FR2728987A1 - METHOD AND DEVICE FOR THE REDUCTION OF FORM INFORMATION AND ITS METHOD OF SEQUENTIAL POLYGONAL APPROXIMATION - Google Patents

Abstract

The present invention relates to a method and a device for reducing shape information for encoding and transmitting moving pictures as well as a method of sequential polygonal approximation, which are capable of predicting the motion compensation of the contour of an object in a series of images, to transmit a prediction error by region of prediction error, and to reduce the amount of information transmitted regarding the shape of an object. According to the invention, a prediction error region is determined and a thresholding is performed (means 12) to remove the information to which the eye is not sensitive, and then to determine the approximation method to be applied. to this region. Application to the coding and to the reduced rate transmission of moving images.

Description

PROCESS INFORMATION INFORMATION REDUCTION DEVICE

  FOR REDUCING SHAPE INFORMATION AND METHOD THEREOF

  SEQUENTIAL POLYGONAL APPROXIMATION

BACKGROUND OF THE INVENTION

  FIELD OF THE INVENTION The present invention relates to a shape information reduction device and method for encoding and transmitting moving images as well as a sequential polygonal approximation method for such a method, and more particularly an improved shape information reducing device and an improved method as well as a sequential polygonal approximation method for such a method, which are capable of predicting the motion compensation of the contour of an object in a series of images, to transmit a prediction error by region of prediction error, and to reduce the amount of information transmitted

concerning the shape of an object.

  2. Description of the known technique

  In a known manner, videophones, teleconferencing systems, image coding / decoding systems, and computer vision systems require transmissions of a very large volume of data at high speed,

  when transmitting or receiving image data.

  It is therefore necessary to reduce the amount of data when transmitting

  shape information, for moving pictures.

  From this point of view, M. Hostter presented an article entitled "Predictive contour coding for an object-oriented analysis-synthesis coder" in the IEEE International Symposium Information Theory, San Diego, California, USA, January 1990 , page 75, and P. Gerken presented an article entitled "Object-oriented analysis-synthesis at low bit rates". IEEE Transaction on Circuits and System for Video Technology, special issue on very low-bit rate video coding, Vol. 4, N 3, June

1994, pages 228-235.

  According to these techniques, it is sought to approximate a complete contour of a region of an object moving, to predict motion compensation

  vertices used for the approximation, and to transmit the error.

  However, when transmitting or receiving a large amount of data, the

  resolution of images deteriorates.

  The polygonal approximation, which is intended to approximate the contour of a moving object region, is adopted in object recognition techniques, object analysis techniques, and coding techniques.

  objects. The range of applications of this polygonal approximation has been expanded.

  R, 13300 \ 13392BIS DOC -29 d & amab, 1995 - / 28 To achieve the polygonal approximation of a contour, in the case of a contour having two endpoints, these two endpoints are connected by a line right, as shown in FIG. 1A, and a point which has a maximum distance to the formation approximation contour is considered a new vertex, and the ends of the contour and the new vertices are connected by

right segments.

  These operations are repeated until the maximum distance between the pixels of the contour and each straight line segment is smaller than a certain threshold value Dmax. The result is a polygon formed of segments, such as

  shown in Figure 1C and 1D.

  Furthermore, in the case of a contour without ends, we first connect the two points or pixels of the contour between which the distance is the most important, then we implement the aforementioned method, so as to obtain a

approximation polygon.

  In this case, the threshold value is the maximum error value or maximum distance between a contour and a line segment, in the case of the approximation

of an outline by a polygon.

  However, the aforementioned polygonal approximation method has the disadvantage that, when the maximum error between the distances to the straight line segments of the polygon is smaller than the threshold value, it is not accurately obtained.

the desired polygon.

  In addition, the number of vertices increases because of the large number of polygon segments required, so that it is difficult to transmit or receive

position information.

SUMMARY OF LINVENTION

  Accordingly, one of the objects of the present invention is to provide a shape information reduction device, and a shape information reduction method, with its sequential polygonal approximation method, which overcome the problems encountered in a conventional form of information reduction device, or in a conventional method, and

  in his method of sequential polygonal approximation.

  Another object of the invention is to provide an improved shape information reduction method and an improved shape information reduction device, with a sequential polygonal approximation method, which are capable of predicting the compensation due to the movement of the contour of an object, in a series of images, to transmit a prediction error corresponding to a region of prediction error, and to reduce the transmission of shape information

for an object.

  To achieve these objects, the invention provides a method of reducing shape information by object in a method of encoding and transmitting moving images, comprising the steps of comprising: (i) obtaining a prediction error region using motion information of a current frame, and a contour of a previous frame in which the motion has been compensated; (ii) extracting a prediction contour and a transmission contour from the obtained transmitted prediction error region, and calculating pixel numbers of these contours; (iii) obtaining an evaluation function, using the number of pixels of the two outlines and comparing said evaluation function with a threshold value; (iv) approximating, when the evaluation function is greater than the threshold value, a transmission contour using a polygon / spline approximation method, and without using a prediction contour due to the error in the extraction of the object; (v) extracting, when the evaluation function is less than the threshold value, the maximum error value in an encoding method using a prediction contour; (vi) comparing the extracted maximum error value and the threshold value used in the polygon / spline approximation; (vii) transmitting, in the case where said maximum error value is greater than said threshold value, the information concerning a certain vertex located in the zone where the maximum error occurs as well as the two ends of the transmission; (viii) not to transmit certain area information when the value

  maximum error is less than the threshold value.

