WO2012128376A1

WO2012128376A1 - A method of processing an image, a method of processing a video image, a noise reduction system and a display system

Info

Publication number: WO2012128376A1
Application number: PCT/JP2012/057665
Authority: WO
Inventors: Marc Paul Servais; Andrew Kay; Graham Roger Jones; Matti Pentti Taavetti Juvonen; Allan Evans
Original assignee: Sharp Kabushiki Kaisha
Priority date: 2011-03-21
Filing date: 2012-03-16
Publication date: 2012-09-27
Also published as: GB2489233A; GB201104700D0

Abstract

A method of processing an image comprises, if the part of the image has been determined to represent human skin applying a first noise reduction process to the part of the image, or, if the part of the image has been determined not to represent human skin, applying a second noise reduction process to the part of the image, the first noise reduction process giving an image with a higher noise and/or texture content than the second noise reduction process. The additional noise/texture present, after processing, in a part of the image that is determined to represent human skin effectively acts as a substitute for natural skin texture which may have been lost during the image capture and transmission process. The method may be applied to a still image, or to a sequence of images such as a video image.

Description

DESCRIPTION

TITLE OF INVENTION : A METHOD OF PROCESSING AN IMAGE, A METHOD OF PROCESSING A VIDEO IMAGE, A

NOISE REDUCTION SYSTEM AND A DISPLAY SYSTEM

TECHNICAL FIELD

This invention relates to a method of image processing for improving the perceived quality of digital TV or video without changing what is broadcast or otherwise distributed . The method may be applied to a still image, or to a sequence of images such as a video image . The invention also relates to a noise reduction system that uses the image processing method, and to a display system including such a noise reduction system . A display system of the invention may be used in devices such as, for example, a television, mobile phone, advertising hoarding, digital photograph display device , computer display, proj ector and other public or personal devices . BACKGROUND ART

Television and video content is delivered to end users in a variety of different formats . This includes analogue and digital terrestrial broadcasting, satellite broadcasting, and internet streaming. For most formats there is usually a significant loss in picture quality between the original captured video, and that which is eventually received and decoded by a TV, computer or other display device .

On analogue TV, noise appears as a time-varying random "snow" pattern or "grain" which is superimposed on the picture . The greater the noise power, the more prominent the snow or grain, and the less visible the picture. One of the main advantages of digital TV is that it allows video to be transmitted without being corrupted by channel noise in the way that that analogue TV is . (Note that very high noise levels can lead to loss in a digital signal) .

However, digital video suffers from another type of noise : compression artefacts. Because of limited bandwidth and storage, video is compressed prior to transmission . Compression artefacts are different to analogue noise, and the most common characteristics are : blocking artefacts (due to block based compression) , mosquito noise (visible around the edges of moving objects) and blurring (caused by a loss of high-frequency detail) .

One known noise reduction technique is selective blurring, which is often used to reduce both analogue and digital noise . It is usually desirable to keep the edges of obj ects sharp, but blurring can be used in the interiors of objects or regions to reduce the visibility of blocking artefacts and unwanted grain. One popular example of selective blurring that operates in this way is the Bilateral Filter (Tomasi and Manduchi, "Bilateral Filtering for Gray and Colour Images" , IEEE ICCV 1998) , and many decoders or displays use this type of filtering as a post-processing step after decoding but prior to displaying an image .

While the use of blurring to reduce noise can be effective, it can have unwanted side-effects such as removing detail. Over-smoothing of a part of an image that depicts human skin can result in it appearing artificial or plastic. This can be particularly noticeable because viewers tend to focus a large amount of their attention on any faces present within a scene .

Various noise reduction techniques have been proposed, but these generally relate to the reduction of analogue noise - and do not discuss how to handle compression artefacts introduced by the coding process:

US 4 689 666 ( 1987) proposes how a variety of noise reduction methods (and parameters) based on spatial filtering can be used for differently-coloured objects (e. g. sky, skin, and background regions) . The patent does not discuss temporal processing, but only spatial filtering.

