into final level information.

According to one embodiment, the main value corresponds to the amplitude of the contour, that is to say the difference between the maximum amplitude of one pixel and the minimum amplitude of another pixel. This embodiment has the advantage of limiting the number of operations to be performed.

Preferably, the main information is normalized, and corresponds to the amplitude of the contour divided by this same amplitude. The main information therefore has a unit value.

More particularly, there is provided a method of processing a digital image which comprises at least one contour zone, comprising a step of processing the contour zones.

Said contours zone processing step comprises for each treated contour zone and for a set of selected pixels, a first step of converting the pixel level information into a main information, a processing step of this main information function. image processing, so as to obtain new main information, and a second step of converting the new main information into another pixel-level information.

In other words, the level information, or amplitude, of the pixels is converted into a main information, in particular a standardized one for which a set of operations specific to it is defined.

Indeed, to determine a main information of the pixel, corresponds to determining a phase information of the pixel. Since this is a periodic function, it has very specific properties. In particular, each image transformation results in a modification of the easily identifiable phase, which has the advantage of not generating information loss, unlike the operations carried out on the amplitude of the pixels, by the methods linear current contour processing, because of the necessarily necessary approximations.

Therefore, to deal with the contour areas of the image, the invention takes a completely different approach from the existing methods, since the processing operations are performed not on amplitudes, but on phase values.

According to one embodiment of the invention, the sub-processing of sharpness comprises for each pixel located inside the outline area, the determination of an intermediate information, and the subtraction of the initial main information. maximum, of this intermediate information, so as to obtain said final main information.

According to one embodiment, the step of processing the edge zones comprises the detection of said contour zones in the initial digital image, by locating for each dimension of the initial digital image, the level information values, by for example, the local minimum, maximum and maximum amplitudes of the pixels, the succession of two pixels respectively having a minimum and maximum amplitude reflecting the existence of a contour zone. According to one embodiment, said main quantity is formed from the difference between the maximum and minimum level information of two successive pixels delimiting said contour zone.

According to one embodiment, the conversion step comprises for each contour, the normalization of the values of the minimum and maximum level information of the pixels delimiting said contour, and pixels of the contour zone, as a function of said difference between the maximum and minimum level information of two successive pixels. According to one embodiment, the processing step can be carried out further, on the sum of the main information of the set of pixels strictly within the contour area.

According to one embodiment, the contour processing phase may comprise the generation of sub-pixels within the contour zone.

According to another embodiment, the digital image may be a color image, for example an image comprising three components according to the RGB format. During this implementation mode, for each contour

the amplitude of the contour is determined from the contour of each component of the color image, and

conversions and sub-processing of the edge area sharpness processing are performed for each component of the color image.

This mode of implementation has the advantage:

to minimize the appearance of noise on the contours,

- to reduce the phase errors at the contours.

For example, according to the aforementioned embodiment, one can determine the amplitude of the contour, taking into account also the direction of variation of the contour of each component.

More precisely, the determination of the amplitude can comprise for each contour:

a multiplication of the contour of each component of the image by the associated direction of variation,

an addition of the multiplied contour of all the components of the image so as to obtain a modified contour,

the calculation of the amplitude of the contour by making the difference between the value of the level information taken by the modified contour at the beginning of the contour and the value of the level information taken by the modified contour considered at the end of the contour.

According to another aspect of the invention, there is provided a device for processing a digital image which comprises at least one contour zone, comprising means for processing sharpness of the contour zone.

According to a general characteristic of the invention, the sharpness processing means comprises first conversion means able to convert pixel level information of the contour zone into initial main information, between 0 and a main value depending on the amplitude of the contour, sharpness sub-processing means able to perform a sharpness treatment on these initial main information so as to obtain final main information, and second conversion means able to convert the final main information into information. final level information.

More particularly, it is proposed, a device for processing a digital image which comprises at least one contour zone, comprising means for processing the contour areas. Said means for processing the contour zones comprise, first conversion means capable of for each processed contour area and for a set of selected pixels, converting the pixel level information into an initial main information, processing means adapted to process this initial main information according to the image processing, in order to obtain a new main information, and second conversion means able to convert the new standardized information into another pixel-level information. According to one embodiment, the sharpness sub-processing means comprise suitable determination means for each pixel located inside the contour zone, to determine an intermediate information, and subtraction means capable of subtracting this information. intermediate of the initial maximum main information, so as to obtain said final main information.

