KR101796192B1 - Transcoding method and apparatus according to prediction motion vector selection - Google Patents

Abstract

The present invention discloses a transcoding method and apparatus for selecting a predictive motion vector that can solve the problem of a decrease in the accuracy of a motion vector and a decrease in a transcoding efficiency due to a decrease in the accuracy of a motion vector due to a decrease in the amount of calculation for motion vector prediction. A transcoding method is a transcoding method of converting a high resolution to a low resolution, comprising: collecting candidate prediction motion vectors of an upper resolution block matching a lower resolution block in consideration of spatial positions; Dividing; Selecting a candidate prediction motion vector that is closest to a center value of the divided space among candidate prediction motion vectors of the divided space as a representative candidate prediction motion vector; Selecting an optimal predicted motion vector from the representative predicted motion vectors; And performing a transform coding by predicting a motion vector of the lower resolution block based on the best predicted motion vector. Therefore, the coding performance of the motion vector of the current block can be greatly improved, thereby increasing the performance of the video transcoding device and greatly improving the quality of the reconstructed image.

Description

TECHNICAL FIELD [0001] The present invention relates to a transcoding method and apparatus for selecting a predictive motion vector,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a transcoding method and apparatus, and more particularly, to a method and apparatus for efficiently performing a transcoding by selecting a predictive motion vector efficiently at a high-resolution to low-resolution transcoding.

When transcoding a high-resolution video compression stream to a low-resolution video compression stream, a problem arises in that a plurality of motion vectors of a high-resolution block must be matched to one low-resolution macroblock in block-based motion prediction do.

1 is a diagram illustrating a down-sampling of a motion vector. As shown in FIG. 1, when an image having a size corresponding to four macroblocks of a higher resolution is reduced to one macroblock size, a situation in which four motion vectors belonging to a higher resolution are merged into one motion vector Lt; / RTI > Although motion vectors that are optimal for macroblocks of lower resolution can be found by performing actual motion based prediction, the enormous amount of computation required in this case is in violation of the fast processing speed, which is an essential requirement of transcoding.

In order to solve the above problem, a method of finding a representative motion vector from a motion vector candidate block of a block of an upper resolution image matching a lower resolution image is proposed in various conventional encoding methods .

The Random method is a method of randomly selecting one motion vector among a plurality of input motion vectors, and the operation speed of the transcoding is high, but it is inefficient in terms of encoding performance (Non-Patent Document 1).

The mean method is a method of selecting a mean value of some candidate motion vectors as a representative motion vector. As a method of finding a candidate motion vector, a method of using motion vectors matching the lower resolution in the original image as candidates, (Non-Patent Document 2) (Non-Patent Document 3), or a method of selecting a candidate group in consideration of correlation with neighboring macroblocks (Non-Patent Document 4). This technique has a disadvantage in that the magnitude of one of the candidate motion vectors greatly affects the result of the motion vector selected when the magnitude of the motion vector is larger than that of the other motion vectors.

The weighted average (WA) method is a method for selecting a representative motion vector through a weighted average of input motion vectors. The weight for each motion vector is determined by the spatial activity of the perspective prediction error (non-patent document 2) 5). This method is weak against the noise of the candidate motion vector group, and when the original motion vectors have various directions, the motion vector may be deflected.

The Weighted Median (WM) method obtains Euclidean distance between motion vectors and selects a representative value as a motion vector having an intermediate value (Non-Patent Document 1) (Non-Patent Document 4) (Non-Patent Document 6). This method shows a high coding efficiency, but requires a large amount of computation in calculating the intermediate value.

The DCmax method is a method of selecting a representative motion vector as a motion vector of a block having a maximum DC residual coefficient among the candidate motion vectors (Non-Patent Document 7). This method requires a slight increase in computation compared to the mean method, but shows improved coding performance compared to the mean method or the WA method.