US 6 856 704 (2005) suggests how different amounts of sharpening can be used within an image - based on the colour of the region being processed. The inventors give the example of boosting the existing texture in green grass (which may have a moderate amount of texture) while reducing the texture present in blue sky (which may have a low degree of texture) .

EP 1 775 936 (2005) and EP 1 99 1 008 (2007) propose ways of enhancing image texture in an image capture device such as a camera. The texture that is added is a function of the difference between the original captured image and a low- pass filtered version of that image . The documents propose that the algorithm can be applied to the whole image or to parts of the image such as skin tones.

The prior art listed above describes colour-adaptive noise reduction and image enhancement algorithms designed for analogue camera noise, where the input image has not been compressed and therefore does not exhibit any coding artefacts .

In "Block-based Content-adaptive Sharpness Enhancement" (IEEE ISCE2009) , Li et al present a spatially- adaptive image enhancement method which adapts to local image characteristics . They postulate that noise in face / skin coloured areas is very annoying and that excessive sharpening in these areas should therefore be avoided. However, they do not address the issue of enhancing previously compressed data.

In "Visually Improved Image Compression by Combining a Conventional Wavelet-Codec With Texture Modeling" (IEEE Trans. Image Proc , 2002) , Nadenau et al emphasize the importance of skin texture in an image as follows: " ... the small amplitude texture of the human skin is extremely well perceived by human observers. The sensitivity for a possible loss due to blurring is very high, because everybody is "trained" on looking at faces."

SUMMARY OF THE INVENTION

A first aspect of the invention provides a method of processing an image comprising:

if a part of the image is or has been determined to represent human skin applying a first noise reduction process to the part of the image, or, if the part of the image is or has been determined not to represent human skin, applying a second noise reduction process to the part of the image, the first noise reduction process giving an image with a higher noise and/ or texture content than the second noise reduction process .

According to an aspect of the invention a part of the image that is determined to represent human skin is processed so that part has a higher noise/ texture content than if the part is determined not to represent human skin . The additional noise/ texture present, after processing, in a part of the image that is determined to represent human skin effectively acts as a effectively acts as a substitute for natural skin texture which may have been lost during the image capture and transmission process, and helps to prevent skin from looking flat or "plastic" and generally provides more visually pleasing results .

The method can be used on its own or in combination with any noise reduction method . Furthermore, it can be applied to images with both analogue noise and digital compression artefacts .

The method is well-suited to use in TVs, mobile phones, computers and other display devices which often receive a low-quality noisy video from a variety of sources . It provides a simple and efficient way of improving the user experience , and can easily be added to an existing image processing pipeline.

A second aspect of the invention provides a method of processing a video image comprising processing at least first and second frames of the video image according a method of the first aspect.

A third aspect of the invention provides a noise reduction system having:

a module for determining whether a part of an image represents human skin; and

a module for, if the part of the image is determined to represent human skin, applying a first noise reduction process to the part of the image, or for applying, if the part of the image is determined not to represent human skin, a second noise reduction process to the part of the image , the first noise reduction process giving an image with a higher noise content than the first noise reduction process .

A fourth aspect of the invention provides a display system having a noise reduction system of the third aspect. A noise reduction system of the invention may be provided in any display device that is required to process an image including, as examples and without limitation, a television, a mobile phone, an advertising hoarding, a digital photograph display device, a computer display, a projector and other public or personal devices that include such a display system.

Aspects of the invention helps to improve the quality of images by enhancing the appearance of human skin in a way that makes it look more natural. This is important because skin in general - and faces in particular - attract a substantial amount of viewer attention when they are present in a scene .

This observation of the importance of skin texture to human observers is significant when considering the picture quality of both digital and analogue video . The present invention comprises an algorithm which describes how noise reduction can be performed, while simultaneously preserving or enhancing texture in skin-coloured areas of an image .

This provides the motivation for the current invention, which is that it can be useful to perform skin-adaptive noise reduction and image enhancement on both analogue and digital video , in order to improve picture quality.

Because the characteristics of analogue and digital noise are different, we adopt a different strategy depending on the video format. In both cases the goal is to perform noise reduction in the image, while achieving a sufficient amount of texture human skin .