According to another embodiment, the digital image may be a color image, in particular an image comprising three components according to the RGB format. In that case,

the means for processing the sharpness of the outline area

(MTRC) may further comprise contour amplitude determining means (MCamplRGB) capable of determining the amplitude of the contour from the contour of each component of the color image, and

the first and second conversion means, and the sharpness sub-processing means can operate on each component of the color image.

For example, the means for determining the amplitude of the contour (MCamplRGB) can determine the amplitude of the contour, taking also into account the direction of variation of the contour of each component.

More precisely, said means for determining the amplitude of the contour may comprise multiplication means capable of multiplying the contour of each component of the image by the associated direction of variation. addition means capable of adding the multiplied contour of all the components of the image so as to obtain a modified contour,

- Contour amplitude calculation means able to make the difference between the value of the level information taken by the contour modified at the beginning of the contour and the value of the level information taken by the modified contour considered at the end of the outline.

According to another aspect of the invention, there is provided a display system, in particular a plasma screen, comprising a device as mentioned above.

The invention can also be applied to LCD type screens or any other type of screen. The invention can be integrated within a software tool, in particular for image processing, for example Adobe Photoshop®.

Other advantages and characteristics of the invention will appear on examining the detailed description of an embodiment of the invention which is in no way limiting and the attached drawings in which: FIG. 1 represents an embodiment of a system According to the invention, FIG. 2 represents an embodiment of a processing line according to the invention, FIG. 3 represents another embodiment of a processing line according to the invention, FIG. implementation of a method according to the invention, FIG. 5a shows an example of contour after an enlargement, FIG. 5b represents a representation mode of the normalized magnitudes of the pixels according to the invention, FIGS. 6 and 7 illustrate a numerical application of the embodiment of FIG. 4. FIG. 8 illustrates another embodiment of a method according to the invention, FIGS. 9 and 10 illustrate a digital application of the embodiment of FIG. 8, FIG. 11 more precisely illustrates a mode of implementation. embodiment of a processing chain according to the invention, when the digital image is a color image, FIG. 13 illustrates more precisely the first embodiment of a processing chain according to the invention. step of an embodiment of the method according to the invention, in the case of a color image, FIG. 14 represents an example of the contours of each component of a color image of RGB type, FIGS. 15 and 16 more particularly illustrate a means capable of determining the amplitude of a contour of a color image, FIG. 17 more particularly illustrates a means of a processing line according to the invention, this means being suitable to intervene on the outlines of a color image, prior to their re-insertion into the image.

Referring to Figure 1. The reference ECR designates a screen, for example a plasma type screen, comprising a CTR processing chain of a digital image. The latter receives as input an IM-IN-ech sampled input image delivered by MECH sampling means from an IM-IN digital image.

The processing chain CTR outputs an IM-OUT image to display means MAFF able to display the output image IM-OUT on a display matrix MAT of the screen ECR.

Referring now to Figure 2 which illustrates one embodiment of the CTR processing chain.

This comprises MTR processing means of the sampled input digital image IM-IN-ech. For example, these processing means may be means for enlarging or reducing the image, means for increasing or decreasing the resolution of the image, or means for processing the fuzziness effects of the image. picture.

Of course, the processing means are not limited to the aforementioned means.

The CTR processing chain also comprises MTRC contour processing means implementing a method according to the invention.

For this, the CTR processing chain further comprises MDET contour detection means.

The latter receive the IM-IN-ech sampled image as input and calculate the local minimums and maximums for the various dimensions of the image. This detection of minimum and maximum can be carried out according to a conventional method known in by those skilled in the art, for example that described on the site: "http://en.wikipedia.org/wiki/Extremum_local".

The succession of two pixels respectively having a local minimum and local maximum amplitude (or vice versa) reveals the presence of a contour in the image, that is to say a separation between two distinct areas of the image.

These local minimum and maximum amplitudes are delivered by means of MTRC contour processing. They output the processed contours to insertion means MI. The MI insertion means also receive as input the image delivered by the MTR processing means.

The processing chain CTR may also comprise MFIL filtering means, for example image smoothing means delivered by the insertion means MI. In addition, the processing means may comprise additional processing means MTRS of the image, for example means of reducing the image if the processing means are magnification means.

The additional processing means output the IM-OUT image.

Of course, just as for the MTR processing means, the MFIL filtering means and additional processing of the MTRS image are possible examples of post-processing, given for information only. Moreover, it is possible to directly exploit the image delivered by the insertion means MI.