 N. Bjork and C. Christopoulos, "Transcoder architecture for video coding," IEEE Trans. Consumer Electron., Vol. 44, pp. 88-98, Feb. 1998.  B. Shen, I. K. Sethi, and B. Vasudev, "Adaptive motion-vector resampling for compressed video down scaling," IEEE Trans. Circuits Syst. Video Technol., Vol. 9, no. 6, pp. 929-936, Sep. 1999.  P. Yin and M. Wu, "Video transcoding by reducing spatial resolution," in Proc. IEEE Int. Conf. Image Processing, vol. 1, 2000, pp. 972-975.  T. Shanableh and M. Ghanbari, "Heterogeneous video transcoding to lower spatial-temporal resolutions and different encoding formats," IEEE Trans. Multimedia, vol. 2, no. 2, pp. 101-110, Jun. 2000.  G. Shen, B. Zeng, Y.-Q. Zhang, and M. L. Liou, "Transcoder with arbitrarily resizing capability," in Proc. IEEE Int. Symp. Circuits and Systems (ISCAS) 2001, vol. 5, Sydney, NSW, Australia, May 2001, pp. 25-28.  J. Xin, M.-T. Sun, K. Chun, and B. S. Choi, "Motion re-estimation for HDTV to SDTV transcoding," in Proc. IEEE Int. Symp. Circuits and Systems (ISCAS) 2002, vol. 4, Geneva, Switzerland, May 2002, pp. 715-718.  M.-J. Chen, M.-C. Chu, and C.-W. Pan, "Efficient motion-estimation algorithm for reduced frame-rate video transcoder," IEEE Trans. Circuits Syst. Video Technol., Vol. 12, no. 4, pp. 269-275, Apr. 2002.

It is an object of the present invention to provide a transcoding method based on a predictive motion vector selection method capable of solving the problem of degradation of accuracy of a motion vector and reduction of transcoding efficiency due to degradation of accuracy of a motion vector due to a decrease in computation amount for motion vector prediction, Device.

Another object of the present invention is to divide candidate prediction motion vectors of a high resolution image block into several spaces in order to select an optimal motion vector of higher accuracy while maintaining a level of complexity similar to that of the conventional methods, And selecting a representative candidate motion vector and selecting a best candidate motion vector that minimizes a difference between the representative candidate motion vector and a motion vector of the current block.

According to an aspect of the present invention, there is provided a transform coding method for transforming candidate prediction motion vectors of an upper resolution block matched to a lower resolution block into high-resolution low-resolution blocks by taking spatial positions into account, Dividing the space based on the predicted motion vectors; Selecting one of the candidate prediction motion vectors in the divided space as a representative candidate prediction motion vector; Selecting an optimal predicted motion vector from the representative predicted motion vectors; And performing a transform coding by predicting a motion vector of the block having the lower resolution based on the best predicted motion vector.

The representative predictive motion vector selection step may include selecting a candidate predictive motion vector that is closest to the center value of the divided space as the representative predictive motion vector.

The optimal predictive motion vector selection step may include a step of selecting a new optimal predictive motion vector as a motion vector for which less distortion occurs in a neighboring area of a block indicated by the best predictive motion vector, . ≪ / RTI >

Wherein the optimal motion vector selection step selects the optimal predicted motion vector from the optimal predicted motion vector when the motion vector having less distortion occurs in the surrounding area of the block indicated by the best predicted motion vector, Pixel to a near area within a pixel.

The optimal motion vector selection step may include adaptively adjusting the peripheral region according to the characteristics of the image.

Wherein the optimal motion vector selection step includes adjusting a range of the surrounding area to a wide range when the optimal motion vector is large and adjusting the range of the surrounding area to a narrow range when the optimal motion vector is small .

The optimal predictive motion vector selection step may include a step of selecting a best predictive motion vector as a motion vector having the best performance in terms of rate-distortion among the representative predictive motion vectors.

The spatial division step may include dividing the space by a method of minimizing dispersion of a difference between a center value of the divided space and a distance of the candidate prediction motion vectors.

Wherein the spatial division step comprises: a first step of setting a center value by selecting any K among the candidate prediction motion vectors; A group of K groups which are matched with the new center value by finding a new center value that minimizes a distance between any one of the candidate predicted motion vectors and a center value of the K groups based on the K center values, A second step of assigning the one candidate predictive motion vector to the candidate motion vector; A third step of searching for the center value of each group as an average value of all candidate prediction motion vectors in each of the newly changed groups; And a fourth step of selecting a result of the clustering result which is no longer changed through the repetition of the second and third steps as an optimal space division and a center value.