In the case of analogue video , any standard noise reduction method (such as a bilateral filter) can be used. Where skin is detected, different parameters should be used in order to perform less blurring of skin-coloured areas . This helps to preserve more detail in skin, even though some of this detail may be grain resulting from analogue noise. In this case the analogue noise present in the image effectively acts as a substitute for natural skin texture which may have been lost during the capture and transmission process.

In the case of digital video , any standard noise reduction method (such as a bilateral filter) can be used throughout the entire image . Where skin is detected, a small amount of random or pseudo-random noise is added to the image. Adding artificial texture amounts to adding noise, which seems counter-intuitive . However, doing so helps to prevent skin from looking flat or "plastic" and generally provides more visually pleasing results . BRIEF DESCRIPTION OF THE DRAWINGS Preferred embodiments of the present invention will be described with reference to accompanying figures, in which:

Figure 1 is a schematic illustration of processing paths for analogue and digital video .

Figure 2 is a block diagram illustrating principal features of skin-adaptive picture quality improvement method according to an embodiment of the invention .

Figure 3 shows the distribution of human skin tones in Hue - Saturation space, in polar coordinates .

Figure 4 is a schematic block diagram of a noise reduction system according to an embodiment of the invention.

DESCRIPTION OF EMBODIMENTS

A method of processing an image according to an embodiment of the invention comprises, initially, determining whether a part of an image represents human skin . Alternatively, the determination whether a part of an image represents human skin may be performed separately (for example the image may be received together with overhead information indicating which parts of the image represents human skin. ) If the part of the image is or has been determined to represent human skin a first noise reduction process is applied to the part of the image, and if the part of the image is or has been determined not to represent human skin, a second noise reduction process is applied to the part of the image. The second noise reduction process is a noise reduction process that leads to an image with a higher noise and/or texture content than the first noise reduction process.

Thus, a part of the image that is determined to represent human skin is processed so that part has a higher noise and/or texture content than if the part is determined not to represent human skin. The additional noise and/or texture present, after processing, in a part of the image that is determined to represent human skin effectively acts as a effectively acts as a substitute for natural skin texture which

s may have been lost during the image capture and transmission process, and helps to prevent skin from looking flat or "plastic" and generally provides more visually pleasing result

Embodiments of the invention will be described in which the "part" of an image is a pixel, but in principle the invention is not limited to this.

In one embodiment of the invention the above ideas for skin-adaptive noise reduction and image enhancement can be summarised for both analogue and digital video using the following equation:

Y = (ps.fo) X + (1 - ps.fo) XNR + (ps.fsT.L(X)) t (1) where: X is an input pixel, which will generally contain components of two or more colours (eg a full colour pixel may have red , green and blue components) ,

XNR is pixel after standard noise reduction proce ssing (e . g . using a bilateral filter) ,

L(X ) is the brightne ss of pixel X - any suitable measure of the pixel brightness may be used, depending on the colour space in which the invention is being implemented ; for example in the YCbCr colour space the Y (luma)_component may be used as the brightness L(X) whereas in the H SV colour space the V (value)_component may be used as the brightne ss L(X) ,

ps is the (estimated) probability that pixel X repre sents human skin ,

fo is the weighting factor of the original image , which represents how much of the original (unproces sed) image it is desired to keep ,

fsT is the weighting factor for skin texture ,

t is a random / pseudo-random number used to add texture , and

Y is the output pixel

By "pseudorandom number" is meant a number in a sequence of numbers that, although the sequence is not truly random in so far as the sequence is determined by a set of initial values, approximates the properties of random numbers. The sequence may be generated using an available random number generator.

A wide range of values can be chosen for the weighting factors fo and fsT. Based on experiments with standard definition analogue and digital TV content, according to one embodiment of the invention the following parameters are recommended: fo ~ 0.5 and fsT = 0 for analogue video; and

f o = 0 and fsT ~ 0.02 for digital video.

