KR19980073963A - Vertex Position Adjustment Method for Retention Vertices in Contour Polygon Approximation during Shape Information Coding - Google Patents

Abstract

The present invention is applicable to a contour predictive coding method that performs temporal prediction for efficient compression encoding of a contour of a binary mask. The present invention relates to a method of adjusting the position of a vertex for a holding vertex in contour polygonal approximation when shape information encoding is made as a new holding vertex while minimizing a given error measure while moving a holding vertex in a constant search window. In the case of the maintenance vertex with the elimination vertex at the adjacent position, the error center vertex search method is not applied, and the motion information of the maintenance vertex can be encoded by using the chain differential coding method, and the motion information of the maintenance vertex is general vertex position. It can be encoded using an encoding method.

Description

Vertex Position Adjustment Method for Retention Vertices in Contour Polygon Approximation during Shape Information Coding

The present invention relates to a shape information encoding method. In particular, in MPEG-4, contour prediction coding is performed to perform temporal prediction for efficient compression encoding of a contour of a binary mask. The contour coding method is applicable to the contour coding method, and when the holding point is determined, the moving point is moved within a constant search window to minimize a given error measure. A vertex position adjustment method for a holding vertex in a polygon approximation.

In general, an object-based moving image coding method is frequently used to encode an ultra low speed video signal.

The object-oriented video encoding method divides an input video signal into a background that does not move and a changed region that is generated due to the movement of an object.

The motion objects of the change area are separated into motion compensated objects when motion estimation and motion compensation are possible, and separated into motion failed objects when motion estimation and motion compensation are impossible.

Herein, the movement compensable object refers to an object moving with a certain principle such as horizontal movement, rotation movement, and linear movement in a state in which an object in three-dimensional space is converted into an image of a two-dimensional object. An object that does not apply to certain principles.

When the image of a certain object is transmitted, the motion compensable object transmits only motion information of the image, and the image of the motion compensable object and the exposed object should be encoded and transmitted in the most effective way. Since the amount is about 60-70% of the total transmission amount, much effort is being made to reduce it efficiently.

Since the image of the object having the shape information and / or the shape information has a large amount of data, the storage capacity should be very large when storing on the recording medium, and when transmitting, it takes a lot of transmission time, making it difficult to transmit in real time.

Therefore, the image of the object is encoded, the motion is estimated, the amount of information is reduced, stored in the recording medium, and transmitted in real time.

A block discrete cosine transform (hereinafter, abbreviated as 'DCT') and a vector quantizer may be used to encode the signal information of an object, and recently, a shape adaptive DCT (hereinafter, referred to as 'DCT') is used. SADCT 'has been proposed to be very effective in the object center coding method using shape information of an object.

SADCT aims to efficiently encode signal information of a predetermined object having an arbitrary shape. The SADCT subdivides an image frame into blocks having a predetermined size, and then encodes only signal information of an object entering the block.

In other words, if the block is filled with all the information to be encoded, the two-dimensional block DCT and the coding efficiency are the same.If not, the SADCT performs DCT on the X axis only for signals corresponding to the region of the object. The final result is obtained by performing one-dimensional DCT on the Y axis again.

When considering the DCT coding gain, the SADCT is more efficient to transform the transform coefficients when the image of the object is filled in the block as much as possible than when the image of the object is separated and encoded.

Therefore, when performing SADCT, it is preferable to effectively fill the image of the object to be encoded in the block and to reduce the number of blocks in which the image of the object that cannot be compensated for motion is located before encoding.

Recently, WG11 under ISO / IEC, which is a global standardization organization, is performing standardization work on MPEG-4, a method of encoding an object having arbitrary shape information unlike MPEG-1,2.

To this end, research on contour polygon approximation for shape information encoding of binary masks is being conducted.

In addition, peripheral research to improve the compression performance is also being performed.

A representative example of the conventional contour prediction encoding method is a vertex list update method as shown in FIGS. 1 to 3.

This method first performs non- / motion compensation on the current frame using the vertices used to approximate the contour 1 in the binary frame, as shown in FIG. As shown in FIG. 2, it is determined from each predicted vertex 2 whether the distance to the current contour 1 satisfies a given accuracy and determines whether to remove (deleted vertex) or maintain (maintained vertex).

Then, as shown in Fig. 3, the retained vertices 3 are used as initial vertices, and new vertices 4 are inserted until the given accuracy is satisfied.

This vertex list correction method has an advantage that it is possible to effectively reduce the amount of contour information.

However, when the holding vertex 2 is used as it is in the predicted position as shown in FIG. 4, a new insertion vertex 4 may occur at a position immediately adjacent to the holding vertex 2, thus reproducing approximation such as a dotted line. There is a problem that the line 5 is generated.

