FR2824689A1 - Public video phones object extraction process having initial image real time evaluated calculation euclidean distance background/current image and finding level/object shape then extracting image. - Google Patents

Abstract

The object extraction process has an initial image (2) of a sequence of images from a camera. The object is evaluated in real time using a background image (11), calculating the euclidean distance between the background image and the current image. The level is then calculated, and the object shape modelled. A representation of the object is then extracted from the current image.

Description

WAP.

  METHOD FOR EXTRACTING OBJECTS FROM AN IMAGE ON A

  UNIFORM BACKGROUND AND INTEGRATION OF THESE OBJECTS INTO A

SCENE.

  s The invention relates to a method for extracting objects from an image, on a uniform background and

  the inlay of these objects in a scene.

  It is intended for real-time applications on PC, as varied as multimedia on the Internet, videoconferencing, remote monitoring, reality

virtual, the game ....

  Although this application should not be considered as limiting, the use of the method of the invention is particularly suitable for videophone for the general public, under conditions

  natural and / or artificial lighting.

  It is known that digital imaging systems allow the user to extract the object from the background and embed this object in another more user-friendly background,

  like an image of the moon, an animated virtual scene.

  Previously, two methods were used to obtain this result: l) the background, consisting of a brilliant screen of uniform color is used behind the object or the character, so as to construct a mask of the character by the "d" method blue screen "or" Chroma key ". An example of this method is described in the document US

5424.781.

  This method can give excellent results, but requires a uniform color background with careful and expensive lighting. In addition, this method uses specific hardware components for its processing.

in real time.

  2) The character can be captured in front of any type of background, and extracted from the background by software tools such as Adobe Photoshop for example. However

  processing is carried out in deferred time.

  Furthermore, a chromatic key can be produced independently of variations in the intensity level of the color signal, as described in document US Pat. No. 608,466, however it requires components

  specific materials for its processing in real time.

  The present invention overcomes these drawbacks by proposing a method for extracting the objects from an image, preferably on a uniform background of blue color and embedding them in a scene, without requiring any particular hardware components and a

specific lighting.

  An algorithm describes the different stages of real-time processing of the process with a tolerance to variations in light intensity, so as to benefit from the usual lighting conditions, such as daylight and / or traditional lighting. a

plece.

  The processing intended to extract the objects from the background is carried out at the base by a Euclidean distance calculation between the points of an initial image representing a uniform background and the points of the image

  current representing the objects in this uniform background.

  In general, the distance is a function of real value d [(jl, kl), (j2, k2)] of two image points (jl, kl) and (j2, k2) which satisfies the following properties: d [(jl, kl), (j2, k2)] 2 0 d [(jl, kl), (j2, k2)] = d [(j2, k2), (jl, kl)] d [(jl, kl ), (j2, k2)] + d [(j2, k2), (j3, k3)] 2 d [(jl, kl), (j3, k3)] There are a number of distance functions that satisfy the properties thus defined such that the Euclidean distance is: dE = [(jl-j2) 2 + (kl-k2) 2] 1/2 Furthermore, taking into account variations in intensity of the light source captured by the camera, allows to overcome noise problems and obtain sufficient information, especially in

natural lighting conditions.

  The invention therefore relates more particularly to a method for extracting objects from an image composed of video objects in the foreground and from a uniform background in the background; this initial image being designated by the current image of a sequence of images captured by

a camera.

  To obtain in real time an extraction of these objects, with a tolerance for variations in intensity of the capLurce light source by the camera, under natural and / or artificial lighting conditions, it includes, according to a first characteristic, the following steps: - in a first phase, a representation of the background image; - in a second phase, a calculation of the Euclidean distance between the representation of the background image

and the current image.

  - in a third phase, the determination of a threshold; - in a fourth phase, the construction of a model of the shape of the object; and in a fifth phase, the representation of

  the object, extracted from the current image.

  s According to particular embodiments: 15. in the first phase, the representation of the background image can be obtained by acquiring a background image, before processing

the current image.