In these embodiments equation (1) reduces to: Analogue: Y = (0.5 p_s) X + 0.5(1 - p_s) XNR (2)

Digital: Y = XNR + 0.02 p_S.L(X)) t (3)

It should be understood that the invention is not limited to the values of fo and fsT, which are given only by way of example. For example, in the analogue case fo may in principle take any value between 0 and 1, and in the digital case fsT may in principle, depending on the characteristics of the added texture, take any value between 0 and around 0.2.

It should be noted that the invention does not require that one of fo and fsT is equal to 0. Analogue content will have inherent grain-type noise whereas in digital content this noise is largely absent - so , in a typical case , texturing the skin is done by mixing in unprocessed content in the analogue case and by mixing in random noise in the digital case . In some cases, however, it may be desirable to use both unproce ssed content and random noise at the same time - that is , both fo and fsT would be non-zero .

It will be seen that, in the case that p_s = 0 , when the pixel definitely doe s not depict skin, equation ( 2 ) and equation (3 ) both reduce to Y = XNR, that is the output pixel is the input pixel after standard noise reduction processing (e . g. using a bilateral filter) .

Figure 1 summarises the recommended processing paths for analogue and digital video . Initially when an input image is received at a noise reduction proce ssor (which may for example be provided in a decoder such as a video decoder) , a determination is made whether the image is an analogue image or a digital image . (If it is already known whether the input image is an analogue image or a digital image , the determination may be omitted . )

If the input image is an analogue image , the image is then processed according to a method of the invention for processing an analogue image . If a pixel of the image is determined to represent human skin , a first noise proces sing technique is applied to that pixel . If a pixel of the image is determined not to represent human skin, a second noise processing technique is applied to the pixel; the second noise reduction process is a weaker noise reduction process than the first noise reduction process, so that the method gives a noisier image in a pixel determined to represent human skin. For example , the image may be processed according to equation (2) above .

If the input image is an digital image, the image is then processed according to a method of the invention for processing an digital image . If a pixel of the image is determined to represent human skin, a first noise processing technique is applied to that pixel. If a pixel of the image is determined not to represent human skin , a second noise processing technique is applied to the pixel of the image ; the second noise reduction process is a weaker noise reduction process than the first noise reduction process, so that the method gives a noisier image in a pixel determined to represent human skin. For example, the second noise reduction process may be a combination of the first noise reduction process and the addition of random texture as in equation (3) .

When adding a small quantity of synthetic texture to skin, the amount of texture added is preferably proportional to a measure of the brightness of the skin. The amount of texture should also be proportional to the probability that an area represents skin.

In principle, a method of the invention may be limited to assigning one of the two values p_s = 0 or p_s = 1 to a pixel in the step of determining whether a part of the image represents human skin. However, allowing the probability p_s of skin to adopt a range of values between 0 and 1 (for example p_s may take the values [0, 0.1, 0.2....0.9, 1] or to be continuous in the range [0,1] helps to reduce the likelihood of sudden transitions between textured and non-textured regions in areas that might be skin.

The texture t to be added to an image in the term (ps.fsT.L(X)) t of equation (1) or (3) can conform to a variety of random distributions. One suitable distribution is additive pink Gaussian noise, since pink noise has a 1/f power spectrum which is typical of natural image content (where "f" is a spatial frequency - the power spectrum of natural images decreases (at a rate of 1/f) as the spatial frequency increases so that there is not much power at high spatial frequencies).

The present invention may be applied to processing of a still image, or to processing of a sequence of image such as a video image. Where a method of the invention is applied to a video image or other image sequence, equations (1) to (3) define how a frame is processed.

For video, there are a number of possibilities regarding the temporal characteristics of any pseudo-random noise, for example including:

The pseudo-random noise pattern can change from frame to frame . This can make the noise "twinkle" , causing it to appear less like natural texture and more like noise .

A fixed pseudo-random noise pattern can be used - that is, the same pseudo-random noise pattern is used for each frame . This generally gives a better result, but can cause the noise to look like dirt on the screen when there is slow obj ect motion.

A combination of the above is also possible - the pseudo-random noise pattern may have one component that is constant between frames and another component that changes from frame to frame, allowing the noise pattern to change more slowly over time . This is not perfect, but gives a more natural result.