  In one embodiment, the method further comprises a step of predicting the motion compensation of a contour, using the redundancy between the shape information of a moving region of the same object in a series of images. and transmitting a prediction error by region in which said prediction error has occurred and reducing the amount

  form information to be transmitted.

  In another embodiment, the method further comprises a step of dividing the contour of a moving object into independent error regions, when predicting a motion compensation of the shape information of a frame. preceding one according to a motion information, and a current frame according to a motion information, and coding and transmission

  motion information for a prediction error region.

  In this case, the method may further comprise a step of encoding and transmitting a certain portion corresponding to the shape information of the frame.

  current for a prediction error region.

  In one embodiment, the method further comprises a step of obtaining the value of an error with respect to a prediction contour, so as to approximate a transmission contour, and polygonal approximation of this information of error, and expression of the approximated error in a frame by

report to the prediction contour.

  In this case, the method may further comprise a step of calculating a difference in the transmission contour after obtaining the error value, and obtaining a real error value having a smaller curvature, and

  approximation of the contour on the basis of the error value thus obtained.

  It is then possible that the method further comprises an adoption step

  a polygonal approximation method, to approximate the value of the error.

  It is also possible that the method of approximating the error value consists in transmitting the error information using three vertices in a polygonal approximation of an error information, using as vertices for the approximation on the one hand the position in which the maximum error occurs

  and on the other hand both ends of the transmission contour.

  In this case, the method of reducing shape information also advantageously comprises a step of combining an approximation error and a prediction contour, in a reception terminal, and reproducing the

outline in a given frame.

  In one embodiment, the method further comprises a step of approximating an outline by separating a well extracted region and a bad region.

  extracted in the entire frame, in case of approximation of a transmission contour.

  In one embodiment, the method includes an evaluation function D of:

INT - NPI

D =

(NT + NP) / 2

  where NT is the number of pixels of the transmission contours, and NP is the number of

pixels of the prediction contours.

  In one embodiment, the method further comprises, during the transmission of a vertex, a step of transmitting an index indicating that a vertex is a representation of two vertices, when two end vertices

  two neighboring transmission contours are close.

  The invention also has a method of reducing object shape information for a motion picture coding and transmission system, comprising: R \ 13300 \ 13392BI5 DOC - 29 luncheon 1995 -4 / 2S (i) obtaining a prediction error region using motion information of a current frame and a contour of a previous frame compensated for the motion; (ii) extracting a prediction contour and a transmission contour from the transmitted prediction error region; (iii) extracting a prediction transmission contour from the transmitted prediction error region; (iv) connecting two neighboring transmission contours within a certain distance to form a single transmission contour; and (v) the polygon / spline approximation of the contour of

transmission thus obtained.

  In one embodiment, said step (iv) comprises binding an approximate final vertex of an upper contour and an initial vertex of a lower contour, when the two contours are adjacent, and the realization of a

  approximation and reduction to an approximation and a vertex.

  Another object of the invention is an object form information reduction device for motion picture encoding and transmission devices comprising: motion compensation prediction means for predicting shape information of a frame current using shape information of a previous frame and motion information of a current frame in a motion compensation method; subtraction means for subtracting from a current motion region the predicted shape information with motion compensation by said motion compensation prediction means, and for calculating an isolated prediction error region; thresholding means for performing a thresholding operation in the isolated prediction error region obtained by said subtracting means and for transmitting or blocking information for a certain time; contour approximation encoding means for predicting a contour of a predicted error region determined by said thresholding means and for encoding transmitting the shape information, with reduction of the amount of information to be transmitted; contour reforming means for reformatting a contour encoded by said contour approximation encoding means to form an isolated prediction error region; and combining means for combining the isolated prediction error region obtained by said contour reforming means and the information of the present invention.

  shape of the current frame predicted with motion compensation.

  In one embodiment, said thresholding means suppresses a region comprising information to which the human eye is not sensitive as regards the error prediction of the shape information, by thresholding based on a characteristic of the eye that is sensitive to the geometric shape of an object and to

complete movement of this object.

  In another embodiment, said thresholding means obtains the pixel difference using two binary images of the predicted shape information and the actual shape information, and generates an error region having tens to hundreds pixels and remove an error region that alternately appears by combining and subtracting the regions in a region having

a narrow outline.

  In this case, said thresholding means can suppress a low error region corresponding to the maximum region size value that is not visually perceptible in the whole images and reduce the transmission rate, unless

  It does not affect the image resolution of the user.

  In another embodiment, said thresholding means blocks some meaningless transmission of shape information, caused by a signal characteristic that is not modified by robjet movement, caused by a certain sensitivity characteristic of the signal. human eye to the shape of an entire object and to the motion of that whole object rather to the position of that object, and caused by rapid variations in the type of shape of the object generated by

  the extraction of the displacement region.

  In another embodiment, the contour approximation encoding means suppresses a prediction contour contained in an outline of a region

  of error and encode a transmission contour.

  In another embodiment, said contour prediction means determines an approximate contour for transmission obtained by the means of

  thresholding using differential coding.

  In one embodiment, said contour prediction means approximates a transmission contour obtained by the thresholding means in

using differential coding.

