KR100307621B1 - Method for encoding and decoding contour line based on base line - Google Patents

Abstract

PURPOSE: An outline encoding and decoding method based on the base line is provided to compensate motion efficiently by sampling the contour line data changing slowly, and to improve performance by keeping the contour line sample continuously. CONSTITUTION: A base line is decided for sampling the outline(60). The sampling rate is extracted according to the size of quantization step(61). The contour line is sampled according to the decided base line and the sampling rate, and the shape of the object is reconstructed(62). Data is generated with extracting outmost samples(63), and the reconstructed shape is generated with using the outmost sample data(64). The shape of the object is subtracted from the outmost sample data(65), and samples of residue contours are extracted(66). Error samples are arranged as non-shape data, and encoded by processing. Overlapping is prevented, and motion is compensated with sampling the contour line.

Description

Baseline Based Outline Coding and / or Decoding Method

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image encoding method, and more particularly, to a reference line-based outline encoding and / or decoding method.

The encoding of the outline of an object plays an important role in the encoding of an object in a still image or a moving image. In particular, in the case of low bit rate encoding such as MPEG4, the outline needs to reduce the burden of coding as much as possible, so the compression effect should be high while reproducing the original outline in a similar manner. In addition, since the outlines are continuous in most cases, they can be composed of one-dimensional data that changes slowly, and thus are suitable for encoding using transforms.

The outline coding method generally used until now is a chain coding method which is widely used in computer vision. The code is defined in advance for each direction, and a code defined in advance for the tracking direction while tracing the outline in a predetermined direction And the like. This chain coding method has the advantage that it can reproduce the same as the original outline because it does not lose information about the outline, but it has a drawback that it takes a large amount of bits generated by encoding and time redundancy in a sequence like video . In order to overcome the disadvantages of such a chain coding method, a method of compromising the reproducibility and the bit generation amount has been proposed. However, in this case, the amount of bits can be reduced, but the fundamental drawback of the chain coding method can not be overcome.

There are many applications in the field of image processing that do not require very accurate outline representation. In this application field, since the chain coding method having a large bit generation amount is a large burden, effective outline encoding methods are not proposed in the present invention. One of these methods is a method of approximating a polygon inside or outside with respect to an entire outline related to one object as a method of a polygonal approximation. However, the polygonal approximation method has a disadvantage in that the reproducibility is poor, though it is very effective in terms of bit generation.

A spline approximation method has been used to overcome this drawback. Conventional polygonal approximation method approximates a straight line to an outline, whereas a spline approximation method approximates to a curve to realize higher reproducibility with respect to the same bit generation amount.

On the other hand, a coding method having high reproducibility for a more complex outline is developed while maintaining the efficiency of the polygonal approximation method in the bit generation amount by using Discrete Sine Transform (DST) instead of the spline technique. In this method, the outline is first approximated by the polygonal approximation method, the difference between the polygon and the outline is sampled at a predetermined number per one straight line of the polygon, and the sampled error value is encoded by performing DST and quantization. According to this method, not only a small amount of bits are generated but also a more complex outline can be reproduced.

However, all of the above-mentioned various outline encoding methods are methods for performing outline encoding on a still image. Since each object in a video is in a similar shape and position within an adjacent time, the outline of each object also has temporal redundancy as in a general video. Therefore, the encoding efficiency can be improved by encoding the outline of the moving image by using temporal redundancy. An example of an outline coding method in a moving picture is to estimate a motion of an object of a k-th frame by motion estimation through motion estimation for an object of a k-th frame, and compensate for the motion, And then encodes it through a chain coding method. However, in this method, when the outline of the generated objects changes very complexly, since the outline to be encoded is generated in a small length, the start point coordinates of the outline must be frequently encoded, which is an inefficient disadvantage.

In order to overcome this disadvantage, a center-point-based outline encoding method for measuring the distance from the center point to the outline with respect to the center line in the outline encoding of the object and encoding the measured distance using the DCT is described in There is a bar. In such a center-point-based outline encoding, the distance between adjacent outline samples selected from the center point is often very different. That is, a very large number of samples are extracted in the case of the outline near the center point, and a very small number of samples are extracted in the outline far from the center point. Therefore, if the outline far from the center point has a complicated shape, the angle for sampling must be dense in order to accurately recover it. In this case, unnecessary samples are extracted at the near outline, which causes the amount of bits generated in the encoded data to increase.