A motion compensated texture pattern could be used - allowing texture to track a moving face. This would produce the most realistic result, but it is the most computationally expensive solution.

Skin detection can be performed in a variety of colour spaces. One of the most popular colour spaces is the HSV colour space, since human skin tones from a variety of races form a fairly tight cluster in Hue - Saturation space, as shown in Figure 3 ("Skin Colour Analysis", Sherrah and Gong, 200 1 - available at http : / / homepages, inf. ed . ac.uk/ rbf/ CVonline / LOCAL_CO PIES / GONG 1 / cvOnline- skinColourAnalysis . html) .

In embodiments of the invention the determination whether a pixel of an image may be based on one or more first components of colour space, and the noise reduction may be applied to a second component of colour space . This is illustrated in figure 2 , which provides a more detailed block diagram to illustrate the operation of the algorithm . Figure 2 shows an embodiment of the invention applied in the HSV colour space, where H denotes "hue" , S denotes "saturation" and "V" denotes to the "value" component in the HSV colour space .)

Initially an image is input at 1 . In the case of a full colour image this is an RGB (red, green, blue) image, and at 2 the image (or a part thereof) is transformed into another colour space, in this example into the HSV colour space .

At 3 , a determination of whether a pixel X of the colour- transformed image represents human skin is made . This determination is carried on one or more first components of the colour space, in this example on the H and S components. The determination comprises assigning a probability p_s to a pixel that represents the probability that the pixel represents human skin . As noted above, the invention may be implemented by assigning one of the two values p_s = 0 or p_s =■ 1 to a pixel of the image , but better results may be achieved by allowing the probability p_s of skin to adopt a range of values between 0 and 1 or to be continuous in the range [0 , 1 ] .

It should be noted that the determination of whether a pixel X of the colour-transformed image represents human skin may be omitted, if it is already known whether the pixel X represents human skin . For example, where the method is being performed at a decoder such as a video decoder, the input data received at the decoder may contain overhead information indicating which pixels represent human skin .

At 4 a conventional noise reduction process is applied to the pixel X, for example by applying a bilateral filter, to produce a noise reduced pixel XNR. The noise reduction process may be performed on at least a second component of the colour space, preferably on a component of the colour space not used in the determination 3 of the probability p_s.

At 5 a pseudo-random skin texture t is generated, using any suitable pseudo-random number generator.

At 6 , an output pixel Y is determined from the input pixel X and one or more of the probability p_s assigned to the pixel at 3 , the noise reduced pixel XNR obtained at 4 , the pseudo-random texture generated at 5 , and the weighting factors fo , fsT . For example, the output pixel may be determined at 6 according to equation ( 1 ) , or according to n equation derived therefrom such as equation (2 ) or (3 ) .

Each pixel of the colour-transformed image is processed according to operations 3-6 , and the noise-reduced pixel are then assembled, and at 7 are transformed back to the RGB colour space, to give an output image 8.

Figure 2 shows the process applied to a single image, for example a still image or one frame of a video image . When the invention is applied to a video image each frame is processed as shown in figure 2 (although, as noted, the texture t may vary from one frame to another) .

As shown in Figure 2 , the use of HSV space allows for the probability that a pixel represents skin to be estimated using the Hue (H) and Saturation (S) components. Noise reduction and texture enhancement is then performed on the Value (V) component. Since the Value term closely correlates with brightness this allows for an efficient and effective implementation of the algorithm.

A method of the invention is not however limited to the HSV colour space but may be performed in other colour spaces . Another suitable colour space is the YCbCr colour space . In this case, the chrominance components Cb and Cr can be used to estimate the probability that a part of the image represents skin, while noise reduction and texture enhancement is performed on the Luminance (Y) component.

A method of the invention may be implemented in any suitable noise reduction system, for example a noise reduction system that can perform the operations 2 -7 of figure 2.

A noise reduction method of the invention may be carried out in, and a noise reduction system for implementing a method of the invention may be provided in, any device that is required to process an image including, as examples and without limitation, a television, a mobile phone , an advertising hoarding, a digital photograph display device, a computer display, a projector and other public or personal devices that include such a display system.