The optimal predicted motion vector selection step may include selecting the best predicted motion vector using a Sum of Absolute Transformed Differences (SATD).

According to an aspect of the present invention, there is provided a transcoding apparatus for converging candidate motion vectors of an upper resolution block matching a lower resolution block in consideration of a spatial position, A spatial divider for dividing a space based on the predicted motion vectors; A representative predicted motion vector selecting unit for selecting one of the candidate predicted motion vectors in the divided space as a representative candidate predicted motion vector; An optimal predicted motion vector selecting unit for selecting an optimal predicted motion vector from among the representative predicted motion vectors; And a transform coding unit for performing a transform coding by predicting a motion vector of the block having the lower resolution based on the best predicted motion vector.

The representative predictive motion vector selection unit may select the candidate predictive motion vector having the closest distance from the center value of the divided space as the representative predictive motion vector.

In order to improve the accuracy of the optimal predictive motion vector, the optimal predictive motion vector selection unit may select, as a new optimal predictive motion vector, a motion vector that generates less distortion in the surrounding region of the block indicated by the best predictive motion vector .

Wherein the optimal motion vector selection unit selects the optimal predicted motion vector from the optimal predicted motion vector when the motion vector having less distortion occurs in the peripheral region of the block indicated by the best predicted motion vector, Of the area.

The optimal motion vector selection unit may adaptively adjust the peripheral region according to the characteristics of the image.

The optimal motion vector selection step adjusts the range of the surrounding area to a wide range when the optimal motion vector is large and adjusts the range of the surrounding area to a narrow range when the optimal motion vector is small.

The optimal predicted motion vector selection unit may select a best predicted motion vector as a best predicted motion vector in terms of rate-distortion among the representative predicted motion vectors.

The space division unit may divide the space by a method of minimizing dispersion of a difference between a center value of the divided space and a distance of the candidate prediction motion vectors.

Wherein the spatial divider comprises: a first center value setting unit for setting a center value by selecting any K among the candidate predicted motion vectors; A new center value that minimizes a distance between any one of the candidate predicted motion vectors and a center value of K groups is searched for and a group matching with the new center value is found A group generating unit that belongs to the one candidate prediction motion vector; A second center value generation unit for searching again the center value of each group as an average value of all candidate prediction motion vectors in each of the newly changed groups; And an optimal space division unit for selecting a result of the point of time at which the clustering result is no longer changed through the repetition of the second and third steps as the optimum space division and center value.

The optimal predicted motion vector selection unit may select the best predicted motion vector using a Sum of Absolute Transformed Differences (SATD).

According to the transform coding method and apparatus according to the predictive motion vector selection method of the present invention, a representative motion vector for a lower resolution block is selected through spatial division of a candidate motion vector obtained from an upper resolution block, Since the motion vector is selected to be smaller than the actual motion vector prediction and the selected motion vector is selected in close proximity to the actual motion vector prediction, the coding performance of the motion vector of the current block is greatly improved, There is an effect that the speed is increased and the image quality of the restored image is greatly improved.

1 is a diagram showing a down-sampling state of a motion vector
FIG. 2 is a flowchart schematically illustrating a transcoding method according to an embodiment of the present invention. FIG.
3 is a diagram illustrating an example of spatial division of a transcoding method according to an embodiment of the present invention;
4 is a detailed flowchart specifically illustrating a spatial division step of the transcoding method according to an embodiment of the present invention;
FIG. 5 is a detailed flowchart specifically illustrating an optimal predictive motion vector selection step of the transcoding method according to an embodiment of the present invention;
6 is a block diagram schematically illustrating a transcoding apparatus according to an embodiment of the present invention.
FIG. 7 is a detailed block diagram specifically illustrating a spatial division unit of a transcoding apparatus according to an embodiment of the present invention;
8 is a detailed block diagram specifically illustrating an optimal predictive motion vector selecting unit of the transcoding apparatus according to an embodiment of the present invention.
FIG. 9 is a graph comparing the performance of the spatial division method of the transcoding method of the present invention with that of another technique,
FIG. 10 is a graph comparing performance of an optimal motion vector selection technique of the transcoding method of the present invention with that of another technique.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail.