Accordingly, it is an object of the present invention to provide an outline sampling method for sampling an outline using an object baseline in order to solve the above-mentioned problems.

It is still another object of the present invention to provide an outline encoding and / or decoding method for encoding and / or decoding outline data sampled by the outline sampling method.

According to an aspect of the present invention, there is provided an outline sampling method comprising: determining a reference line and a sampling interval for an object in a still image or a moving image; Generating outline data by sampling an outline of the object using the reference line and a sampling interval, and restoring the shape of the object from the outline data; Extracting outermost samples of the object from the sampled outline data; Restoring an outline using the outermost samples; And extracting error samples by obtaining a difference between the reconstructed outline from the shape of the reconstructed object.

According to an aspect of the present invention, there is provided a method of sampling an outline, comprising: determining a reference line and a sampling interval for a shape mask of an object in a still image or a moving image; Sampling an outline of the object in accordance with a priority in consideration of a traveling direction with a position of a sample having a largest distance value on the reference line as a first sample point; Determining whether the corresponding sample point corresponds to a turning point in the shape mask of the object in the sampling step; If the corresponding sample point is a turning point, repeating the coordinate value corresponding to the turning point twice on the reference line and extracting the position data of the turning point; And extracting distance data from the reference line to the corresponding sample point when the corresponding sample point is not a change point in the change point determination step.

According to another aspect of the present invention, there is provided a method of encoding and decoding a reference line-based outline, comprising the steps of: generating an outline data and error data by sampling an outline of an object in a still image or a moving image using a reference line; Encoding the outermost data and error data using a motion compensation-transform structure, respectively;

Reconstructing an outermost sample and an error sample from the encoded outermost data and error data; Subtracting an error sample from the reconstructed outermost sample to reproduce reconstruction data to which error due to quantization is added to original sample data; And decoding the reproduced restoration data using a motion compensation-transform structure.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an apparatus for executing an outline encoding and decoding method according to the present invention; FIG.

FIG. 2 is a flowchart for explaining an operation according to the first embodiment of the outline sampling unit in FIG. 1; FIG.

Figure 3 is an example of an outline and a baseline of an object.

4 is an example of the outline data of FIG. 3 extracted by the method of FIG.

5 is an example of outline sampling data in FIG.

6 shows an example of data reproduced from the outermost data (data 0) in FIG.

FIG. 7 shows an example of data reproduced from the error data (data1, data2) in FIG.

FIG. 8 is a flowchart for explaining an operation according to the second embodiment of the outline sampling unit in FIG. 1; FIG.

FIG. 9 is a diagram showing a priority order for finding a next sample point in consideration of a traveling direction in FIG. 8; FIG.

10 shows an example of a turning point.

11 is an example of the outline data of Fig. 3 extracted by the method of Fig.

12 shows an example of data restoration using distance data.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a block diagram illustrating an outline encoder and a decoder according to the present invention. The outline encoder 10 includes an outline sampling unit 20, a motion estimator 21, a subtracter 22, a converter 23, a quantizer 24, a variable length coder (VLC) 25, an inverse quantizer 26, an inverse transformer 27, an adder 28, a reconstruction contour storage 29 and a motion compensator 30, The decoder 11 includes a variable length decoder (VLD) 40, an inverse quantizer 41, an inverse transformer 42, an adder 43, a restored contour storage 44, and a motion compensator 45.

The basic processing schemes of the outline coder 10 and the decoder 11 shown in Fig. 1 are motion compensation and transcoding.

In the outline coder 10, the outline sampler 20 samples the outline data based on the reference line from the mask image. The motion estimator 21 divides the data generated by the outline sampling unit 20 into blocks of a predetermined size and refers to the previous outline data stored in the restoration outline storage unit 29 for each of the blocks, do. The motion compensator (30) refers to the previous outline data and generates motion compensated data from the motion estimator (21). The motion compensation data thus generated is subtracted from the original data generated by the outline sampler 20 in the subtractor 22. [ The difference data obtained by the subtractor 22 is subjected to transformation and quantization through the transformer 23 and the quantizer 24. The data quantized by the quantizer 24 is encoded by the VLC 25 and is transmitted to the outline decoder 11. The data quantized by the quantizer 24 is sent to the VLC 25 and the data generated by the motion compensator 30 are sequentially sent to the adder 28 through the inverse quantizer 26 and the inverse transformer 27 Combine to create previous outline data. This previous outline data is stored in the restoration outline storage 29 and referred to in motion estimation and compensation of the incoming data. The VLC 25 also encodes the quantized data supplied from the quantizer 24 as well as the processing mode 32 and the motion information 31 generated in the encoding process and transmits the encoded data to the outline decoder 11.