  20. in the first phase, to be insensitive to variations in intensity, the representation of the background image can be obtained from an area of interest of the current image, when this area represents the background. the area of interest can comprise at least the first luminance line of the current image, preferably the first four, in order to anticipate the

  object detection in this first line.

  the detection of the background in the area of interest of the current image can be carried out according to the following steps: s - a detection of the contour, preferably in column and line of the first five lines of luminance of the current image; - a threshold of around 10; - a median filter, preferably in column and row, using 3 structuring elements; and a test of at least a first binary line, of

preferably the first four.

  the background image can be reconstructed, according to a copy of the first of the current image, in all the respective lines of the background image,

  when this first line represents the background.

  a sliding average of the pixels of the first line of luminance and chrominance of the current image, can be carried out before the copying of the first line of the current image, in order to obtain

  a first homogeneous reference line.

  the background image can be refreshed according to the following steps, when the first luminance line of the current image represents the background: in a first part, a subtraction between the average value of the pixels of the first luminance line of the current image and the average pixel value of the first luminance line of the image

initial background.

  - in a second part, an addition between the signed value of this subtraction, and the voltage level of each pixal of the luminance signal of the initial background image, when the absolute value of this subtraction is greater than a threshold value not

  previously defined, for example 5.

  in the second phase, to reduce the calculation time, the Euclidean distance between the representation of the background image, defined by the luminance (Yf), chrominance (Uf, Vf) signals and the current image, defined by the luminance (Ypf) and chrominance (Upf, Vpf) signals can be calculated by the following chrominance relation: dE W = [(Uf Upf) 2 + (Vf - Vpf) 2] in the second phase, the distance Euclidean between the representation of the background image, defined by the luminance (Yf), chrominance (Uf, Vf) signals and the current image, defined by the luminance (Ypf), chrominance (Upf, Vpf signals) ), can be calculated by the following Luminance and Chrominance relation: dE YW = [(yf _ ypf) 2 + (Uf - Upf) 2 + (Vf - Vpf) 2] in the third phase, to extend its operating range , it can include an adaptive threshold, based on the minimum value of a histogram of the Euclidean distance between the representation of the image

background and the current image.

  the adaptive threshold may also include a dynamic operating interval, based on the

maximum value of the histogram.

  the adaptive threshold, based on the minimum and maximum value of the histogram of the Euclidean distance between the representation of the background image and the current image, can be calculated on a sliding average of the previous values of the adaptive threshold, for example

of the last five values.

  in the fourth phase, to obtain a shape model of the object, the construction of a mask can be carried out by the threshold binary image of the Euclidean distance between the representation of the image

background and the current image.

  in the fourth phase, to obtain intermediate levels of tension in the contour region of the mask, it can comprise an Alpha plane defined according to an equation of type: Alpha Yn = ... (An-2). (Yn-2) + (An-l). (Yn-1) + An.Yn + (An + l). (Yn + 1) ..... o An are the respective coefficients of the intermediate voltage levels and Yn the pixcls of l image of the mask. to obtain intermediate steps in the contour region of the mask, for example and more advantageously 5 intermediate voltage levels for the luminance and 3 intermediate voltage levels for the chrominance, the respective Alpha planes can be defined according to the following equations, for the luminance: Alpha Yn3 = (An-3). (Yn-3) + (An-2). (Yn-2) + (An-l). (Yn 1) + An.Yn + (An + l). (Yn + 1) + (An + 2). (Yn + 2) with the coefficients: An-3 = 1/12, An-2 = 2/12, An-1 = 3/12, An = 3/12 , An + 1 = 2/12, An + 2 = 1/12; which gives for a step {0, 255}, the successive levels {0, 21.25, 63.75, 127.50,191.25, 233.75, 255} for the chrominance: Wn4 = (An-2). (Wn-2) + (An -l). (Wn-l) + An.Wn + (An + l). (Wn + l). with the coefficients An-2 = 1/6, An-1 = 2/6, An = 2/6, An + l = 1/6; which gives for a step of {0, 255}, the successive levels {0, 42.50, 127.50, 212.50, 255} in the fourth phase, to reduce the calculation time of the Alpha plan, the latter can be built in a zone of interest of the mask representing the region

contour of objects.

  the zone of interest intended for the calculation of the Alpha plane can be defined, from the contours of the mask, by a change of origin of the equation of the Alpha plane the zone of interest intended for the calculation of the Alpha plane can be defined by building a mask

contour of the object mask.

  in the fifth phase, to extract the object from the current image, it can perform a first

  multiplication between the current image and the Alpha plane.