  Finally, a subject of the invention is a polygonal shape information reduction approximation method, comprising the steps of: (i) determining a first vertex to begin the polygonal approximation of a certain contour; (Ii) drawing a segment between the first vertex and the next pixel of the contour, and calculating the distance between the pixels of the contour lying between the ends of the segment and the segment; (iii) determining whether the calculated maximum distance to step (ii) is greater than the predetermined thresholding value; (iv) drawing a segment between the vertex and the next pixel of the contour, and calculating the distance between the pixels between the ends of the segment and the segment, when the maximum distance calculated in step (iii) ) is smaller than the predetermined threshold value; and (v) fixing the preceding pixel as a new vertex, and plotting a segment between the new vertex and the next pixel, and calculating the distance between the pixels of the contour lying between the ends of the segment and the segment, when the maximum distance calculated in step (iii) is greater than the value of

predetermined thresholding.

  In one embodiment, step (i) consists in determining that one of the two ends constitutes the first vertex when a given contour has two ends. In another embodiment, step (i) consists of determining that a certain pixel among the pixels constituting the contour is the first vertex, when

  given contour is of the endless type.

  In one embodiment, said threshold value is determined by the

  maximum effective distance between a real contour and a polygon segment.

  In another mode of implementation, said thresholding value is constituted

by the number of vertices.

BRIEF DESCRIPTION OF THE DRAWINGS

  FIGS. 1A to 1D are views of a known method of approximation

polygonal shape.

  FIG. 2 illustrates a moving image separation method for the region of the head and the shoulder region in a conventional shape information reduction device, its shape information reduction method and its

  sequential polygonal approximation method.

  FIG. 3 is a view of the shape information showing the shape information of the object on the previous frame and shape information of the object on the current or current frame, in the region of the head and in the shoulder area, in a conventional form information reduction device, a shape information reduction method and its polygonal method

sequential approximation.

  A1 is a view of a moving image compensated on an image of a previous frame and on an image of a current or current frame, in a conventional shape information reduction device, with a method

  corresponding, and its method of sequential polygonal approximation.

  FIG. 5 is a block diagram of an information reduction device of FIG.

form according to the present invention.

  FIGS. 6A to 6C are views of a method of approximation by

  polygon / spline, according to the present invention.

  FIG. 7 is a flowchart of a method of approximation by

  polygon / spline according to the present invention.

  FIGS. 8A to 8C are views showing a method of linking two

  adjacent contours of Figure 7 according to the present invention.

  FIGS. 9A to 9C are views of a method of polygonal approximation of a prediction error value using a prediction contour according to FIG.

present invention.

  FIG. 10 is a flowchart of a prediction coding method of the

  movable compensation contour according to the present invention.

  FIG. 11 is a flow chart of a polygonal approximation method

  sequential according to the present invention.

  - Figures 12A to 12E show views of approximation methods

  sequential polygons according to the present invention.

  DETAILED DESCRIPTION OF THE INVENTION

  A shape-reducing device and a shape-reduction method and its method of sequential polygonal approximation according to the present invention aim at separating an animated image obtained by object-coding a moving image into a region. of variation of the object and in a background or

  unmodified region, in which the signal is not modified.

  The background region does not require any method of information analysis or data transmission, and the previously received signal is used on the

receiving terminal.

  The moving image information is obtained using an object model and a

  displacement model in the moving region of the extracted object.

  In addition, travel information and form information from the

  mobile object region are transmitted to the receiving terminal of the system.

  The image is reproduced using compensated motion prediction, motion information, and shape information of the region of the object,

  issued from the transmitting terminal of the system.

  R \ 13300 \ 13392BIS DOC -29 db. The present invention provides improved prediction performance over a conventional block encoding method, because the displacement information is predicted as a function of the moving object, and the fact that he

  is possible to reduce block effects and increase the resolution of the image.

  Furthermore, because a form information transmission is intended to provide different displacement information to two neighboring pixels with respect to a boundary between objects, the points that appear in the methods by

  block do not appear, and there is no border distortion.

  The moving object region may include regions in which there are motion compensation prediction errors, and these regions correspond to regions in which the assumed object model is not appropriate, and in which the model movement is not appropriate. From this bridge of sight, phenomena mentioned above can occur in certain regions,

  like for example the eyes or the mouth.

  In addition, because these regions are important for users, it is necessary to code the color information so that it actually reproduces the information.

  original signals on the receiving terminal.

  When a background is discovered, because of the movement of an object, it is normally necessary to provide some information corresponding to the movement of the object, corresponding to the movement of the object, because this

  information is not provided in previously transmitted images.

  Figure 2 shows a background of a head and shoulders, a region in which the model is effective, a region in which the model is not effective, and

an open background.

  The shape information of a region of an object in motion can be expressed by a binary image or a contour with indication of a region of failure. The shape information of the region of an object on the move, for the same

  object, has a fairly high redundancy in a series of images.

  A method of contour prediction coding aims, in the above-mentioned method, to predict a contour in a motion compensation method, to transmit a prediction error, and to reduce the quantity.

  form information to be transmitted.

  The contour prediction coding method that is adopted to implement the objects of the present invention relates to a thresholding operation for selecting a transmission prediction error, so as to transmit shape information or error information concerning a region in R \ 13300 \ 13392BIS DOC - 29 of December 1995 - 9/28 that there is a prediction error, as well as an approximation coding of the contour. This method has advantages in that it makes it possible to reduce the transmission rate without deterioration of the image resolution, and makes it possible to control the coding parameters, compared to a conventional contour prediction coding. In addition, the sequential polygonal approximation method used in the present invention determines vertices in the order of the pixels of the given contours. Thus, in the case where the maximum error on a segment of a polygon is less than the threshold value, it is possible to obtain a gain, by reducing the

  number of vertices, and the amount of vertex data transmitted.

  The shape information reduction device, and the shape information reduction method with its sequential polygonal approximation method

  according to the present invention are now described in detail.