Figure 4 is a schematic block diagram of a noise reduction system according to an embodiment of the invention . A noise reduction system of the invention may comprise a module 10 for, if a part of the image is determined to represent human skin, applying a first noise reduction process to the part of the image, or for applying, if the part of the image is determined not to represent human skin, a second noise reduction process to the part of the image, the first noise reduction process giving an image with a higher noise content than the second noise reduction process. It may optionally further comprise a module 1 1 for determining whether a part of an image represents human skin, and may optionally further comprises other modules (not shown in figure 4) for performing other features of a method of the invention. The module or modules may be implemented as hardware, or they may be implemented On one or more processors . (Figure 4 shows the modules 10, 1 1 as separate from one another for clarity, but in principle the modules could be implemented on the same processor.) The module 1 1 for determining whether a part of an image represents human skin may be adapted to assign a probability that the part of the image represents human skin.

Although the invention has been shown and described with respect to certain embodiments, equivalent alterations and modifications may occur to people skilled in the art upon the reading and understanding of this specification and the annexed drawings. In particular regard to the various functions performed by the above described elements (modules components, assemblies, devices, compositions, etc. ) , the terms used to describe such elements are intended to correspond, unless otherwise indicated, to any element which performs the specified function of the described element (i. e . , that is functionally equivalent) , even though not structurally equivalent to the disclosed structure which performs the function in the above exemplary embodiments of the invention. In addition, while a particular feature of the invention may have been described above with respect to only one or more of several embodiments, such feature may be combined with one or more other features of the other embodiments, as may be desired and advantageous for any given or particular application . A method of the first aspect may further comprise determining whether a part of an image represents human skin . In this embodiment, the determination of whether part of an image repre sents human skin and the proce ssing of that part of the image using the first or second noise reduction process , as appropriate , may be carried out together, for example in a single device that receives and processes the image . Alternatively, the proces s of determining whether a part of an image represents human skin may be carried out separately.

Determining whether a part of an image repre sents human skin may comprise assigning a probability that the part of the image represents human skin .

The first noise reduction process and the second noise reduction process may be noise reduction processes of the same type .

The first noise reduction proce ss may comprise adding random or pseudorandom noise to the part of the image .

Applying the first noise reduction process may comprise applying the second noise reduction proce ss and adding random or pseudorandom noise .

The method may comprise proce ssing a part of the image according to :

Y = (ps. fo) X + ( 1 - ps. fo) XNR + (ps. fsT. L(X) ) t ( 1 ) where X is the input part, XNR is the input part after standard noise reduction processing (e . g. using a bilateral filter) , L(X) is the brightness of the part X, ps is the (estimated) probability that the part X represents human skin, fo is the weighting factor of the original image , fsT is the weighting factor for skin texture, t is a random / pseudorandom number used to add texture, and Y is the output part.

The determination whether a part of an image represents human skin may be performed on one or more first components of the image in a colour space .

The method may comprise applying the first or second noise reduction process on a second component of the image in the colour space.

The colour space may be the HSV colour space , the determination whether a part of an image represents human skin may be performed on the Hue or Saturation of the part of the image, and the first or second noise reduction process may be applied on a Value of the part of the image .

Alternatively the colour space may be a luminance- chrominance colour space, the determination whether a part of an image represents human skin may be performed on one or more chrominance components of the part of the image , and the first or second noise reduction process may be applied on a luminance component of the part of the image .

A method of the second aspect may comprise processing the first frame by adding first pseudorandom noise to a part of the first frame determined to represent human skin and processing the second frame by adding second pseudorandom noise different from the first pseudorandom noise to a part of the second frame determined to represent human skin .

A method of the second aspect may comprise processing the first frame by adding pseudorandom noise to a part of the first frame determined to represent human skin and processing the second frame by adding the same pseudorandom noise to a part of the second frame determined to represent human skin .

A method of the second aspect may comprise processing the first frame by adding motion-compensated pseudorandom noise to a part of the first frame determined to represent human skin and processing the second frame by adding motion-compensated pseudorandom to a part of the second frame determined to represent human skin .