It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the relevant art and are to be interpreted in an ideal or overly formal sense unless explicitly defined in the present application Do not.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In order to facilitate the understanding of the present invention, the same reference numerals are used for the same constituent elements in the drawings and redundant explanations for the same constituent elements are omitted.

Transcoding method

2 is a flowchart schematically illustrating a transcoding method according to an embodiment of the present invention. 2, the transform coding method according to an exemplary embodiment of the present invention includes a motion vector aggregation step 210, a spatial division step 220, a representative predictive motion vector selection step 230, Step 240 and perform transcoding step 250. [0040]

Referring to FIG. 2, in a step 210 of collecting motion vectors, a transcoding device (not shown) appropriately selects candidate motion vectors of an upper resolution block matching a lower resolution block by considering spatial positions of respective motion vectors It is gathering. The K-means algorithm can be used for aggregation.

FIG. 3 is a diagram showing an example of spatial division of the transcoding method according to an embodiment of the present invention. As described above, a space is divided by using a K-means algorithm. As shown in FIG. 3, a solid line indicates a candidate predictive motion vector, and a dotted circle denotes a group of motion vectors that are grouped spatially. The darkly colored circle is the center value of the group, and what is lightly colored represents a representative predicted motion vector of each group in the predicted motion vector.

Referring to FIG. 3, there are a plurality of candidate motion vectors, which are grouped in consideration of a spatial positional relationship. And can be grouped into K groups through the spatial position relation, which will be described in more detail with reference to FIG. Next, in a spatially aggregated group, the center value is calculated. This means the center value of the divided space. If there are K groups, there are also K center values. A candidate prediction motion vector located closest to the center value among the candidate prediction motion vectors included in each of the following groups is selected as the representative prediction motion vector of each group. Therefore, when there are K groups of representative predictive motion vectors, K number is selected. Also, the space is divided based on the positional relationship of the grouped groups. Fig. 3 shows a state in which four spaces A, B, C and D are divided.

4 is a detailed flowchart illustrating the spatial division step 220 of the transcoding method according to an exemplary embodiment of the present invention. The spatial division step 220 according to an exemplary embodiment of the present invention may use a K-means algorithm. In general, the K-means algorithm is an algorithm for grouping given data into K groups, and operates in such a manner as to minimize dispersion of the difference between the center value of each group and the distance of data values. That is, it can be expressed by the following equation.

Here, S _i is an i-th candidate motion vector group among K groups, x _j is a j-th candidate motion vector, u _i is a center value of S _i , S is a center value of K S _i , _{j means} that the sum of the variances of _j is minimized.

Referring to FIG. 4, the transform coding apparatus finds an upper resolution block matching the lower resolution block and extracts a candidate prediction motion vector of the upper resolution block (Step 410). An arbitrary K of the candidate predicted motion vectors x _j of the extracted block is selected to set an initial center value (420). Next, motion vectors close to K center values are collected (430). Next, K groups are generated based on the aggregated motion vectors (440). The center values of the K groups are reset (450). The mobilized step 430. Looking at the resetting center value of step 450 in more detail since, looking at the u _i which minimizes the Euclidean distance between one of the candidate prediction motion vectors x _j and the center value of the K groups u _i that this matching Let x _j belong to group S _i . This can be expressed as follows.

The center value u _i of each group is again searched as an average value of all candidate motion vectors in each group that is newly changed. This can be expressed by the following equation.

After the above process, the re-found center value is compared with the leading center value (460). As a result of the comparison, if the leading center value differs from the current center value, the process returns to the motion vector gathering step 430 based on the center value of K based on the current center value. As a result of comparison, if the center values are equal to each other, K candidate motion vectors are extracted and the space is divided. That is, through the iterative process, the result of the point at which the clustering result is no longer changed is selected as the optimum grouping and center value, and the next step is performed.

Although it is not shown in the drawing, if the spatial division can be performed more efficiently based on the spatial position through another spatial division algorithm other than the K-means algorithm, the space can be divided by another method. It is not necessarily limited to the above method.

Returning to FIG. 2, in the representative predictive motion vector selection step 230, the transcoding apparatus selects a representative candidate motion vector in each group of the candidate motion vector space divided through the spatial division step 220. In the selection method, the candidate predicted motion vector having the closest distance from the center value of each group is selected as the representative predicted motion vector. The mathematical expressions related to selection are as follows.