Referring now to Figure 3 which illustrates another embodiment of the CTR processing chain. In this example, the detection means MDET and the MTRC contour processing means receive in input either the sampled initial image IM-IN-ech but directly the image processed by the MTR means. Indeed, the pixels of the edges of the untreated image are indexed by conventional addressing methods well known to those skilled in the art, which make it possible to reference the pixels of the output image with respect to the image. input.

In the following description, unless otherwise indicated, it is considered that the main information that will be discussed below, are standardized, that is to say between 0 and 1.

Referring now to Figure 4 which illustrates an embodiment of the method according to the invention in the particular case where an enlargement of a digital image is carried out, or when increasing the resolution of the image.

The magnification can be done with an amplification factor (or gain) integer or not integer.

Each contour is detected according to a dimension of the image, for example a line, a column, or another dimension in the case of a 3-dimensional or larger image.

What will be described now applies for each of the dimensions of the image.

A first step (step 1) consists in calculating the norm associated with the treated contour. This must be recalculated for each processed contour.

This corresponds to the difference (principal value) between the maximum and minimum amplitudes of the pixels delimiting the contour. This difference is called contour amplitude. During a second step (step 2), the pixels of the outlines are normalized.

This operation returns, for each pixel of the contour, to subtract from the amplitude of said pixel, the minimum value of the pixel limiting the contour, and to divide the result by the norm calculated in step 1.

This operation is equivalent to calculating a phase value (initial principal information) for the pixel considered, according to the relation below:

Or :

Φ _GA (ÎI) represents the phase of the pixel at the position n within the contour,

CA (n), CA (min) and CA (max) respectively represent the amplitudes of the contour pixel located at the position n, the pixel having the minimum amplitude, and the pixel having the maximum amplitude.

FIG. 5a illustrates an exemplary rising edge, comprising a pixel PXm of minimum amplitude, two intermediate pixels PXI1, PXI2, and a pixel PXIM of maximum amplitude. The different amplitudes CA (min), CA (n), CA (n + 1) and CA (max) of these pixels are represented by a circled black dot.

The SPX pixels in light gray are those added by the processing (here the magnification) and will be called subpixels.

Moreover, performing the normalized phase / amplitude transformation can be represented on a unit circle as illustrated in FIG. 5b. The amplitude of the contour (CA (max) -CA (min)) corresponds to the circumference of the unit circle, and the phase associated with 2π. The amplitude CA (n) -CA (min) corresponds to the portion of circle associated with the phase φc _A (n). Referring again to Figure 4.

During a step 3, the new amplitude (intermediate information) of each pixel strictly inside the contour is calculated by multiplying its normalized amplitude by the gain G of the enlargement.

Indeed, there exists between the phase of a pixel of a given image, and this same pixel in the enlarged image, a relation: φ ' _CA (n) = G * φ _CA (n) (2), where φ'c _A ( _n) is the phase of the pixel in the enlarged image, and φc _A ( _n ) is the pixel phase of the initial image.

Then, in order to obtain amplitudes for the subpixels generated by the enlargement of the image (or the increase of the resolution), we compare (step 5) the new amplitude of each pixel CA (k) ( k successively taking the value of the index of the processed pixel) with a threshold α equal to the amplitude of the normalized contour (that is to say 1, in this case) multiplied by an index m, the latter having been initialized to 1 during a step 4.

If the new amplitude of the CA (k) pixel is greater than the threshold α, the value 1 is assigned to the sub-pixel (step 6), and 0 otherwise (step 7). In the case where one does not consider a normalized amplitude, one affects the amplitude of the contour that is to say (CA (max) - CA (min)) to CA (k + m).

The value of m is incremented (step 8) and steps 5 to 8 are repeated as long as m is strictly less than the gain. During a step 9, the amplitude of each pixel that was amplified during step 3 is shifted so that it is found within the normalized range (here between 0 and 1). This operation amounts to performing a "modulo" operation on the amplitude CA (k), which is possible because of the relation between a normalized amplitude and a phase.

Finally, the amplitudes obtained for all the pixels of the contour and of the sub-pixels are sorted according to the direction of variation thereof (step 10).

Finally, we find the real amplitudes (final level information) of each pixel and sub-pixel by performing the reverse process of step 2 (step 1 1). In other words, the values obtained in the previous step are multiplied by the norm, here the amplitude of the contour.