  30. the method for embedding objects in a scene, can use the method intended for extracting objects from an image, for carrying out a reconstruction of the luminance and chrominance signals of the image representing the scene, with the objects extracted from

the current image.

  to embed objects in a scene, it can include the following steps: - construction of a background shape model by inverting the object's Alpha plane; - a representation of a scene background by a second multiplication between the scene image and the reverse of the Alpha plane; - a reconstruction of the image by an addition between the background from the second multiplication and the object of the current image from

of the first multiplication.

  to perform a clipping, it may include the following steps: - detection of the outline of the mask W; - loading of contour values to a value

  no chroma, such as 128.

  The accompanying drawings illustrate the invention: - Figure 1 schematically illustrates the steps of image processing, according to the method according to the invention, in its basic operating principle. - Figure 2 shows schematically the image processing steps, according to the method according to the invention, in its complete embodiment, relating to the standard W. - Figure 3 is an image representing a histogram in the foreground and the mask of a character,

background.

  - Figure 4 schematically shows the steps of image processing, according to the method according to the invention, in its complete embodiment, relating to the standard Y W. - Figure 5 shows schematically the steps

  reconstruction of the background image.

  - Figure 6 is a schematic representation

  steps to refresh the background image.

  - Figure 7 shows schematically the steps

  of the optimized construction of the Alpha plan.

  - Figure 8 shows an example of images illustrating the different stages of construction

  optimized from the Alpha plan of figure 7.

  - Figure 9 is a graphical representation of

  stages of the optimized construction of the Alpha plan.

  - Figure 10 schematically shows the

reconstruction of the image.

  - Figure 11 schematically illustrates the steps

background image clipping path.

  - Figure 12 graphically represents an optimized construction of the Alpha plane of the signal

luminance.

  - Figure 13 is a graphical representation of

  construction of the alpha plane of the chrominance signal.

  - Figure 14 graphically represents the

  construction of the Alpha plan, after a clipping.

  With reference to these drawings, FIG. 1 represents the steps of the process that can be implemented

  in accordance with the invention, in its basic principle.

  According to a nonlimiting example, a camera (1) takes an initial image of a uniform background (3), then the images (2) composed of a model placed in front of a uniform background. These two steps provide a background image (3) and a composite image which will be considered as

the current image (2).

  To these two images, a first treatment will be applied, allowing to define an area of interest of

the image.

  This first step can be implemented by calculating the Euclidean distance (4) between the background image

(3) and the current image (2).

  A threshold (5) makes it possible to define different classes (at least two) of gray levels. For example, at least one of these classes corresponds to the background of the image and is not retained: the background mode is then reduced to 0 in the source image. From this operation, we can therefore deduce a model of form (6) of the object, which

  allows to define an area of interest of the image.

  The object (8) extracted from the current image (2) is obtained by a multiplication (7) between the shape model (6) of the object and the current image (2). This object (8) therefore contains only the elements of the image which are

chosen as significant.

  We then carry out a global processing (global approach) of the image: the object (8) will be embedded (9) in an image composing a scene, to display a

recomposed image (10).

  FIG. 2 illustrates the processing chain of the invention process, according to the W standard. This chain makes it possible, in a first step, to extract the object from a video image. This image being considered as the current reference image (2). This first step consists in making a representation of the background image (11), from the current image (2). The calculation of the Euclidean distance (12) between the chrominances of the representation of the background image (11) and the current image (2), makes it possible to obtain a basic representation of the object composing the image.

current (2).

  The minimum value of the histogram (13) of the Euclidean distance (12) makes it possible to define an adaptive threshold (14) while a sliding average (15) of this threshold (14) participates in the construction of a model of shape of the object characterized by the construction of a mask (16) and an alpha plane (18) of luminance (16) and

a chrominance mask (17).

  The second step is to embed the object in a scene. For this, the operations of reconstruction of the luminance (19) and chrominance (20) signal will make it possible to obtain, after a clipping (21) on the chrominance, a recomposed image (10) representing

the object in the background.

  FIG. 3 is an example of an image which represents, in the background a binary image of the mask (16) of a character and in the foreground the histogram (13) of the Euclidean distance (12) between the current image ( 2) and

  the representation of the background image (11).

  The description of this histogram (13) highlights

  evidence of three regions: - a first region (or "high mode") which contains only the background of the image; - a central region, which mainly contains information dividing the areas of interest of the image; - a third region (or "low mode") which mainly contains the character information. The central region of the histogram is delimited by a dynamic operating interval (S1, S2),

  set to the maximum value (S1) of the histogram (13).

  While the adaptive threshold (14) is determined by the minimum value of the histogram (13), in the interval

considered (S1, S2).