If there are K candidate motion vector spaces, K representative candidate motion vectors are also selected. Referring to FIG. 3, there are four candidate motion vector spaces, and thus four representative candidate motion vectors exist.

Next, in the optimal predictive motion vector selection step 240, the transcoding apparatus selects an optimal predictive motion vector among the representative predictive motion vectors selected in the representative predictive motion vector selection step 230. [

5 is a detailed flowchart illustrating an optimal predicted motion vector selection step 240 of the transcoding method according to an exemplary embodiment of the present invention. Referring to FIG. 5, the optimal predicted motion vector selection step 240 includes a step 510 of selecting a best predicted motion vector as a motion vector showing the best performance in terms of rate-distortion among the representative motion vectors, And selecting (520) a motion vector that causes less distortion in the surrounding area of the pointing block as a new optimal predicted motion vector.

Referring to FIG. 5, in an optimal predicted motion vector selection step 510, a best predicted motion vector is selected as an optimal predicted motion vector in terms of rate-distortion among representative candidate predicted motion vectors. In this step 510, the rate-distortion cost of each representative predicted motion vector is calculated to compare the rate-distortion cost. The rate-distortion cost is sum of absolute difference (SAD). Sum of Absolute Transformed Difference (SATD) or Sum of Absolute Integer Transformed Difference (SAITD). And selects an optimal predicted motion vector based on the calculated rate-distortion cost.

In an exemplary embodiment of the present invention, an optimal motion vector is selected from among the motion vectors selected in the representative predictive motion vector selection step 230 in terms of rate-distortion.

은 x를 부호화하는데 들어가는 비트량이다. 또한, R _r 은 현재 블록의 잔여 신호(residual signal) 값을 부호화하는데 들어가는 비트량,

은 모드를 부호화하는데 들어가는 비트량,

는 움직임 벡터를 부호화하는데 들어가는 비트량,

과

은 가중치이다. D는 움직임 벡터가 가리키는 블록과 실제 블록 간의 왜곡을 의미한다.here,

Is the amount of bits entering x to encode x. Also, R _r denotes a bit amount for encoding the residual signal value of the current block,

Is the amount of bits entering the mode encoding,

The amount of bits to be used for coding the motion vector,

and

Is the weight. D is the distortion between the block indicated by the motion vector and the actual block.

Next, in order to improve the accuracy of the optimal motion vector selected in the previous step 510, in the step of selecting a new optimal predicted motion vector, less distortion occurs in the peripheral region of the block indicated by the selected motion vector Find the motion vector. In this case, the region for searching for an improved motion vector is limited to a region within a few pixels from the optimal motion vector selected in the previous step 510. If the area for finding an improved motion vector is large, the time required for the corresponding calculation is greatly increased, thereby failing to meet the purpose of realizing fast change coding. Therefore, it is necessary to adjust the motion vector search range adaptively according to the characteristics of the image. It is necessary to set the search range for estimating the motion vector to a wide range because image quality deterioration may increase in a motion image having a large motion. Therefore, the search range for improving the accuracy is adjusted according to the magnitude of the motion vector selected by the motion vector selection method.

Referring back to FIG. 2, finally, in the transcoding step 250, the transcoding apparatus decides a motion of a block having a lower resolution to be transcoded on the basis of the best predicted motion vector selected in the optimal predicted motion vector selection step 240 The vector is predicted and the transform coding is performed.

Transcoding device

6 is a block diagram schematically illustrating a transcoding apparatus according to an embodiment of the present invention. 6, the transcoding apparatus according to an exemplary embodiment of the present invention includes a spatial divider 610, a representative predictive motion vector selecting unit 620, an optimal predictive motion vector selecting unit 630, 640 < / RTI >

The spatial division unit 610 appropriately aggregates candidate motion vectors of an upper resolution block matching the lower resolution block in consideration of the spatial positions of the respective motion vectors. The K-means algorithm can be used for aggregation. There are a number of candidate motion vectors, which are grouped considering spatial relationships. And can be grouped into K groups through the spatial position relation, which will be described in more detail with reference to FIG. Next, in a spatially aggregated group, the center value is calculated. This means the center value of the divided space. If there are K groups, there are also K center values. And divides the space based on the positional relationship of the group.