More precisely, for a rising edge, if CA (in) corresponds to the real amplitude of a pixel at the position n inside a contour, by performing the normalization operation, it comes: CA (nl ) <- (CA (in) - CA (il)) / contour_contrast with contour_contrast = abs (CA (kl) - CA (il)) and with ni = (il + l), ..., (kl -l) , nor corresponding to the position of a pixel inside the considered contour, where, where <kl, it and kl being two integers correspond to the positions of the pixels having respectively a minimum and maximum amplitude and delimiting the contour,

CA (il) represents the minimum amplitude of the contour delimiting pixel, of index ll, CA (kl) represents the maximum amplitude of the contour delimiting pixel, of index kl,

The width of the contour 1 is therefore equal to:

1 = (kl - il - 1), in number of pixels. The new amplitude CA 'of the index pixel ni in the initial image is equal to:

CA '(nl * g) = gain * (CA (nl) -CA (il)); where: gain represents the amplification factor, g represents the entire portion of the gain, n1 * g represents the new index in the processed image of the index pixel and in the initial image.

The amplitude CA "of the sub-pixels added by the image processing is determined by the relation:

C * rCAfld) ifCA '(nl * g)> = m * (CA (kl) -CA (il)) gm; | _{CA (il)} jfCA '(n1 * g) <m * (CA (k1) -CA (il)) with m = 1, ..., (g-1).

We shift the new amplitude CA (nl) that we had multiplied by the gain, so that it is included in the normalized difference, that is to say that we apply a modulo operation, in term phase:

CA "(n1 * g) = modulo {CA '(n1 * g), (CA (k1) - CA (II))} + CA (II), where CA" (n1 * g) denotes the normalized amplitude of the pixel strictly inside the contour, index n * g.

The magnitudes of the contours are sorted in the ascending order between the minimum and maximum values, so as to obtain for each index the final amplitude C August: CAout (nl * g + m0) ≈ tri {CA (nl * g + m0)} asc with mO = 0.1, ..., (g-1).

Similarly, for a falling edge, if CA (in) corresponds to the real amplitude of a pixel at position n inside a contour, by carrying out the normalization operation, it comes: CA (n2) + - (CA (in) - CA (k2)) / contour_contrast with contour_contrast = abs (CA (k2) - CA (i2)) and with n2 = (i2 + 1), ..., (k2 -l), n2 corresponding to the position of a pixel within the contour considered, i2 <k2, i2 and k2 being two integers corresponding to the positions of the pixels having respectively a maximum and minimum amplitude and delimiting the contour,

CA (k2): minimum amplitude of the pixel delimiting the contour, of index ll, CA (i2) denotes the maximum amplitude of the pixel delimiting the contour, of index kl.

The width of the contour 1 is therefore equal to: 1 = (k2 - i2 - 1), in number of pixels.

The new amplitude CA 'of the index pixel ni in the initial image is equal to;

CA '(n2 * g) = gain * (CA (n2) -CA (k2)); where: gain denotes the amplification factor, g the integer part of the gain, n2 * g denotes a new index in the processed image of the index pixel n2 in the initial image.

with m = 1, ..., (g-1).

The new amplitude CA (n2) is shifted so that it is included in the normalized deviation, that is to say that a modulo operation is applied in terms of phase: CA "(n2 * g) = modulo {CA '(n2 * g), (CA (i2) - CA (k2))} + CA (k2), where

CA "(n2 * g) designates the normalized amplitude of the pixel strictly inside the contour, of index n * G. The amplitudes of the contours are sorted in ascending order between the minimum and maximum values, so that obtain for each index the final amplitude C August:

CAout (n2 * g + mO) = tri {CA "(n2 * g + mO)} with mO = 0, 1, ..., (g-1) Referring now to FIG. a numerical example of the embodiment described in FIG.

In this example, the image has been enlarged by a gain equal to 3. The strict interior of the outline area here comprises a single pixel, and two sub-pixels will be generated by the enlargement.

At the end of normalization, one finds oneself in step 2 where the amplitudes CA (min), CA (n) and CA (max) are respectively the values of 0; 0.65 and 1.

For step 3, the amplitude CA (n) is multiplied by the gain, that is to say 3. Its new value is equal to 1.95.

Then we proceed to step 5 where the value of m is equal to 1. As 1, 95 is greater than 1, we assign to the amplitude CA (n) 1 of the first sub-pixel, the value of 1.

Then we repeat the same operation with the value of m = 2. Since 1, 95 is less than 2, the amplitude CA (n) 2 of the second sub-pixel is assigned the value of 0.