  A sliding average (15) of the threshold (14) makes it possible to obtain a threshold value adapted to the construction

of the mask (16).

  With reference to FIG. 4, the processing chain of the process of the invention, according to the standard Y W. in principle takes up an approach similar to the processing

  previous according to the W. standard illustrated in figure 2.

  The representation of the background image (11) is

  made from the current image (2).

  The calculation of the Euclidean distance (22) of luminance and chrominance, between the representation of the background image (11) and the current image (2), makes it possible to obtain a basic representation of the object.

composing the current image (2).

  The minimum value of the histogram (13) of the Euclidean distance (22) makes it possible to define an adaptive threshold (14) while a sliding average (15) of this threshold (14) participates in the construction of a model of shape of the object, relating to the luminance and chrominance signals. The construction of this shape model comes in three parts: the construction of a mask (23, 17), a median filter (24, 25) and an Alpha plane (18, 26)

  relating to the luminance and chrominance signals.

  The second step is to embed the object in a scene. For this, the operations of reconstruction of the luminance (19) and chrominance (20) signal will make it possible to obtain, after a clipping (21) on the chrominance, a recomposed image (10) representing

the object in the background.

  Furthermore, the processing chains according to the W and YUV standard, illustrated in FIGS. 2 and 4, are characterized by the following three comparative points: 1). the use of this second processing chain according to the Y W standard makes it possible to obtain an optimum representation of the contours of the object, due to the initial definition of the luminance mask (23), which corresponds to the same definition as the

  luminance signal of the current image (2).

  While the first processing chain according to the standard W. presented in FIG. 2, has a luminance mask (16), calculated on the basis of the definition of the chrominance signal; or with four times less pixels. However the Alpha plane (18) makes it possible to improve the edges of the mask (16, 23) of

significantly.

  2). the processing time of the second chain Y W is greater than the first W. taking into account

operations that are added.

  As an example, a PCO type "general public" machine using a 133Mhz Pentium processor, an advanced programming language like "C language", and images in QCIF format defined by 176x144 pixels, will process the process. invention, under the following conditions: 5. According to the standard Y W. with reference to FIG. 4, the calculation time is of the order of 260 ms, ie approximately:

  - 3ms for the construction (11) of the background image.

  - 100ms for Euclidean distance calculation according to

  equation (22) of luminance and chrominance.

  - lms for the calculation of the sliding average (15) of the

  adaptive threshold (14), resulting from the histogram (13).

  - 7 ms for the construction of the Y mask (23).

  - 26ms for the 3-element Y filter (24)

structuring in row and column.

  - 50ms for the optimized construction (18) of the Alpha Y plan.

  - 27ms for reconstruction Y (19).

- 4 ms for the mask W (17).

  - 7 ms for the 3-way W (25) median filter

structuring in row and column.

  - 5ms for the optimal construction of the Alpha W plan (26). According to the W. standard with reference to FIG. 2, the calaul time is of the order of 14 ms, or approximately:

  - 3ms for the construction (11) of the background image.

  - 3Oms for Euclidean distance calculation (12) according to

the chrominance equation.

  - lms for the calculation of the sliding average (15) of the

  adaptive threshold (14), resulting from the histogram (13).

  - 3ms for the construction of the mask Y (16) and W (17).

  - 50ms for the optimized construction of the Alpha Y plan (18).

  - 27ms for reconstruction Y (19).

  - 15ms for reconstruction W (20) and clipping (21). -lOms for copying the Y W signals into the image pointer, intended for display (10). 3). In an application relating to consumer video telephony, the lighting conditions may vary

not negligible.

  For this purpose, the second processing according to the YW standard is relatively sensitive to variations in brightness, with reference to the terms Ypf which is the luminance signal of the current image representing the character and the background Yf which is the luminance signal of the background image, in the following relation of Luminance and Chrominance (22):

  dE Y W = [(yf _ ypf) 2 + (Uf - Upf) 2 + (Vf - Vpf) 2].

  While the first processing W is less sensitive to variations in brightness because it depends only on the chrominance signal, according to the terms Upf Vpf which are the chrominance signals U and V of the current image and Uf Vf which are the chrominance signals U and V of the background image of the following equation (12):

  dE W = [(Uf - Upf) 2 + (Vf - Vpf) 2].

  With reference to FIG. 5, the stages of construction of the background image are carried out from

of the current image (2).