The outline decoder 11 receives the encoded data and reproduces the outline image. In the VLD 40, data transmitted from the VLC 25 is analyzed and classified into quantization information, processing mode, and motion information. The quantization information is subjected to an inverse quantization and inverse transform process by the inverse quantizer 41 and the inverse transformer 42 and the motion mode and the motion information are calculated by referring to the previous outline data stored in the reconstruction contour storage 44, ) To perform motion compensation. The adder 43 combines the data output from the inverse transformer 42 and the data output from the motion compensator 45 to construct a final reproduction outline image. The data is stored again in the restoration outline storage 44, Is used as reference data for compensating the motion of the data coming from the memory.

FIG. 2 is a flowchart for explaining an operation according to the first embodiment of the outline sampling unit 20 in FIG. 1. In operation 60, a reference line for sampling an outline is selected. To be longer than the vertical line and the horizontal line that are generated. In FIG. 3, reference numeral 10 denotes an object, and reference numeral 11 denotes an example of the generated reference line.

In step 61, a sampling interval is determined. The sampling interval can be determined according to the size of the quantization step. That is, since the maximum error generated in the quantization process is 1/2 of the quantization step, the sampling interval can be set to 1/2 of the quantization step, so that the maximum error in the vertical and horizontal directions can be made equal. In operation 62, outline sampling is performed according to the determined reference line and the sampling interval.

Figure 4 shows the baseline 11 and the outline samples performed according to the same sampling interval. Sample points (0 to 28) are selected at the same interval on the reference line to extract the outer points of the object. A grid from the selected sample points to the object is formed, and the intersection of the outline of the object and the grid is set as a sampled outlier point. Here, in the arrangement of these outer points, the points closest to the reference line among the sampled outer points starting from one end of the reference line and arranging to the other are arranged, and then arranged again in the opposite direction, Use the array method. In FIG. 4, the shape data 0 indicates that the object 10 of FIG. 3 is extracted by the above method. After the outline data is extracted as described above, sampled outline points that have not yet been extracted are extracted by the same method. An example of this data is 'non shape data 1' and 'non shape data 2' in FIG.

However, the outline data extracted by this method may cause a sudden change in the data value. In FIG. 4, 20, 21, 22, and 23 are such cases, and this discontinuity is represented by separating 'non-shaped data 1' and 'non-shaped data 2'. Discontinuity in such encoding is the biggest factor that lowers the compression effect of the data. That is, one outermost sample (nL, nH, where n is a sample number) and zero or more error samples (pLq, pHq, where p is a sample number and q is an error data number) do. FIG. 5 shows the shape (org) of the sampled object generated through the process of FIG.

Returning to FIG. 2, in operation 63, the outermost samples are generated as data. The 'shape data 0' of FIG. 4 shows data of the outermost samples. Effective encoding is possible in the case of data that changes slowly in encoding using transforms. Therefore, when converting data into one-dimensional data, it is advantageous to make a small difference in the process of arranging from H to L in order to effectively encode. It is advantageous to arrange from 28L to 0L as shown in 'shape data 0' of FIG. 4, and to arrange from 0H to 28H in the order of 0L to 28L and 28H to 0H. In such an effective arrangement, the sizes of 0H-0L and 28H-28L are compared and the array is started on the larger side. At this time, in order to not use the indication bit for the arrangement method, when starting from 0, start from 0H and arrange clockwise. When starting from 28, start from 28L and arrange in the counterclockwise direction. Alternatively, the outline decoder 11 may be arranged in the same manner as described above. The 'shape data 0' generated in operation 63 is divided into a plurality of blocks of a predetermined size and processed and coded in a processing structure as shown in FIG.