During the ninth step (step 9), the operation of Modulo is carried out where the value 1 is subtracted, that is to say the difference between CA (max) and CA (min), of the amplitude CA (n) of the contour pixel, so that its amplitude is between 0 and 1.

Finally, in step 10, the set of standardized amplitudes obtained, here in ascending order, are sorted. FIG. 7 illustrates exactly the same type of treatment, that is to say an enlargement by a factor of 3, but in the case where initially the contour comprises two intermediate pixels, here having a standardized amplitude of 0, 85 for CA (n) and 0.35 for CA (n + 1). Two sub-pixels are added per pixel strictly within the contour area, i.e. 4 subpixels in all in this example.

Moreover, a decreasing contour is considered here.

Steps 2 to 9 are identical to those described in FIG. 6, but repeated for each of the pixels of the contour zone CA (n) and

CA (n + l).

Therefore, the sorting step 10 is performed on all the amplitudes obtained, here six pixels, and in descending order, since we consider here a descending edge. Referring now to Figure 8, which illustrates another embodiment of a method according to the invention, particularly suitable in the case of an increase, a reduction or a change in resolution of the image.

In this embodiment, during step 3, not every normalized amplitude of the pixels situated strictly within the contour zone is considered, but the sum of these normalized amplitudes is operated.

Therefore the comparison of step 5 is made with this sum. In addition, the stopping criterion at the end of step 8 is not the comparison between the index m and the gain but between the index m and the variable 1, which corresponds to the number of pixels strictly at 1. inside the contour at the end of the treatment. Furthermore, the modulo operation of step 9 is performed on the sum calculated during step 3.

Referring now to Figure 9 which gives a numerical example of the embodiment shown in Figure 8.

A downward contour with two intermediate pixels having normalized amplitudes of 0.85 for CA (n) and 0.35 for CA (n + 1) respectively is considered here.

The processing performed here is again an enlargement with a gain of 3, 4 sub-pixels are generated for the contour, in addition to the initial 2 intermediate pixels. Therefore, in step 3, the sum of the amplitudes equal to 0.85 and 0.35, multiplied by the gain of 3, is assigned as the new amplitude to the CA (n) pixel. The resulting amplitude is then equal to 3.6. Then, as already done previously, steps 4 to 8 are performed as long as the index m is strictly less than the number of pixels 1 inside the contour zone.

During step 9, the Modulo operation is carried out on the amplitude calculated during step 3, so that it is again normalized, that is to say equal to 0.6 in that case.

Finally, during step 10, the set of values obtained is sorted.

Reference is now made to FIG. 10 which illustrates a numerical example of the embodiment of FIG. 8, in the particular case where the gain is equal to 1. This example corresponds to an image processing where neither the format nor the resolution of the image is changed. For example, the processing may be that of the blur effects of a digital image.

This numerical example corresponds to a downward contour comprising two intermediate pixels CA (n) and CA (n + 1) having, respectively, as standardized amplitudes, the values of 0.85 and 0.35.

At the end of step 3, the amplitude allocated to CA (n) is equal to 1, 2, that is to say the sum of the normalized amplitudes of CA (n) and CA (n + 1) to step 2.

The comparison made during steps 4 to 8 makes it possible to assign the value of 1 to pixels having the amplitude CA (n + 1).

Steps 9 and 10 are carried out in the same manner as previously described. Referring now to Figure 1 1 which shows in more detail an embodiment of the CTR processing chain capable of implementing a method according to the invention.

The MCTR contour processing synthesis means comprise a first MCamp block capable of calculating the amplitude of the contour considered from the local minimum and maximum amplitudes detected by the MDET means.

MClarg means are able to calculate the size or width of the contour, that is to say the number of intermediate pixels within this contour. Addressing means MCadr memorize for each contour its location within the IM-IN-ech input image.

Finally, MDpol means are able to detect the polarity of the contour, that is to say if the contour varies from a minimum amplitude to a maximum amplitude (rising edge) or a maximum amplitude at a minimum amplitude (falling edge).

The MCTR edge processing means also comprise memory means MM capable of storing the value of the gain gain gain. Calculation means MCAL implement, from the data delivered by the means

MCampl, MClarg, MCadr and gain gain gain, a mode of implementation of the method according to the invention, for example steps 4 to

8 of the examples illustrated in FIGS. 4 or 8. The new normalized amplitudes are delivered to MDEC offset means capable of performing the modulo operations of the outlinear pixel amplitudes (step 6 of FIG. 5).