A noise reduction system of the third aspect may further comprise a module for determining whether a part of an image represents human skin .

The module for determining whether a part of an image represents human skin may be adapted to assign a probability that the part of the image represents human skin.

Claims

1 . A method of processing an image comprising:

if a part of the image has been determined to represent human skin applying a first noise reduction process to the part of the image , or, if the part of the image has been determined not to represent human skin, applying a second noise reduction process to the part of the image , the first noise reduction process giving an image with a higher noise and / or texture content than the second noise reduction process .

2. A method as claimed in claim 1 and comprising determining whether the part of an image represent human skin .

3. A method as claimed in claim 2 wherein determining whether a part of an image represents human skin comprising assigning a probability that the part of the image represents human skin .

4 A method as claimed in claim 1 , 2 or 3 wherein the first noise reduction process and the second noise reduction process are noise reduction processes of the same type.

5. A method as claimed in claim 1, 2 or 3 wherein the first noise reduction process comprises adding random or pseudorandom noise to the part of the image.

6. A method as claimed in claim 5 wherein applying the first noise reduction process comprises applying the second noise reduction process and adding random or pseudorandom noise.

7. A method as claimed in any preceding claim wherein the method comprises processing a part of the image according to:

Y = (ps.fo) X + (1 - ps.fo) XNR + (ps.fsT.L(X)) t (1) where X is the input part, XNR is the input part after standard noise reduction processing (e.g. using a bilateral filter), L(X) is the brightness of the part X, ps is the (estimated) probability that the part X represents human skin, fo is the weighting factor of the original image, fsT is the weighting factor for skin texture, t is a random / pseudorandom number used to add texture, and Y is the output part.

8. A method as claimed in claim 2 or in any of claims 3 to 7 when directly or indirectly dependent from claim 2 wherein the determination whether a part of an image represents human skin is performed on one or more first components of the image in a colour space .

9. A method as claimed in claim 8 and comprising applying the first or second noise reduction process on a second component of the image in the colour space .

10. A method as claimed in claim 8 or 9 wherein the colour space is the HSV colour space, the determination whether a part of an image represents human skin is performed on the Hue or Saturation of the part of the image, and the first or second noise reduction process is applied on a Value of the part of the image .

1 1 . A method as claimed in claim 8 or 9 wherein the colour space is a luminance-chrominance colour space, the determination whether a part of an image represents human skin is performed on one or more chrominance components of the part of the image, and the first or . second noise reduction process is applied on a luminance component of the part of the image .

12. A method of processing a video image comprising processing at least first and second frames of the video image according a method as defined in any one of claims 1 to 1 1 .

13. A method as claimed in claim 12 , wherein the method comprises processing the first frame by adding first pseudorandom noise to a part of the first frame determined to represent human skin and processing the second frame by adding second pseudorandom noise different from the first pseudorandom noise to a part of the second frame determined to represent human skin.

14. A method as claimed in claim 12 , wherein the method comprises processing the first frame by adding pseudorandom noise to a part of the first frame determined to represent human skin and processing the second frame by adding the same pseudorandom noise to a part of the second frame determined to represent human skin .

15. A method as claimed in claim 12 , wherein the method comprises processing the first frame by adding motion-compensated pseudorandom noise to a part of the first frame determined to represent human skin and processing the second frame by adding motion-compensated pseudorandom to a part of the second frame determined to represent human skin .

16. A noise reduction system having: a module for, if a part of an image is determined to represent human skin , applying a first noise reduction process to the part of the image , or for applying, if the part of the image is determined not to represent human skin, a second noise reduction process to the part of the image, the first noise reduction process giving an image with a higher noise and/ or texture content than the second noise reduction process .

17. A noise reduction system as claimed in claim 16 and further comprising a module for determining whether a part of an image represents human skin .

18. A noise reduction system as claimed in claim 17 wherein the module for determining whether a part of an image represents human skin is adapted to assign a probability that the part of the image represents human skin .

19. A display system having a noise reduction system as defined in claim 16, 1 7 or 1 8.