  In an object-driven image encoding operation, the shape information should be transmitted with the highest priority, together with

  displacement information for motion compensation prediction.

  The shape information transmission provides different displacement information to neighboring pixels, and has the advantage that it prevents the formation of spots or dots that appear in an object-based coding method, and border distortions, as well as the deterioration of the resolution for the user. Thus, the transmission of shape information has an appreciable feature that the object coding method at a low transmission rate has an image resolution for the user that is better, compared to

to the block coding method.

  Methods for expressing the border regions include an infographic method, a character recognition method, and an object synthesis method. For example, in the industry, a string difference coding method is well known ( in English "chaim-difference coding method"), an S-curve curve method, an approximation method

  polygonal, and the Fourier description method.

  However, because these methods are not intended for transmission, they are actually impossible to adopt in the case where the outline of an object region

  mobile is transmitted for each frame ,.

  There are many similarities in types and positions between the shape information of a moving object region obtained for the same object in a series of frames. . It is therefore possible to predict current or current shape information from past shape information. In addition, the prediction of the motion compensation with respect to the shape information is made possible by predicting the displacement information.

of a moving object.

  When the extraction of the moving object region and the motion information prediction are accurate, the transmission of the shape information is not necessary. This method of encoding or decoding shape information is called

  predictive method of contour coding.

  However, because the transmission rate is low, the ratio between this rate and the shape information is higher, and it is necessary to substantially reduce the shape information to obtain a coding gain, when compared to a sequential type coding, in which the transmission of the shape information is

not necessary.

  In this embodiment of the present invention, a thresholding operation and a contour approximation method are used to substantially reduce the shape information to select a transmission prediction error. As a result, in the case of predicting a compensation of the movement of the current shape information, regions of prediction error appear.

isolated.

  Fig. 3 shows shape information of an object in a previous frame ("N1" th frame) and a current frame ("N" th frame), and Fig. 4 shows nine isolated prediction error regions. when the shape information of the previous frame ("Nl" th frame) moves according to a motion information, and when the current frame is predicted in a method of

compensation of the movement.

  In such a case, the region corresponding to the shape information of the frame

  current for the nine prediction error regions is coded and transmitted.

  The prediction error transmission may include certain information that does not influence the human eye. Information that influences the resolution of the image for the user is transmitted, and information that does not influence the resolution of the image for the user is not transmitted, so as to ensure a

  coding with a low rate transmission.

  The present invention thus relates to the deletion of the information which does not influence the resolution of the image for the user, by the thresholding operation,

  so that this information is not transmitted.

  In this case, the thresholding operation is performed using a region in which a certain error occurs. The approximation method is adopted to transmit the contour and effectively reduce the shape information, so as to transmit the error region

  prediction that must be transmitted.

  The contour approximation method includes a polygon / spline approximation, and a polygonal approximation method that are characterized

  by approximating the prediction error value.

  Fig. 5 shows a contour prediction and motion compensation encoder for reducing shape information, which includes a motion compensation prediction unit 16 for predicting shape information motion compensation. of a current frame, using the shape information of the previous frame, a subtraction unit 11, to calculate the value of a difference between an output signal of the motion compensation prediction unit 16 and a an object region signal corresponding to the current frame, a thresholding unit 12 for calculating an isolated prediction error region by virtue of the subtraction unit 11, and for performing a thresholding operation in this region of isolated prediction error, and to transmit some information that is considered to have an influence on the human eye and to block the transmission of information that is cons which is believed to have no influence on the human eye, a contour approximation coding unit 13 for reducing the contour information by using an approximation method in the contour remission, so as to transmit the region of a prediction error that the thresholding unit 12 considered to be transmitted, a contour reforming unit 14, for reformatting in a prediction reproduction error region the contour coded by the thresholding unit 12 and the an approximate coding unit 13, and an adding unit 15 for adding the isolated prediction error region by the contour reforming unit 14, and the output of the motion compensating prediction unit 16 to calculate a shape information of the current frame and provide the information

  thus calculated at the motion compensation prediction unit 16.

  The thresholding unit and the contour approximation coding unit 13 are

  important elements in the embodiment of the present invention.

  R \ 13300 \ 13392BIS DOC - 29december 1995 - 12/21 We will now explain the prediction error region of two moving object regions, the thresholding operation of these regions, and the coding

  contour approximation for transmitting an error region.

  To begin, the thresholding operation for selecting the information error to be transmitted has an influence on the resolution for the user and the user.

amount of data transmitted.

  The thresholding operation includes adopting shape and size characteristics of the error region. From this point of view, we now explain the error of

  prediction of shape information.

  The shape information of the region of the moving object and the information of

  movement are extracted from two real images.

  That is, the process of extracting the current image is independent of

  the shape information of the previous image.

  The extraction of the moving object region and the prediction of the displacement or motion information must be accurate to allow more accurate prediction between two shape information, and this is desirable, to avoid

  that the transmission of form information is not necessary.

  However, a shape information prediction error occurs because of the limitations of the motion information prediction method or

  motion, and signal characteristics of the actual images.

  In addition, the moving object region includes a moving object, and a background

discovered.

  Since this region does not depend on motion information, it is impossible to make a motion compensation prediction, and there are

errors.

  A motion information prediction error may include information to which the human eye is not sensitive, because the human eye is sensitive to a geometric appearance and to the complete motion of an object. Therefore, according to the invention, it is provided to eliminate these errors by a thresholding operation,

and not to transmit them.