Next, in order to extract the error sample data, in operation 64, reconstructed images are created using the outermost sample data generated in operation 63. FIG. 6 is the outermost sample data rec0 restored using 'shape data 0' of FIG. In operation 65, a process of subtracting 'org' sample data generated in operation 62 from the outermost sample image generated in operation 64 is performed. 7 is the error sample data generated in operation 65. FIG. In step 66, error samples generated as shown in FIG. 7 are separately arranged as 'non-shape data 1' and 'non-shape data 2' in FIG. 4, and the arrangement method is the same as that of the outermost samples . Here, the arranged data are divided into blocks each having a predetermined size, and are processed and coded in a processing structure as shown in FIG.

On the other hand, in Fig. 1, the outline decoder 11 generates restored data in the reverse of the sampling method of Fig. First, if data like the sixth diagram is generated from the reproduced outermost data, and then data similar to the seventh diagram is generated by using the reconstructed error data, restoration data to which an error due to quantization is added to the original sample data can be reproduced . Finally, the outline is restored using the reproduced sample data.

In the case of the outermost data, the motion estimation uses the motion information of the positional displacement having the smallest difference value within a certain range at the current position with reference to the outermost data of the previous block in which the block of a predetermined size is reproduced. In the case of error data, the most similar error data among the previously reproduced error data is selected, and the positional displacement having the smallest difference value within a certain range in the selected data is regarded as motion information. Therefore, in the case of error data, the previous error data number referenced is also used as the motion information. If there is no sample data to be partially referenced due to the difference in the number of data in the error data, both ends are used in both directions. In the case of motion compensation, the reference image is determined in the same way as the motion estimation. In the case of error data, both ends are used as reference error data in both directions.

FIG. 8 is a flowchart for explaining an operation according to the second embodiment of the outline sampling unit 20 in FIG. 1. In operation 100, a reference line is selected for the input shape mask. And the long axis of the vertical axis is selected as the reference line. In step 101, the accuracy of the outline is adjusted, and the sampling interval is determined. At this time, if the sampling interval is wide, the number of generated data is small, but the quality of the reproduced outline is not good, and if the interval of sampling is small, the number of data generated increases, but the quality of the reproduced outline is improved. In operation 102, a portion having the largest distance value among the sampling data at the reference line 0 is selected as a portion for searching for the initial data to be extracted. In operation 103, the next sample point is found in the current sample point in consideration of the traveling direction. In this case, the samples connected to eight positions in each of the four directions are considered, and the next sample point is selected by confirming whether there is a sample point that is not processed at the corresponding position according to the priority order considering the traveling direction as shown in FIG. In operation 104, it is determined whether all sample points have been processed. In operation 105, it is determined whether the current sample point is a turning point. At the change point, the coordinate value on the reference line is repeated twice. On the other hand, in the case of returning from one sample point, this judgment point can be generalized by repeating the sample point once. This is shown in Fig. In operation 106, if it is determined that the change point is a change point in operation 105, the change point location value is extracted and added to the change point list. In operation 107, if it is determined in operation 105 that the change point is not a change point, the distance from the reference line to the current sample point is extracted and added to the distance list.

The position and distance data of the switching points are finally extracted and sent to the outline coder as shown in FIG. 1 through the above process. 11 is an example of the outline data of FIG. 3 extracted by the r process of FIG. The outline sample data extracted by the procedure of FIG. 8 can be efficiently encoded through a conversion such as DPCM (Differential Pulse Code Modulation) or DCT. That is, since the continuity of data is excellent, the data concentration effect is excellent through DPCM or DCT.

12 shows that restoration is possible by using the distance data and the turning point data extracted by the process of FIG. the data is restored as shown in a) by the n-th data 21 and restored as shown in b) by the 20 data n-1. Since there is information that there is a turning point in 20 times, the direction of restoration is changed. Also, at the transition point, since it is necessary to repeatedly restore the next input sample data at the turning point, it is restored as in (c) in the (n + 2) th. (n + 3) -th sample is restored as the restoration direction has already been changed. In this way, all data can be used to restore all the data at the end. The method and / or the method for decoding a reference line based contour line according to the present invention can be used not only for contour coding of a moving image but also for contour coding of a still image.

As described above, according to the method of encoding and / or decoding a reference line based contour according to the present invention, the importance of each sample is equalized using a simple processing structure of the existing center-point based contour coding method and effective elimination of temporal redundancy. , It is possible to prevent the duplication of data in the center-based outline encoding method.