  Firstly, an area of interest of the current image (2) is defined, according to a nonlimiting example, by the first line of the current image (2); this line being brought to represent the most

often the bottom.

  For this purpose, a detection of contours (27) of the first lines of the luminance signal of the current image (2), followed by a thresholding (28) and a median filtering (29), makes it possible to detect (30 ) objects in

this area of interest.

  Secondly, the presence of objects in this area leads to the use of the last reconstructed background image (32); While the absence of objects allows the calculation of the sliding average (31) of the first line of luminance and chrominance (33), followed by a copy (34) of these first lines (31, 33) in all the respective lines of the background image, in order to obtain a

  new reconstructed background image (11).

  FIG. 6 represents the processing chain for

the refreshed background image.

  It partially repeats the stages of reconstruction of the background image, presented in the previous figure 5, to take into account, the content

of the initial background image (3).

  Indeed, an area of interest of the current image (2) is defined, according to a nonlimiting example, by the first line of the current image (2); this line

  being brought more often to represent the bottom.

  To this end, a detection of contours (27) of the first lines of the luminance signal of the current image (2), followed by a thresholding (28) and a median filtering (29), makes it possible to detect (30) the objects in

this area of interest.

  Secondly, the presence of objects in this area results in the use of the last refreshed background image (40); While the absence of objects makes it possible to determine the difference in brightness between the current image and the initial image of the background, by the difference (37) between the average value of the pixels of the first line of luminance of the current image (36) and that of the initial background image (35) and thus to refresh the background image (11) by assigning this difference (39) to each pixel of the luminance signal of the initial image of the background. background (3), when the absolute value of this difference is greater than a previously defined value n (38) by

example 5.

  It is recalled that the initial image of the background is a single image, taken by the camera (1), prior to the first processing of the current image (2), with reference

in figure (1).

  The optimized construction of the Alpha plane, presented in FIG. 7, comprises, according to a first characteristic, a detection (41) of the outline of the mask (23) of the object while a thresholding (42) allows

  obtain a noise-free contour image.

  A change of origin (43) of the equation of the Alpha plane is based on the contours present in the thresholded image (42). This makes it possible to construct (44) the Alpha plane only in the contour region of the mask (23),

  to obtain an optimized Alpha plan (18).

  The images illustrated in figure 8, shows the different stages of construction of the Alpha plan

  optimized, from an example of a mask image (23).

  The thresholded image (42) makes the mask contours appear significantly, while the construction of the Alpha plane carried out only in the mask contour region is presented in the following image (44). Finally, the optimized construction of the Alpha plane which is illustrated in the last image (18), makes it possible to check the quality of the Alpha plane, comparable to a traditional treatment, carried out on the entire mask with a longer treatment time, in a report. two to three, depending on the representation of the object. Referring to Figure 9, the graphical steps of the optimized construction of the Alpha plan show the operations that are carried out in the regions of

contours of the mask (23).

  The graph (42) of the threshold binary image presents the contours of the mask (23), in the form of an impulse, in the transient region of the mask, while the graph (43) represents the change of origin of the equation of the Alpha plane, with a shift to the left, sufficient to correspond to the median value of the Alpha plane constructed (44) only in the

mask outline region.

  The processing chain presented in the figure illustrates the operating mode of a reconstruction of

the image.

  A first multiplication (45) between the current image (2) and the Alpha plane (18) of the object makes it possible to obtain the object extracted from the current image (2) while a second multiplication (48) between the reverse Alpha plane (18) (47) and a scene image (46) makes it possible to restore a background which is added (49) to the object extracted from the current image, in order to obtain an image

  (19) recomposed of the object in a scene.

  The construction of the trimming which is presented in FIG. 11, includes a contour detection (50) of the

chrominance mask (17). A loading of the contour (51) at a zero chroma value, by

  example 128, in the contour region of the chrominance mask (17), makes it possible to obtain a

  clipping (21) of the background image (3).