All the amplitudes of the pixels of the contour are delivered to MTRI sorting means which classifies them according to the data delivered by the MDpol means, that is to say if the contour is upward or downward.

The MTRI sorting means also determines the final amplitudes by reversing the normalizing operation.

The new synthesized amplitudes are delivered to the insertion means MI which, as a function of the addresses of the contours delivered by the MCadr means, insert them into the processed image, according to a method well known to those skilled in the art.

The contour processing means may also comprise MInt interpolation means capable of controlling the insertion of the contours carried out by the MI means in the case where the processing carried out by the MTR means comprises the application of a non-integer gain. Furthermore, the MCAL means can perform all the operations of an embodiment of the method according to the invention only in a preferred dimension of the image.

This preferred dimension is chosen as a function of the number of pixels inside the processed contour.

Then, it is possible to determine the amplitude of the pixels added by the processing means MTR in the other dimensions of the image, directly from the amplitudes calculated for these pixels in the preferred dimension. Indeed, for the pixels added by the processing means, there is a relationship between the phases calculated for the preferred dimension and that of the other dimensions.

For example, for a two-dimensional image x and y, and for a processing implementing a gain, it comes:

Where: spp _y : position of the pixel added by the processing within the contour, φcoy '• slope of the contour, φc _A (n _x ) represents the phase of the pixel at the position n _x within the contour, for the initial image ,

Φ ' _CA (H _X ) represents the phase of the pixel at the n _x position within the contour, for the processed image. Reference is now made to FIG. 12, which shows an embodiment of the CTR processing chain in the case of processing a digital color image. The embodiment illustrated in FIG. 12 is given solely by way of example, the CTR processing chain being adaptable to any processing performed on the color image.

Conventionally, in the context for example of the processing of a television signal, the digital image is represented by two chrominance components and a luminance component.

The image is processed in this format to improve the luminance and chrominance quality. Then the digital image is converted into the so-called RGB format ("Red, Green, Blue") before being projected onto the television screen.

However, the image obtained in the RGB format reveals many artifacts in the colors of the image such as: a change in the colors along the contours, strong contrast variations on either side of the contours; contrast errors on the pixels of the contour, these errors being related to a bad position of the contour because of the quantification of the amplitude of the pixels during the conversion of the image to the RGB format. The CTR processing chain, an embodiment of which is shown in FIG. 12, makes it possible to solve these problems. This processing chain has the advantage of preserving all the information of important phases on the colors of the outline. This treatment chain avoids - distorted variations of contrasts, and obtaining visibly false colors at the contours.

The processing chain presented in FIG. 12 shows the elements of the processing chain illustrated in FIG. However, because of the three RGB components of the new format of the color image, the sampled image is transmitted via a bus comprising three transmission channels, one for each component of the image. Therefore, each module of the CTR processing chain performs the previously described processing for a black and white image, for each RGB component of the color image.

The determination means MDETRGB are adapted to the color image. At the end of the detection of the minima and maxima of the image, they deliver the outlines for each component of the image.

These contours are referenced Cr, Cg, Cb, respectively for the red, green and blue components. Each contour comprises the pixels belonging to the contour as well as their amplitude, for the component of the image considered.

Moreover, the determination means MDETRGB deliver the ranks of the pixels n _m j _n and n _max of the two pixels bounding the contour considered.

The means for determining the direction of variation of the MDpol contours deliver three variables POLR, POLG and POLB. Each variable indicates the direction of variation of the contour (also called the polarity) respectively of the red component, the green component and the blue component of the image.

The means for determining the amplitude of the contours of the image MCamplRGB receive the variables Cr, Cg and Cb, as well as the ranks of the pixels n _m i _n and n _max .

They also receive the polarities POLR, POLG, POLB contours of each component of the image. From these different variables, the means for determining the amplitude of the MCamplRGB contours calculate the amplitude of the CA contour of the digital image in colors.

The MTRIRGB means are sorting means adapted to color images, in particular of the RGB type. They will be described more precisely below.

The MTRIRGB sorting means receive at input the variables delivered by the means MInt, MDpol, MCadr, MClarg, as well as the amplitude CA determined by the MCamplRGB means. The MTRIRGB means also receive the amplitude of the edge pixels for each RGB component, these amplitudes having been calculated by the MCAL means and then shifted by the delay means MDEC.

The MCamplRGB means also deliver to the MTRIRGB sorting means:

the amplitude of the contours for each component CA (r), CA (g) and CA (b), before the computation performed by the means MCAL,

- the lower limit of the contour for each component CA (rmin), CA (gmin), CA (bmin). Sorting means MTRIRGB deliver to the insertion means

MI, contours SR, SG and SB for each RGB component.