This problem is quite disadvantageous not only in terms of information compression but also in terms of subjective quality.

The present invention solves the problems of the conventional vertex list correction method as described above, and an object of the present invention is to maintain a vertex in a contour polygon approximation for adjusting a maintenance vertex in a vertex list correction method, which is a kind of contour prediction coding scheme. The present invention provides a method for adjusting the vertex position.

Such an object of the present invention can be achieved by performing the maintenance vertex adjustment by the error-centered vertex search method in the vertex list correction method, which is a kind of the contour prediction coding method.

In addition, an object of the present invention can be achieved by setting the size of a search window to 3 × 3 or 5 × 5 when searching for a maintenance vertex as an error-centered vertex in a vertex list correction method, which is a kind of contour prediction coding.

An object of the present invention can be achieved by not applying an error-centered vertex searching method in the case of a vertex list correction method, which is a kind of contour prediction encoding, in the case of a maintenance vertex having an elimination vertex at an adjacent position when the maintenance vertex is adjusted.

In addition, an object of the present invention is achieved by transmitting information on correction of a maintenance vertex to the receiver as flag and motion information in a vertex list correction method, which is a kind of contour prediction coding scheme.

In addition, an object of the present invention is achieved by encoding motion information of a sustained vertex using a chain encoding method in a vertex list correction method, which is a kind of contour prediction encoding method.

In addition, an object of the present invention can be achieved by encoding the motion information of a maintenance vertex using a general vertex position encoding method in a vertex list correction method, which is a kind of contour prediction encoding method.

1 is a view for explaining a prediction vertex in the conventional vertex list correction method

2 is a view for explaining a removal vertex in the conventional vertex list correction method

3 is a view for explaining the insertion vertex in the conventional vertex list correction method

4 is an overall explanatory diagram of a conventional vertex list correction method.

5 is a block diagram illustrating the contour polygon approximation in the P-VOP encoding according to the present invention.

6 is an operational flowchart of the contour polygon approximation of FIG.

* Description of the symbols for the main parts of the drawings *

10: number setting unit 20: contour extraction unit

30: contour polygon prediction encoder

Hereinafter, with reference to the accompanying drawings, an embodiment of the vertex position adjustment method for the holding vertices in the contour polygonal approximation when the shape information encoding of the present invention will be described in detail.

FIG. 5 is a diagram illustrating the overall configuration of contour polygon approximation of general P-VOP (Predictive VOP) encoding, and includes a number setting unit 10 for performing number labeling on an object VOP to be encoded; Contour polygon prediction which extracts the contour of each region (consisting of the same numbered pixels) and contour contour extracted by the contour extractor 20 to perform contour polygon prediction coding. The encoder 30 is configured.

In other words, contour polygon approximation of P-VOP encoding performs region laveling on the object or region to be encoded first, and then extracts the contour of each object or region, and then extracts the contour with the contour extracted from it. Perform approximation

FIG. 6 is a flowchart illustrating an operation of the present invention in which the contour polygon approximation block of FIG. 5 is divided into a VOP portion and a contour-unit, and each signal processing sequence is shown.

First, the VOP motion is estimated (S1), and the VOP motion mode is determined by using four vertices, a mode according to contour matching movement, or a no MC mode, and then the VOP motion is corrected (S2).

Then, it is determined whether or not to encode the VOP (S3). If there is the same VOP, the operation is immediately finished without encoding at all, and when encoding, similar contours are found and the motion of the contours is estimated (S4, S5). .

Here, all contours are equally unchanged where the VOP itself does not change. That is, if the change of the entire VOP is less than or equal to a certain level and the sickle coding state is used as it is, the coded / not coded need not be performed for each outline.

Where the VOP change is constant or not, rather than the contour change, all the contour information can be controlled with the VOP information.

In other words, giving a signal across the VOP eliminates the need to signal each contour.

On the other hand, after estimating the contour motion in step S5 (S6) to determine whether the intra mode or Inter mode (S6) is too poor to fit the intra mode if it does not fit well in this case the intra mode vertex Select and encode the output to the multiplexer (S7), perform prediction detection in the state where the vertex is selected, and encode the error to output the multiplexer (S8).

In this case, the step S8 is performed to transmit the error to the receiver when the vertex is selected as an error.

On the other hand, if it is not the intra mode in step S6, since it is an inter mode, it is determined in this case whether it is a motion compensation (MC) mode (no MC) (S9). If not, go to the step of determining whether the VOP is complete.

In the state of performing the motion compensation mode, it is determined whether to perform encoding again or not (S11). If the encoding is not performed, it is determined whether the VOP is completed. If the encoding is performed, vertex rearrangement is performed. After (S12), the holding vertex is adjusted and the vertex is inserted (S13, S14).