In addition, by using characteristics of outline data with gentle change, it has good reproducibility in small bit generation in outline coding of regular image through conversion, and excellent reproducibility in small bit generation in outline coding of moving image through conversion and motion estimation I have. In addition, since the apparatus for realizing the method of encoding and / or decoding a reference line based on the present invention has a conventional motion compensation-transform coding structure, it is possible to reduce the cost of hardware implementation.

Since the outline of the moving image has a tendency to change smoothly, motion estimation and compensation can be effectively performed even on the one-dimensional data sampled using the outline sampling method according to the present invention. Therefore, it is possible to perform coding more effectively than the conventional outline coding method. In addition, since the extracted outline samples maintain continuity, they have excellent performance in encoding through conversion such as DPCM or DCT. In addition, since it is not separated into various types of data, the structure is simplified even when motion compensation is performed. In particular, since the previous outline samples can be referred to, excellent performance is obtained in motion compensation.

Claims (13)

Determining a reference line and a sampling interval for an object in a still image or a moving image; Generating outline data by sampling an outline of the object using the reference line and a sampling interval, and restoring the shape of the object from the outline data; Extracting outermost samples of the object from the sampled outline data; Restoring an outline using the outermost samples; And And extracting error samples by obtaining a difference between the reconstructed outline from the shape of the reconstructed object. 2. The method according to claim 1, wherein the reference line is set to be longer than a vertical line and a horizontal line generated by projecting the object vertically and horizontally. 3. The method of claim 2, wherein the sampling interval is determined according to a size of a quantization step. 4. The method according to claim 3, wherein the sampling interval is set to 1/2 of a size of the quantization step. When the outermost samples and the error samples are arranged in order to convert the sampled outline data into the one-dimensional data, a difference value between the upper and lower outlines of each edge sample on the left and right sides of the object And the array is started from the side having a larger difference value. 6. The method of claim 5, wherein, when starting the array from the left edge sample, the samples are arranged in a clockwise direction starting from the sample of the left edge upper edge outline, and when arranging is started from the right edge sample, Lt; RTI ID = 0.0 > direction. ≪ / RTI > The method of claim 1, further comprising: sampling an outline of an object in a still image or a moving image using a reference line to generate outlier data and error data; And And encoding the outermost data and error data using a motion compensation-transform structure, respectively. Reconstructing an outermost sample and an error sample from the encoded outermost data and the error data, respectively; Subtracting an error sample from the reconstructed outermost sample to reproduce reconstruction data to which error due to quantization is added to original sample data; And And decoding the reconstructed reconstructed data using a motion compensation-transform structure. The method of claim 1, further comprising: sampling an outline of an object in a still image or a moving image using a reference line to generate outlier data and error data; Encoding the outermost data and error data using a motion compensation-transform structure, respectively; Reconstructing an outermost sample and an error sample from the encoded outermost data and error data; Subtracting an error sample from the reconstructed outermost sample to reproduce reconstruction data to which error due to quantization is added to original sample data; And And decoding the reconstructed reconstructed data using a motion compensation-transform structure. ≪ Desc / Clms Page number 19 > The method as claimed in any one of claims 7 to 9, wherein, in the motion estimation and compensation of the outermost samples, referring to the previous outermost sample data in which blocks of a predetermined size are reproduced, And the positional displacement having the smallest difference value is used as the motion information. 10. The method according to any one of claims 7 to 9, wherein, in the case of the motion estimation and compensation for the error samples, the most similar error data among the previously reproduced error data is selected, And the positional displacement having the smallest difference value is used as the motion information. Determining a baseline and a sampling interval for a shape mask of an object in a still image or a moving image; Sampling an outline of the object in accordance with a priority in consideration of a traveling direction with a position of a sample having a largest distance value on the reference line as a first sample point; Determining whether the corresponding sample point corresponds to a turning point in the shape mask of the object in the sampling step; If the corresponding sample point is a turning point, repeating the coordinate value corresponding to the turning point twice on the reference line and extracting the position data of the turning point; And And extracting distance data from the reference line to the corresponding sample point if the corresponding sample point is not a change point in the change point determination step. The method of claim 12, further comprising: sampling an outline of an object in a still image or a moving image using a predetermined reference line to generate position data and distance data of a turning point; And And encoding the position data and the distance data of the switching point using a motion compensation-conversion structure, respectively.