  With reference to the first graph in FIG. 12, the 1S luminance signal (18) of the object's Alpha plane, is constructed, from the mask (23), on five intermediate levels, according to the equation: Alpha Y = (An-3). (Yn-3) + (An-2). (Yn-2) + (An-l). (Yn-l) + An.Yn + (An + l). (Yn + l ) + (An + 2). (Yn + 2) with the coefficients: An-3 = 1/12, An-2 = 2/12, An-1 = 3/12, An = 3/12, An + l = 2/12, An + 2 = 1/12; which gives for a step {0, 255}, the successive levels {0, 21.25, 63.75, 127.50, 191.25, 233.75, 255} The reverse Alpha plane (47) representing the background overlaps with the Alpha plane (18 ) of the object in the middle part of the transition of the luminance signal (18). The graph in FIG. 13 represents the chrominance signal (26) of the alpha plane of the image, constructed from the mask (17), according to the following equation: Alpha W = (An-2). (Wn- 2) + (An-l). (Wn-1) + An.Wn + (An + l). (Wn + 1). with the coefficients An-2 = 1/6, An-1 = 2/6, An = 2/6, An + 1 = 1/6; which gives for a step of {0, 255}, the levels

  successive {0, 42.50, 127.50, 212.50, 255}.

  With reference to the graph in FIG. 14, the clipping occurs on two pixels (51) in the contour region of the mask (17), in order to obtain a refined clipping of the chrominance signal (21) of the reconstructed image.

Claims (23)