FIG. 7 is a detailed block diagram specifically illustrating a space division unit 610 of the transcoding apparatus according to an embodiment of the present invention. 7, the space division unit 610 includes a first center value setting unit 710, a group generating unit 720, a second center value generating unit 730, and an optimal space dividing unit 740 . As described above, the K-means algorithm is an algorithm for grouping given data into K groups. The K-means algorithm divides the center value of each group and the difference of distance of data values Lt; / RTI >

The first center value setting unit 710 finds an upper resolution block matching the lower resolution block, extracts a candidate prediction motion vector of the upper resolution block, and calculates an arbitrary number K of candidate prediction motion vectors x _j To set the initial center value.

Next, the group generating unit 720 combines motion vectors close to the K center values (430). Next, K groups are generated based on the aggregated motion vectors.

The second center value generator 730 sets the center values of the K groups again. In this case, the x _j belonging to the group S _i to be looking at the u _i which minimizes the Euclidean distance between one of the candidate prediction motion vectors x _j and the center value of the K groups u _i this matching. Then, the center value u _i of each group is again searched as an average value of all candidate motion vectors in each group that is newly changed.

After the above process, the optimal space division unit 740 compares the center value generated by the second center value generation unit 730 with the preceding center value. As a result of the comparison, if the leading center value differs from the current center value, the first center value setting unit 710 returns to the first center value setting unit 710 based on the center value of K based on the current center value. As a result of comparison, if the center values are equal to each other, K candidate motion vectors are extracted and the space is divided. That is, an optimal grouping and a center value are selected as a result of a point in time at which no clustering result is changed through an iterative process.

Although not shown in the figure, if the space division unit 610 can perform spatial division more efficiently based on the spatial position through a spatial division algorithm other than the K-means algorithm, the spatial division unit 610 may divide the space . It is not necessarily limited to the above method.

6, the representative predictive motion vector selection unit 620 selects a representative candidate motion vector in each group of the candidate motion vector spaces divided through the spatial division unit 610. [ In the selection method, the candidate predicted motion vector having the closest distance from the center value of each group is selected as the representative predicted motion vector. As described above, if there are K candidate motion vector spaces, K representative candidate motion vectors are also selected.

8 is a detailed block diagram specifically illustrating an optimal predictive motion vector selection unit 630 of the transcoding apparatus according to an embodiment of the present invention. 8, the optimal predictive motion vector selection unit 630 may include a rate-distortion cost calculation unit 810, a vector selection unit 820, and an accuracy calculation unit 830. [

Referring to FIG. 8, the optimal predictive motion vector selection unit 630 compares the rate-distortion cost to select the best predicted motion vector as the best predicted motion vector in terms of rate-distortion among the representative candidate predictive motion vectors. Therefore, the rate-distortion cost calculator 810 calculates the rate-distortion cost of each representative predicted motion vector to compare the rate-distortion cost. The rate-distortion cost is sum of absolute difference (SAD). Sum of Absolute Transformed Difference (SATD) or Sum of Absolute Integer Transformed Difference (SAITD). An optimal predicted motion vector can be selected based on the calculated rate-distortion cost.

The vector selecting unit 820 selects an optimal predicted motion vector based on the rate-distortion cost of each representative predicted motion vector calculated by the rate-distortion cost calculator 810. [

Next, in order to improve the accuracy of the optimal motion vector selected by the vector selecting unit 820, the accuracy improving unit 830 finds a motion vector in which a less distortion occurs in the peripheral region of the block indicated by the selected motion vector. At this time, the region for searching for an improved motion vector is limited to a region within a few pixels from the optimum motion vector selected by the vector selection unit 820. If the area for finding an improved motion vector is large, the time required for the corresponding calculation is greatly increased, thereby failing to meet the purpose of realizing fast change coding. Therefore, it is necessary to adjust the motion vector search range adaptively according to the characteristics of the image. It is necessary to set the search range for estimating the motion vector to a wide range because image quality deterioration may increase in a motion image having a large motion. Therefore, the search range for improving the accuracy is adjusted according to the magnitude of the motion vector selected by the motion vector selection method.