  Failure to transmit these errors can make a reduction in the actual transmission possible, while preserving the resolution of the image for the user. The thresholding operation in the contour prediction coding device according to the present invention relates to the deletion of minor error information

  and small variations in shape information.

  The contour of shape information can be considered as simple, and in general, more complex

contour by pixel.

  When making a difference per pixel, using binary images of the predicted shape information and the actual shape information, error regions having tens or hundreds of pixels are generated due to the aforementioned form information, and there may be regions of addition and

  subtraction, although the prediction is correct.

  It is therefore necessary to remove the regions of lower error, so that

  remove small error regions.

  The choice of the threshold value needed to remove the regions of lower error is likely to go up to the maximum value of the size of the region to which the human eye is not sensitive to the entire image . From this point of view, it is possible to set this value preferably from 1 to 5 PEL (picture elements) or 2 to

3 PEL.

  Thus, the removal of the lower error region having a correct threshold value makes it possible to reduce the transmission rate without having any influence on the resolution of the image for the user, and can easily be achieved by dividing the

  contour of the moving object into a plurality of independent error regions.

  In addition, the human eye is more sensitive to shape and movement

  of an object rather than a positional error.

  As a result, the human eye does not capture rapid changes in the shape of an object produced by signal characteristics, which are not in a region modified by the movement of the object, and do not capture plus the extraction of

the region of the movement of the object.

  In order to transmit the aforementioned information, a large amount of data is needed, and some of which adds nothing to the image resolution

viewed by the user.

  As a result, the variation of shape information should not be transmitted.

  The error region, which is determined to be transmitted, is encoded by its contour. The outline of the error region includes an outline (such as a

  prediction outline) of shape information.

  The contour that is transmitted is therefore an outline in which the contour of

prediction has been removed.

  The amount of transmission data is increased in the case where a chain difference coding is used, in which there is no error in the transmission contour encoding, as for an error region having a shape. of

  narrow band in images including a head and shoulders.

  Thus, in the present invention, it is necessary to adopt a method which is not suitable for use in the present invention.

  Contour coding using an approximation.

  As a result, there is a partial position error.

  However, because the human eye is sensitive to the geometric shape of an object, an image resolution likely to cause a sensation can be avoided.

  unpleasant for the human eye because of this minor position error.

  FIGS. 6A to 6C show a method of approximation by

  polygon / spline that is used for this contour approximation.

  The contour to be transmitted or transmission contour 20 of the error region to be transmitted is approximated by a polygonal method, as shown in FIG.

Figure 6A.

  The number of segments of polygon 30 depends on the transmission contour

  real 20 and an approximation polygon 30.

  This level of approximation indicates the maximum difference between the real transmission contour 20 and the approximation polygon 30. Here, when the maximum difference is large, the approximation becomes inaccurate, and the number of vertices decreases, and when the maximum difference is small, the approximation

  becomes precise and the number of vertices increases.

  Moreover, as shown in FIG. 6B, it is possible to obtain a spline 40 passing through each of the approximation vertices, and the distance between each pixel

* the spline 40 and the transmission contour 20 can be detected.

  When the distance d1 exceeds the threshold value dmax, as shown in FIG. 6B, the polygon 30 is used as an approximation of the distance for the part comprising this pixel, instead of using the spline 40. When the distance d2 is lower at the threshold value dmax, we approximate the part of the

  contour including this pixel by the spline 40, as shown in Figure 6C.

  In addition, an index indicating whether the approximation for each segment, or between each pair of vertices is a polygonal approximation or a spline approximation, is transmitted to the receiving terminal, so as to obtain a polygon

and / or a spline.

  The contour approximation in which polygons and splines are used makes it possible to obtain more natural contours with a lower number of vertices,

  compared to the polygonal approximation.

  The procedure for approximating a plurality of transmission contours is now explained, using the method of approximation by

  polygon / spline mentioned above.

  After carrying out the thresholding operation, the procedure comprises a first step of detecting a transmission contour in a region of error transmitting; a second step in which two neighboring transmission contours in a certain range and for a certain transmission contour are connected, and a third step of producing the polygon / spline approximation for the

  transmission contours thus connected.

  The above procedures will now be explained in more detail, in

reference to Figure 7.

  The transmission contour is detected in an error region to be transmitted,

  after a thresholding operation (SI).

  At this moment, there may be unnecessary vertices, between outlines

neighboring transmission systems.

  In order to solve this problem, a step of connecting two

  neighboring contours in a certain range and a certain contour (S2).

  In addition, a polygon / spline approximation is performed for the contours

thus connected (S3).

  FIGS. 8A to 8C show a procedure for connecting two contours

  neighbors and some transmission outline.

  As shown in FIG. 8A, when two transmission contours are adjacent, an approximated final vertex of the upper contour and an approximated initial vertex of the lower contour are connected, and these vertices are approximated again, so as to remove a final approximated vertex and an approximated peak

  initially, which reduces the amount of information to be transmitted.

  The classical contour prediction coding method consists of approximating the set of contours in a polygon or spline method, and predicting in a motion compensation method using the vertices

approximated.

  The polygon / spline approximation method of the present invention, on the other hand, aims at approximating by dividing the entire contour into a plurality of transmission contours. In addition, it aims to approximate the reduction of the amount of transmission by removing regions of low error and

  approximating the plurality of smaller transmission contours.

  In addition, it is possible to control the transmission of shape information by using a threshold value that can be provided for each transmission contour, and a threshold value that is used for a plurality of contours.

transmission smaller.

  Figures 9A and 9C show a method of polygonal approximation of error information using a prediction contour, which is another method

for the approximation of an outline.