  1. Method for extracting objects from an image composed of video objects in the foreground and of a uniform background in the background; this initial image being designated by the current image (2) of a sequence of images captured by a camera (l), characterized in that, in order to obtain in real time an extraction of these objects, with a tolerance for variations d intensity of the light source captured by the camera (l), in natural and / or artificial lighting conditions for taking pictures, it comprises the following steps: - in a first phase, a representation of the background image ( ll); - in a second phase, a Euclidean distance calculation (4) between the representation of the background image (ll) and the current image (2); - in a third phase, the determination of a threshold (5); - in a fourth phase, the construction of a shape model (6) of the object; and - in a fifth phase, the representation (8)   of the object extracted from the current image (2).   2. Method according to claim l, characterized in that, in the first phase, the representation of the background image (ll) is obtained by acquiring an initial background image (3), prior to processing of the current image (2).   3. Method according to claim 1, characterized in that, in the first phase, to be insensitive to variations in intensity, the representation of the background image (11) is obtained from an area of interest of the current image (2), when this area represents the bottom.   4. Method according to claim 3, characterized in that the zone of interest comprises at least the first luminance line of the current image (2), preferably the first four, in order to anticipate the   object detection in this first line.   5. Method according to claims 3 and 4,   characterized in that, the detection of the background in the area of interest of the current image (2) is carried out according to the following steps: - an edge detection (27), preferably in column and line of the first five lines of luminance of the current image (2); - a threshold (28) of the order of 10; - a median filter (29), preferably in column and row, using 3 structuring elements; and - a test (30), of at least a first line   binary, preferably the first four.   6. Method according to claims 3 to 5,   characterized in that the background image is reconstructed, according to a copy (34) of the first line of the current image (2), in all the respective lines of the background image (11), when this first line represents the background.   7. Method according to claim 6, characterized in that a sliding average of the pixels of the first line of luminance (31) and of chrominance (33) of the current image (2), is effectuce prior to the copying (34 ) of the first line of the current image (2), in order to obtain a first homogeneous reference line.   8. Method according to claims 3 to 5,   characterized in that, the background image is refreshed according to the following steps, when (30) the area of interest of the current image (2) represents the background: - in a first part, a subtraction (37) between the average value (36) of the pixels of the first luminance line of the current image (2) and the average value (35) of the pixels of the first line of   luminance of the initial background image (3).   - in a second part, an addition (39) between the signce value of this subtraction (37), and the voltage level of each pixcl of the luminance signal of the initial background image (3), when the absolute value of this subtraction (37) is greater than a threshold value n previously defined (38), by example 5.   9. Method according to claims 1 to 8,   characterized in that in the second phase, to reduce the calculation time, the Euclidean distance (4, 12) between the representation of the background image (3, 11), defined by the luminance signals (Yf), of chrominance (Uf, Vf) and the current image (2), defined by the luminance (Ypf), chrominance (Upf, Vpf) signals, is calculated by the following chrominance relation:   dE W = [(Uf - Upf) 2 + (Vf - Vpf) 2].   10. Method according to claims 1 to 9,   characterized in that in the second phase, the Euclidean distance (4, 22) between the representation of the background image (11), defined by the signals of luminance (Yf), of chrominance (Uf, Vf) and the current image (2), defined by the luminance (Ypf), chrominance (Upf, Vpf) signals, is calculated by the following Luminance and Chrominance relation:   dE YUV = [(yf _ ypf) 2 + (Uf - Upf) 2 + (Vf - Vpf) 2].   11. Method according to claims 1 to 10,   characterized in that in the third phase, to extend its operating range, it includes an adaptive threshold (14), based on the minimum value of a histogram (13) of the Euclidean distance (12, 22) between the representation of the background image (11) and the current image (2).   12. Method according to claim 11, characterized in that the adaptive threshold (14) further comprises a dynamic operating interval (S1, S2), based   on the maximum value of the histogram (13).   13. Method according to claims 11 and 12,   characterized in that the adaptive threshold (14), based on the minimum and maximum value of the histogram (13) of the Euclidean distance (12, 22) between the representation of the background image (11) and the image current (2), is calculated on a sliding average (15) of the previous values of the adaptive threshold (14), for example last five values.   14. Method according to claims 1 to 13,   characterized in that in the fourth phase, in order to obtain a shape model (6) of the object, the construction (16, 17, 23) of a mask is carried out by the threshold binary image of the Euclidean distance (12, 22 ) between the representation of the background image (11) and the current image (2).   15. Method according to claims 1 to 14,   characterized in that in the fourth phase, to obtain intermediate tension levels in the contour region of the mask (16, 17, 23), it comprises an Alpha plane (18, 26) defined according to an equation of type: Alpha Yn = ... (An-2). (Yn-2) + (An-l). (Yn-1) + An.Yn + (An + l). (Yn + 1) ..... o An are the respective coefficients of the levels of   voltage and Yn the mask pixels (16, 17, 23).   16. Method according to claim 15, characterized in that, for obtaining intermediate voltage levels in the contour region of the mask (16, 17, 23), for example and more advantageously 5 intermediate levels for the luminance and 3 for the chrominance, the respective Alpha planes are defined according to the following equations, for luminance:   Alpha Yn3 = (An-3). (Yn-3) + (An-2). (Yn-2) + (An-l). (Yn-   1) + An.Yn + (An + l). (Yn + 1) + (An + 2). (Yn + 2) with the coefficients: An3 = 1/12, An-2 = 2/12, An- 1 = 3/12, An = 3/12, An + 1 = 2/12, An + 2 = 1/12; which gives for a step {0, 255}, the successive levels {0, 21.25, 63.75, 127.50, 191.25, 233.75, 255} for the chrominance: Wn4 = (An-2). (Wn-2) + (An -1). (Wn-1) + An. Wn + (An + l). (Wn + 1). with the coefficients An-2 = 1/6, An-1 = 2/6, An = 2/6, An + 1 = 1/6; which gives for a step of {0, 255}, the levels   successive {0, 42.50, 127.50, 212.50, 255}.   17. Method according to claims 1 to 16,   characterized in that, in the fourth phase, to reduce the calculation time of the Alpha plane (18, 26), the latter is constructed (41) in an area of interest of the mask (16, 17, 23) representing the region contour Objects.   18. Method according to claim 17, characterized in that the zone of interest, intended for the calculation of the Alpha plane (18, 26), is defined, from the contours (41) of the mask, by a change of origin ( 43) of the equation of the Alpha plane.   19. The method of claim 17, characterized in that the area of interest intended for the calculation of the Alpha plane (18, 26) is defined by the construction of a   contour mask of the mask (16, 17, 23) of the object.   20. Method according to the preceding claims,   characterized in that, in the fifth phase, to extract the objects from the current image (2), it performs a first multiplication (45) between the current image (2) and the Alpha plan (18).   21. Method for embedding objects in a scene, characterized in that it uses the method intended to extract objects from an image, on a background   uniform according to any of the claims   previous, to perform a reconstruction of the luminance (19) and chrominance (20) signals, of the image representing the scene (46), with the objects extracts from the current image (2).   22. Method according to claim 21, characterized in that in order to embed the objects in a scene (46), it comprises the following steps: - a construction of a shape model of the background by an inversion (47) of the Alpha plane ( 18) of the object; - a representation of a scene background by a second multiplication (48) between the scene image (46) and the reverse (47) of the Alpha plane (18); - a reconstruction of the image by an addition (49) between the background from the second multiplication (48) and the object of the current image (2)   from the first multiplication (45).   23. Method according to claims 21 and 22,   characterized in that to carry out a trimming (21), it comprises the following stages: - a contour detection (50) of the mask W (17); - a loading (51) of the contour values at a