FIG. 13 illustrates a mode of implementation of the method of determining the amplitude of the CA contour for a color image of RGB type.

These steps are substituted for step 1 of FIGS. 4 and 8.

First, the method, according to this embodiment, comprises a step of calculating the polarity of the POLR, POLG and POLB contour for each red, green and blue component (step 100). This step is performed by the polarization determination means MDpol.

For example, if the outline is ascending for the red component of the image, the variable POLR takes the value +1. Then, in order to take into account the direction of variation of the contour, each contour Cr, Cg, Cb is multiplied by the corresponding polarity (step 101). Thus, the contour Cr is multiplied by the polarity POLR.

Then, a modified contour C (n) is determined which is equal to the sum of the contours Cr, Cg and Cb multiplied by the corresponding polarity (step 102).

In other words, one can write for each pixel n:

C (n) = Cr (n) • POLR + Cg (ή) • POLG + Cb {n) • POLB Finally, we calculate the amplitude of the contour CA by subtracting the value taken by C (n) to the pixel referenced n _m i _n to the value taken by C (n) at the referenced pixel n _max . The amplitude of the CA contour is equal to the absolute value of this difference (step 103).

The various steps mentioned above are illustrated by an example. In this example, we consider that the contour Cb is described by the following equation:

where n is the rank of the pixel inside the contour. In the same example, the outline on the green component of the image is represented by the following equation:

We consider that the outline is null on the red component of the image: Cr (n) = 0

FIG. 14 represents these different contours for the green and blue components, respectively by the Cg (n) and Cb (n) curves. The black dots on the curves represent the different pixels of the image. The phase of the contour corresponds to the difference between the beginning of the contour (pixel of rank n _m i _n ) and a point located on the contour.

The variables CA (g) and CA (b) represent the amplitude of the contour respectively on the green and blue components. According to this example, the polarity of the outline on the blue component Cb (n) is equal to +1 (POLB + 1). The polarity of the outline on the green component of the image is equal to -1 (POLG = -1).

The value of the modified contour C (n) is equal in this example:

C (n) = Cb (n) * POLB + Cg (n) * POLG = Cb (n) * 1 + (l - Cb (n)) * (- 1) = 2 * Cb (n) -l The curve C (n) shown in Figure 14 illustrates the variation of the modified contour.

Finally, the amplitude of the CA contour is calculated according to the following expression:

IT ) J. Another possibility would be to calculate C (n) by considering the sum of the absolute values of the outlines on each RGB component:

C (n) .

FIG. 15 illustrates an embodiment of the MCamplRGB amplitude determining means so as to implement the various steps described above. The MCamplRGB amplitude determining means comprises modified MSOM contour determination means.

The latter are able to receive the components Cr (n), Cg (n) and

Cb (n) and to deliver the modified contour C (n) which is a function f of the aforementioned RGB components.

C (n) = [Cr (n), Cg (n), Cb (n)].

The MCamplRGB means also comprise MDIF calculation means able to calculate | C (n _max ) - C (n _m i _n ) |.

The calculation means MDIF deliver the amplitude of the contour CA.

More precisely, as illustrated in FIG. 16, the means

MSOM comprises a MULR multiplier able to multiply the Cr (n) component and the POLR polarity. They also include another MULG multiplier to multiply the Cg (n) component with the POLG polarity.

An SOMG adder adds the results from the two multiplications performed by the MULG and MULR multipliers.

Finally, the MSOM means comprise a third multiplier MULB for multiplying the component Cb (n) with the polarity POLB. An adder SOMB sum the result delivered by the adder SOMG and the result delivered by the multiplier MULB.

Finally, the adder SOMB delivers the modified contour C (n).

Figure 17 illustrates an embodiment of the MTRIRGB means. For the purpose of simplification, only the means intended to form the contour to be inserted of the red component SR are represented. These same means are repeated for the contours to be inserted SB and SG respectively blue and green components. MNRGB means order the pixels of the contours of each component according to the contour polarity on the considered component (step 10, FIG. 4 or 8).

Then an intermediate contour CTR (n) is developed by auxiliary means MNRGBAX. This intermediate contour CTR (n) is developed according to the contours on each component, delivered by the means of shift MDEC. At this stage, the values taken by

CTR (n) are always normalized.

An inverter X ^"1 receives the amplitude of the contour CA. As can be seen in FIG.