  R \ 13300 \ 13392BIS DOC -29 deb ,, e 1995 -16/28 Indeed, even if the extraction of the region of the moving object is relatively well done, the prediction contour and the transmission contour in a region error that occurs because of the problems of predicting information from

  movement have similar shapes.

  In this case, it is possible to approximate the transmission contour by

  reformatting, using a prediction contour.

  This method aims to reduce the number of vertices that are

  necessary to approximate the transmission contour.

  Figs. 9A-9C show an approximation method using a prediction contour. Fig. 9A shows an approximation method for a transmission contour 51 of an error region E, which results from the motion information prediction rather than the motion object region prediction. FIG. 9B, the error between the transmission contour 51 and the prediction contour 52 can be obtained, and the error value is reformatted as a function of a curvilinear index or abscissa on the prediction contour 52, so that the information corresponding to the transmission contour 51 is not lost, and it is thus possible to transform the transmission contour 51 of FIG.

  contour 53 having a smaller radius of curvature.

  The outline of Figure 9B is approximated by a plurality of vertices as

represented in FIG. 9C.

  It is possible to obtain an apparent contour approximating the transmission contour 51, by adding the approximated transmission contour 54 to the prediction

  of contour 52 in the original image frame.

  The purpose of this method is to reduce the need for a spline approximation to obtain a more natural appearance, since the apparent outline depends on

a prediction outline.

  The thresholding operation of FIG. 5 presents a region in which the approximation is well carried out, and a region in which the approximation is not well carried out, as a function of the contour prediction error region in FIG.

the whole of a frame.

  In this case, another method of approximation can be adopted in each region, which comprises a first step, in which the number of pixels is calculated by detecting a transmission contour and a prediction contour, a second step in which one judges if the extraction of an object has a desired region using the number of pixels of these two outlines, a third step in which a transmission contour is coded using a polygon / spline approximation method, because this is not a desired region, that is to say that the extraction of the object is not well carried out, a fourth step that R 13300 \ 13392BIS DOC -29devebb 1995- 17/23 extract the value of maximum error using a polygonal approximation method of error information, using a prediction contour, when the region has undergone a correct object extraction, a fifth step in which it is judged whether the error value max extracted is greater than a threshold value dmax, obtained in the polygon / spline approximation, and a sixth step in which one approximates a position in which the maximum error occurs, as well as both ends of the transmission contour, as approximated vertices in the case

  o the maximum error value is greater than the threshold value dmax.

  The above-mentioned steps are now explained in more detail, with reference

in Figure 10.

  First, a transmission contour and a prediction contour are detected in the transmission error region, and the number of pixels (Si) is counted. A judgment is made to determine whether the extraction of the object is

  performed using the number of pixels of these two outlines (S2).

  At this time, an evaluation function D that provides a reference for this judgment is obtained using the following expression:

NT - NPI

D = -----

(NT + NP) / 2

  where NT is the number of pixels of the transmission contour and NP is the

  number of pixels of the prediction contour.

  In the region where the extraction of the object is not well done, the

  distance between two pixels is important, and the evaluation function D increases.

  For a given threshold value, if the evaluation function D determined is greater than the threshold value Dth, it is judged that there is a problem in the extraction of the object, and a method of approximation is adopted. to polygon / spline, which does not consider the prediction contour, to encode the transmission contour (S3). When the evaluation function D of the error region is smaller than the threshold value Dth, an encoding method is adopted using a prediction contour, which comprises a step of extracting the maximum error value MAX ERROR (S4), and a step of determining whether the maximum error value extracted is greater than the threshold value Dmax used in the approximation by

polygon / spline (S5).

  The region in which the error value MAX ERROR is lower than the threshold value dmax is not transmitted because the approximation error is small. This corresponds to an error region in the form of a long narrow band. When the maximum error value MAX ERROR is greater than the threshold value d transmission outline are considered as highs

  approximated, and the values of the three vertices (S6) are transmitted.

  The method of encoding the transmission contour using a prediction contour does not include a step of combining a neighboring error region, so as to maintain a uniformity of the error characteristic, with respect to the

  polygon / spline approximation method.

  In the case where the vertices corresponding to the two ends of two neighboring transmission contours are close, two vertices are represented by a

  only vertex, and we transmit an index that corresponds to these.

  In addition, a method of polygonal approximation of contour-based error information has the advantage that a transmission reduction and spline approximation are not realized, compared with the conventional method, when the we approximate by three vertices for each transmission contour. Fig. 11 shows a flow chart of a sequential polygonal approximation method according to the present invention, which comprises a first contour detection step (Si 1) for starting a polygonal approximation and selecting one of the two pixels for the case of a contour having two ends and the determination of the pixel thus selected as the first vertex, and the selection of a given pixel in the case of an endless contour, and

  determining that the pixel thus selected constitutes a first vertex (i.e.

  say VIl1, v, I ,: o "Cn" indicates the "n" th contour pixel or its position, and "Vn" indicates the "nth vertex or its position", a second step of connection by a straight line between the pixel Cc determined to be the first vertex ("C") of the first step, and the contour pixel C (c + 1) of the first step (Si), a third step of calculating the distance (d (c + j), forj = 1 to it, which is the distance between dn and Cn) of the pixels of the contour between the two ends Cc and C (c + 1) and the straight line formed in the second step, a fourth step (S4) of determining whether the maximum distance d (c + j) obtained in the third step is greater than the threshold value dmax, and of passing to the next pixel (i + l1) when the maximum distance is not greater than at the threshold value, and then leading back to the second step (S2), and a fifth step (S5) which determines the immediately preceding pixel as the new vertex, then sque the maximum distance (d (c + j) obtained in the third step is greater than the threshold value (Vv (c + il), v + l, c + i-1, I) and realization again of the second step (S2) starting from the new vertex, the operation ending when the number R '13300 \ I13392 1S DOC - 29 december 1995- 19/2 total contour to approximate by polygonal approximation is reached (c + i = PEL

  contour, o "PEL contour" is the total number of contours to be approximated).