6, finally, the transform coding unit 640 predicts a motion vector of a block having a lower resolution to be transcoded on the basis of the optimal predicted motion vector selected by the optimal predictive motion vector selecting unit 630, And performs encoding.

FIG. 9 is a graph comparing performance of a spatial division method of the transcoding method according to the present invention with another technique. In order to verify the performance of the spatial partitioning technique of the present invention, the performance was compared using the median method and the conventional median method. Experiments were performed on JSVM (Joint Scalable Video Model) 9.19.7, and 1280 x 720 city, crew, harbor, and soccer images were used as experimental images. The reduction in spatial resolution was halved in both width and height. That is, four macroblocks are merged into one macroblock. As shown in FIG. 9, the anchor technique is a result of performing all the motion prediction using the FDFE (Full Decoding Full Encoding) technique. Therefore, although the complexity is high, the best performance is obtained because an accurate motion vector can be obtained. On the other hand, the spatial division technique and the existing algorithms (Mean and Median) of the present invention are the results of motion compensation with motion vectors calculated through motion vector merging without motion prediction. From the results, it can be seen that the performance of the space division scheme of the present invention is improved by 1.5 dB or more compared with the conventional algorithm. Particularly, it can be seen that the performance of the present invention is remarkably excellent in fast and complex images such as City and Soccer. It can be seen that the conventional method can not be properly solved in a complicated motion, but the spatial division technique of the present invention effectively copes with a complicated movement through clustering. Although there is some degradation in performance compared to the FDFE method, the present invention exhibits a complexity that is not comparable with the FDFE method in terms of complexity. However, in order to narrow the difference in image quality from the FDFE method, an accuracy improvement algorithm is applied when selecting the optimal motion vector based on the motion vector intensity.

FIG. 10 is a graph comparing performance of an optimal motion vector selection technique of the transcoding method of the present invention with that of another technique. In particular, the present invention is for evaluating the performance when the optimal motion vector selection and the accuracy improvement algorithm proposed in the present invention are applied. As shown in FIG. 10, when the accuracy improvement algorithm is applied to the test results, the performance is almost the same as that of the FDFE algorithm.

Also, in order to evaluate the complexity, the motion prediction time of the transcoder based on the transcoding method of the present invention is compared with that of the FDFE-based transcoder. As a result, the transcoding method of the present invention has a motion prediction time reduction of about 30 times . Therefore, the transcoding method of the present invention shows almost the same performance as the FDFE method, but shows a reduction of the motion vector estimation amount by about 30 times as compared with the FDFE algorithm.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the inventions as defined by the following claims It will be understood that various modifications and changes may be made thereto without departing from the spirit and scope of the invention.

Claims (20)