CA ≈ G4 (max) - G4 (min)

= CA (r max) - CA (r min) + CA (g max) - CA (g min) + CA (b max) - CA (b min)

= CA (r) + CA (g) + CA (p).

The sorting means MTRIRGB furthermore comprise: a multiplier MUL10 capable of multiplying the intermediate contour CTR (n) by the inverse of the amplitude of the contour 1 / CA, another multiplier MULI 1 capable of multiplying the polarity POLR of the contour on the component considered (here red) by the result delivered by the multiplier MUL10, namely CTR (n) / CA, - another multiplier MULl 2 able to multiply the amplitude CA (r) of the contour for the considered component, here red by the result delivered by the multiplier MULI 1, namely POLR * CTR (n) / CA, and a summator SOM 13 able to add the result delivered by the multiplier MULl 2 to the lower bound of the contour on the red component CA ( rmin). Multipliers MULlO, MULI l and MULl 2 and the summator SOM13 make it possible to perform step 1 1 (FIG. 4 or 8). Finally he comes:

Similarly, it comes for the green and blue components:

Contours SR, SG, and SB comprise the edge pixels respectively of the red, green and blue components, these pixels having a real final amplitude and no longer normalized.

Claims

A method of processing a digital image that has at least one contour area, including sharpness processing of the outline area, characterized in that the sharpness processing includes a conversion of the pixel level information of the contour area into initial key information (step 2), between 0, and a principal value depending on the amplitude of the contour, a sub-processing of sharpness performed on this initial main information so as to obtain final main information ( steps 3 to 10), and a conversion of the final master information into final level information (step 1 1).

2-Method according to the preceding claim, wherein the subtraction of sharpness comprises for each pixel located inside the contour area, the determination of an intermediate information, and the subtraction of the maximum initial main information, of this intermediate information, so as to obtain said final main information.

3. Method according to one of the preceding claims, wherein the digital image is a color image, in particular an image comprising three components (Cr, Cg, Cb) according to the format.

RGB, and in which for each contour

4-Process according to the preceding claim, wherein the amplitude of the contour is determined, taking also into account the direction of variation of the contour of each component. 5-Process according to the preceding claim wherein the determination of the amplitude comprises for each contour:

a multiplication (101) of the contour of each component of the image by the associated direction of variation,

an addition of the multiplied contour of all the components of the image so as to obtain a modified contour (102),

the calculation of the amplitude of the contour by making the difference between the value of the level information taken by the contour modified at the beginning of the contour and the value of the level information taken by the modified contour considered at the end of the contour.

6. Device for processing a digital image which comprises at least one contour zone, comprising edge area sharpness processing means (MTRC), characterized in that the sharpness processing means comprises first means conversion device (MCAL) capable of converting pixel level information of the contour area into initial main information, between 0 and a main value depending on the amplitude of the contour, means for sub-processing of sharpness suitable for performing a sharpness processing on these initial key information so as to obtain final main information, and second conversion means (MTRI) able to convert the final main information into final level information. 7-Device according to the preceding claim, wherein the sharpness sub-processing means comprise determination means suitable for each pixel located inside the contour area, to determine an intermediate information, and subtraction means ( MDEC) able to subtract this intermediate information of the initial maximum main information, so as to obtain said final main information.

8-Device according to claim 6 or 7, wherein the digital image is a color image, in particular an image comprising three components according to the RGB format and in which

the means for processing the sharpness of the outline area

(MTRC) furthermore comprise contour amplitude determination means (MCamplRGB) able to determine the amplitude of the contour from the contour of each component of the color image, and

the first and second conversion means, and the sub-processing means of sharpness are able to operate on each component of the color image.

9- Device according to the preceding claim, wherein the contour amplitude determination means (MCamplRGB) are able to determine the amplitude of the contour also taking into account the direction of variation of the contour of each component.

10- Device according to the preceding claim, wherein said contour amplitude determination means (MCamplRGB) comprise

multiplication means (MULR ₅ MULG, MULB) able to perform a multiplication of the contour of each component of the image by the associated direction of variation,

adding means (SOMG, SOMB) capable of adding the multiplied contour of all the components of the image so as to obtain a modified contour,

- Computation means (MDIF) of the amplitude of the contour able to make the difference between the value of the level information taken by the contour modified at the beginning of the contour and the value of the level information taken by the modified contour considered at the end of the contour.

11- Display system, in particular a plasma screen, comprising a device according to one of claims 6 to 10.