  When an outline such as that of FIG. 12 is provided, the pixel of this contour which will be approximated by polygon is detected. When the given contour is a contour having two ends, one of the two pixels is chosen, and in the case of a

  contour without end, we choose a certain pixel of the contour.

  The pixel thus selected becomes a first vertex. Next, the first vertex and the third pixel are connected by a straight line and the distance between the pixels of the contour (here, reference is made to the second pixel of the contour) and the straight line drawn between the pixels. two ends (i.e. between the first and third pixel of the contour) is calculated, and it is determined whether the maximum distance to this straight line is greater than the given threshold value. When the maximum distance is less than the threshold value, as shown in FIG. 12A, the first pixel and the fourth pixel are connected by a straight line, and the distance between the edge pixels (the second pixel of the contour and the third pixel). pixel of the contour) and the straight line thus obtained between the two ends (the

  first pixel of the contour and the fourth pixel of the contour) is calculated.

  When applying the previous operation many times, and when performing a straight line with the "n" th pixel, and the maximum distance of the pixels from the contour to the straight line is greater than the threshold value as shown in Fig. 12B, the pixel "n-1" becomes a new vertex, as shown in Fig. 12C, and the preceding process is repeated, until the

  contour is completed, as shown in Figures 12D and 12E.

  The number of vertices of the polygonal approximation of the present invention is decreased relative to the number of vertices in a conventional polygonal approximation method, since the maximum distance of the actual contour to each of the

  segments of the polygon is the threshold value.

  In particular, when the vertex position information is transmitted in a video encoding, since the number of vertices is decreased, one can

get coding gain.

  When the threshold value of the number of vertices (not the threshold value) is known, the polygonal approximation method is carried out until the

  number of vertices correspond to this given number.

  It is therefore possible to achieve the objects of the present invention more easily in the

  polygonal approximation method.

  In addition, for the polygonal approximation, two threshold values are available. One of these is the maximum actual distance dmax between the contour

  real and one polygon segment, and the other is the number of vertices.

  R \ 13300 \ 13392BIS DOC - 29 december 1995 - 20/28 The first aims to control the precision of the approximation polygon, and the second allows to control more easily the quantity of

data to be coded.

  When coding the shape information of an object, because it is its visual characteristic that is the most important, accuracy is more important than the number of vertices. From this point of view, the polygonal approximation method achieves a high coding gain, compared to the method

  classic, because one reaches the threshold value.

  Although the embodiments of the present invention have been described for purposes of illustration, those skilled in the art will appreciate that various modifications, additions and substitutions are possible without departing from the spirit of the invention.

  the invention described in the appended claims.

  R I 3300 \ 13392B [SDOC - 29 December 1995 - 2I13

Claims (21)