Transforming candidate prediction motion vectors of an upper resolution block matched to a lower resolution block into high resolution to low resolution in consideration of a spatial position and dividing a space based on the combined candidate prediction motion vectors;
Selecting one of the candidate prediction motion vectors in the divided space as a representative predicted motion vector;
Selecting an optimal predicted motion vector from the representative predicted motion vectors; And
And performing a transform coding by predicting a motion vector of the block of the lower resolution based on the best predicted motion vector,
Wherein the optimal predicted motion vector selection step comprises:
Setting a temporary optimal prediction motion vector among the representative prediction motion vectors;
Determining a distortion of a neighboring area of the block indicated by the temporary optimal prediction motion vector;
And selecting a motion vector for an area in which a relatively small amount of distortion occurs from a neighboring area of a block indicated by the temporary optimal prediction motion vector as an optimal predicted motion vector.
2. The method of claim 1, wherein the representative predictive motion vector selection step
And selecting the candidate prediction motion vector having the closest distance from the center value of the divided space as the representative prediction motion vector.
delete 2. The method of claim 1,
And setting the neighboring region to a neighboring region within a few pixels from an area pointed by the temporary optimal prediction motion vector.
[5] The method of claim 4,
And adaptively adjusting the surrounding region according to characteristics of the image.
[6] The method of claim 5,
Adjusting the range of the surrounding region to a wide range when the temporary optimal prediction motion vector is large and adjusting the range of the surrounding region to a narrow range when the temporally optimal prediction motion vector is small, Lt; / RTI >
The method according to claim 1,
Wherein the optimal predictive motion vector selection step includes a step of selecting a motion vector having the best performance in terms of rate-distortion among the representative predictive motion vectors as the temporally optimal predictive motion vector. Transcoding method.
2. The method according to claim 1,
And dividing the space by a method of minimizing dispersion of a difference between a center value of the divided space and a distance of the candidate prediction motion vectors.
2. The method according to claim 1,
A first step of selecting an arbitrary K of the candidate prediction motion vectors and setting a center value;
A group of K groups which are matched with the new center value by finding a new center value that minimizes a distance between any one of the candidate predicted motion vectors and a center value of the K groups based on the K center values, A second step of assigning the one candidate predictive motion vector to the candidate motion vector;
A third step of searching for the center value of each group as an average value of all candidate prediction motion vectors in each group that is newly changed; And
And a fourth step of selecting a result of a point of time at which the clustering result is no longer changed through the repetition of the second and third steps as an optimum space division and a center value.
2. The method of claim 1,
And selecting the temporary optimal prediction motion vector using a Sum of Absolute Transformed Differences (SATD).
A spatial partitioning unit for grouping candidate prediction motion vectors of an upper resolution block matched with a lower resolution block in consideration of a spatial position in a transcoding from a high resolution to a low resolution and dividing a space based on the combined candidate prediction motion vectors;
A representative predicted motion vector selecting unit for selecting one of the candidate predicted motion vectors in the divided space as a representative predicted motion vector;
An optimal predicted motion vector selecting unit for selecting an optimal predicted motion vector from among the representative predicted motion vectors; And
And a transform coding unit for performing a transform coding by predicting a motion vector of the block having the lower resolution based on the best predicted motion vector,
Wherein the optimal predicted motion vector selection unit comprises:
Setting a temporary optimal prediction motion vector among the representative prediction motion vectors;
Confirming a distortion of a surrounding area of a block indicated by the temporally optimal predictive motion vector;
And a motion vector estimating unit for selecting a motion vector for an area in which a relatively small amount of distortion occurs, from the surrounding area of the block indicated by the temporally optimal predictive motion vector, as an optimal predictive motion vector.
12. The apparatus of claim 11, wherein the representative predictive motion vector selection unit
And selects the candidate predicted motion vector having the closest distance from the center value of the divided space as the representative predicted motion vector.
delete 12. The apparatus of claim 11, wherein the optimal predictive motion vector selection unit
And sets the neighboring area to a neighboring area within a few pixels from an area indicated by the temporary optimal prediction motion vector.
15. The method of claim 14,
Wherein the optimal predictive motion vector selection unit adaptively adjusts the surrounding region according to characteristics of an image.
16. The apparatus of claim 15, wherein the optimal predicted motion vector selection unit
Wherein the control unit adjusts the range of the surrounding area to a wide range when the temporary optimal prediction motion vector is large and adjusts the range of the surrounding area to a narrow range when the temporary optimal prediction motion vector is small. .
12. The method of claim 11,
Wherein the optimal predictive motion vector selection unit selects a motion vector having the best performance in terms of rate-distortion among the representative predictive motion vectors as the temporally optimal predictive motion vector.
12. The apparatus as claimed in claim 11, wherein the space division unit
And dividing the space by a method of minimizing dispersion of a difference between a center value of the divided space and a distance of the candidate prediction motion vectors.
12. The apparatus as claimed in claim 11, wherein the space division unit
A first center value setting unit for setting a center value by selecting any K of the candidate prediction motion vectors;
A new center value that minimizes a distance between any one of the candidate predicted motion vectors and a center value of K groups is searched for and a group matching with the new center value is found A group generating unit that belongs to the one candidate prediction motion vector;
A second center value generation unit for searching again the center value of each group as an average value of all candidate prediction motion vectors in each group that is newly changed; And
And an optimal space division unit for selecting a result of a point in time at which the clustering result is no longer changed through the second and third steps as an optimum space division and a center value.
12. The apparatus of claim 11, wherein the optimal predictive motion vector selection unit
Wherein the temporary optimal prediction motion vector is selected using a Sum of Absolute Transformed Differences (SATD).

Images