Next, it is determined whether the VOP is over (S15). If not, the process goes to the step S4 and finds the contour again.

When the maintenance peak is determined in step S13 of FIG. 6, when the maintenance peak is moved within a predetermined search window, an error-centered vertex search method is performed by minimizing a given error measure. do.

This concept is similar to the concept of finding a motion vector so that the mean absolute error is minimized within the search range when estimating motion information in units of blocks.

That is, a detailed description of the error vertex searching method centering on the k-th holding vertex V _k is as follows.

First, the number of holding vertices is N _V , and the two adjacent holding vertices are V _k-1 (V _Nv-1 when k = 0) and V _{k + 1} ( _V when k = N _{V-1). 0} ).

The point on the outline that maximizes.

Is the distance between two vertices, V _k-1 and the straight line connecting V _k and the i th contour pixel, C _i .

At this time, V _k (a, b) = ((V _{k, X} -a), (V _{k, y} -b)) -ω≤a, b≤ω (1)

The proposed method

Is to find a and b.

Then, in the k-th vertex V _k instead of V _k (a, b).

Ω in (1) determines the size of the search window, and ε {·} depends on a predetermined error measurement.

In the present invention, the following two error measuring means are used.

V _k-1 ≤ jV _{k, i} and V _{k, i} ≤ jV _{k + 1, i} (5)

V _{k, i} indicates how many contour pixels the kth vertex is.

Since the error-centered vertex search method does not guarantee the effect when the distance from the adjacent vertices is far, the error-centered vertex search method is not performed in the case of the maintenance vertex in which the immediately adjacent vertex is removed.

If the position is shifted when the error center vertex searching method is performed for the maintenance vertex, information about the vertex should be transmitted to the receiving end.

The information on the adjustment method of the holding vertex is composed of a flag of whether the position has been moved and movement information of how the movement is performed.

In case of maintaining the position as it is (in case of the maintenance vertex with the removal vertex around, or the position does not change when performing the error center vertex search), only the flag is transmitted.

Otherwise, the motion information is transmitted along with the flag. As a method of encoding motion information, a method generally used for vertex position encoding may be used.

However, it is dangerous to increase the size of the search window because there is a risk that this information amount increases when the search window becomes large.

If the size of the search window is 3x3 or 5x5, the motion information can be encoded using chain differential coding, so that the information can be encoded with a small amount of information.

As described above, the present invention can be applied when performing the vertex list correction method for compressing the amount of information by eliminating redundancy on the time axis of the binary mask.

In addition, the number of inserted vertices decreases by about 50% and the amount of inserted vertex transmission information decreases by about 50% .In addition, the number of inserted vertices decreases the total number of vertices representing shape information and reduces the amount of flag information. You can get the effect.

In addition, the subjective image quality can be improved, and the effect of reducing the reproduction error can be obtained.

Claims (11)

In the contour prediction encoding method for numbering an object or region to be encoded, extracting the contour of each object or region, and performing contour polygon approximation with the extracted contour, the VOP motion mode is estimated in VOP units and VOP Determining and correcting the motion, finding the contour in the contour unit, estimating the contour motion, rearranging the vertices by performing encoding in the motion compensation mode in the inter mode, and maintaining the vertex when the maintenance vertex is determined. A method of adjusting the position of a vertex for a holding vertex in contour polygonal approximation during shape information encoding, characterized in that it comprises a step of minimizing a given error measurement while moving within a predetermined search window. The method of claim 1, wherein the adjustment of the holding vertex is performed by an error-centered vertex searching method. The method of claim 2, wherein the size of the search window is 3 × 3 when the error center vertex is searched. The method of claim 2, wherein the size of the search window is 5 × 5 when searching for an error center vertex. 3. The method of claim 2, wherein a search for error center vertices is not performed when the remaining vertices have elimination vertices at adjacent positions. The method according to claim 3 or 5, wherein the motion information of the holding vertices is encoded using a chain differential coding scheme. 6. The method according to claim 4 or 5, wherein the motion information of the holding vertex is encoded using a general vertex position coding method. The method of claim 2, wherein if the position of the holding vertex does not change during error center vertex searching, only a flag is transmitted to the receiving end. The method of claim 2, wherein when the position of the holding vertex changes during the error center search, information about the correction of the holding vertex is transmitted to the receiving end as flags and motion information. How to adjust the vertex position for. The method of claim 2, wherein the error measurement, A vertex position adjustment method for holding vertices in contour polygonal approximation during shape information encoding. The method of claim 2, wherein the error measurement, A vertex position adjustment method for holding vertices in contour polygonal approximation during shape information encoding.