  A method of reducing shape information by object in a motion picture coding and transmission method, comprising the steps of: (i) obtaining a prediction error region using motion information of a current frame, and a contour of a previous frame in which the movement has been compensated; (ii) extracting a prediction contour and a transmission contour from the obtained transmitted prediction error region, and calculating pixel numbers of these contours; (iii) obtaining an evaluation function, using the number of pixels of the two outlines and comparing said evaluation function with a threshold value; (iv) approximating, when the evaluation function is greater than the threshold value, a transmission contour using a polygon / spline approximation method, and without using a prediction contour due to the error in the extraction of the object; (v) extracting, when the evaluation function is less than the threshold value, the maximum error value in an encoding method using a prediction contour; (vi) comparing the extracted maximum error value and the threshold value used in the polygon / spline approximation; (vii) transmitting, in the case where said maximum error value is greater than said threshold value, the information concerning a certain vertex located in the zone where the maximum error occurs as well as the two ends of the transmission; (viii) not to transmit certain area information when the value   maximum error is less than the threshold value.   The method of claim 1, further comprising a step of predicting the motion compensation of a contour, using the redundancy between the shape information of a moving region of the same object in a series of images and transmission of a prediction error by region in which said prediction error occurred and reduction of the amount   form information to be transmitted.   The method of claim 1, further comprising a step of dividing the contour of a moving object into independent error regions, when predicting a motion compensation of the shape information of a frame. preceding a motion information, and a frame R 13300 \ 13392B15 DOC - 29 decomble 1995 - 22/28 current according to a motion information, and coding and transmission   motion information for a prediction error region.   4. The method of claim 3, further comprising a step of coding and transmitting a certain portion corresponding to the information of   shape of the current frame for a prediction error region.   The method of claim 1, further comprising a step of obtaining the value of an error with respect to a prediction contour, so as to approximate a transmission contour, and polygonal approximation of that information of error, and expression of the error approximated in a frame by report to the prediction contour.   The method of claim 5, further comprising a step of calculating a difference in the transmission contour after obtaining the error value, and obtaining a real error value having a greater curvature.   low, and contour approximation based on the error value thus obtained.   The method of claim 6, further comprising a step of adopting a polygonal approximation method, for approximating the value of the error. The method of claim 6, wherein the method of approximating the error value comprises transmitting the error information using three vertices in a polygonal approximation of error information. using as vertices for the approximation on the one hand the position in which the maximum error occurs and on the other hand the two ends of the transmission contour.   The method of claim 8, wherein the shape information reduction method further comprises a step of combining an approximation error and a prediction contour, in a receiving terminal, and   contour reproduction in a given frame.   The method of claim 1, further comprising a step of approximating an outline by separating a well extracted region and a bad region.   extracted in the entire frame, in case of approximation of a transmission contour.   A method according to claim 1, comprising an evaluation function D of: INT-NICs D = (NT + NP) / 2   where NT is the number of pixels of the transmission contours, and NP is the number of pixels of the prediction contours.   12. The method of claim 1, further comprising, during the transmission of a vertex, a step of transmitting an index indicating that a vertex is a representation of two vertices, when two end vertices.   two neighboring transmission contours are close.   A method of reducing object shape information for a motion picture encoding and transmission system, comprising: (i) obtaining a prediction error region using a motion information information; a current frame and a contour of a previous frame compensated for the movement; (ii) extracting a prediction contour and a transmission contour from the transmitted prediction error region; (iii) extracting a prediction transmission contour from the transmitted prediction error region; (iv) connecting two neighboring transmission contours within a certain distance to form a single transmission contour; and (v) the approximation by the polygon / spline methods of the contour of transmission thus obtained.   The method of claim 13, wherein said step (iv) comprises binding an approximate final vertex of an upper contour and an initial vertex of a lower contour when the two contours are adjacent, and the   making an approximation and reducing to an approximation and a vertex.   An object form information reduction device for motion picture encoding and transmission devices comprising: motion compensation prediction means (16) for predicting shape information of a current frame using shape information of a previous frame and motion information of a current frame in a motion compensation method; The subtraction means (11) for subtracting from a current motion region the predicted shape information with motion compensation by said motion compensation predicting means (16); motion, and to calculate an isolated prediction error region; thresholding means (12) for performing a thresholding operation in the isolated prediction error region obtained by said subtracting means (11) and for transmitting or blocking information for a certain time; contour approximation encoding means (13) for predicting a contour of a predicted error region determined by said thresholding means (12) and for encoding transmitting the shape information, with reduction of the amount of information to be issued; the contour reforming means (14) for reformatting a contour encoded by said contour approximation encoding means (13) to form an isolated prediction error region; and combining means (15) for combining the isolated prediction error region obtained by said contour reforming means (14) and shape information of the predicted current frame with motion compensation. The device of claim 15, wherein said thresholding means (12) suppresses a region comprising information to which the human eye is not sensitive with respect to the error prediction of the information of form, by thresholding based on a characteristic of the eye that is sensitive to the   geometric form of an object and the complete movement of that object.   The device of claim 15, wherein said thresholding means (12) obtains the pixel difference using two binary images of the predicted shape information and the actual shape information, and generates a region of error with tens to hundreds of pixels and remove an error region that appears alternately by combining and subtracting regions   in a region with a narrow outline.   The device of claim 15, wherein said thresholding means suppresses a low error region corresponding to the maximum region size value that is not visually perceptible in the entire images and reduces the transmission rate, unless it affects the image resolution of the user. R I1 3300 \ 13392815 DOC 29 decemnbr 1995 - 25/25 t 26 2728987 F '26   The device of claim 15, wherein said thresholding means blocks some meaningless transmission of form information caused by a signal characteristic that is not modified by the movement of the object, caused by a certain sensitivity characteristic of the human eye to the shape of an entire object and the movement of this whole object rather to the position of this object, and caused by rapid variations in the shape of the object generated by the object. extraction of the displacement region 20. The device of claim 15, wherein the contour approximation encoding means (13) suppresses a prediction contour contained in   an outline of an error region and encode a transmission contour.   The device of claim 15, wherein said contour prediction means (13) determines an approximate contour for transmission.   obtained by the thresholding means (12) using a differential coding.   The device of claim 20, wherein said contour prediction means (13) approximates a transmission contour obtained by the   thresholding means (12) using differential coding.   23.- Method according to one of claims 1 to 14, wherein the métiode   polygonal approximation comprises the steps of: (i) determining (S1) a first vertex to begin polygonal approximation of a certain contour; (ii) plotting (S2) a segment between the first vertex and the next pixel of the contour, and calculating (S3) the distance between the pixels of the contour lying between the ends of the segment and the segment; (iii) determining (S4) whether the maximum distance calculated in step (ii) is greater than the predetermined thresholding value; (iv) plotting (S2) a segment between the vertex and the next pixel of the contour, and calculating (S3) the distance between the pixels lying between the ends of the segment and the segment, when the calculated maximum distance at step (iii) is smaller than the predetermined threshold value; and (v) fixing (S5) the preceding pixel as a new vertex, and plotting (S2) a segment between the new vertex and the next pixel, and calculating (S3) the distance between the pixels of the contour between the ends of the segment and the segment, when the maximum distance calculated at rec (iii) is   greater than the predetermined threshold value.   The method of claim 23, wherein step (i) comprises determining that one of the two ends constitutes the first vertex when contour given at two ends.   25. The method according to claim 23, wherein step (i) consists in determining that a certain pixel among the pixels constituting the contour is the first   vertex, when a given contour is of endless type.   The method of claim 23, wherein said threshold value is determined by the maximum effective distance between a real contour and a polygon segment. The method of claim 23, wherein said thresholding value   is the number of vertices.   The 1330M33921IDOC -26 I 1996 - 13:41 - 27S