JP2006033433A - Method and apparatus for motion estimation, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a motion estimating technique for performing accurate motion type prediction and obtaining high robustness, even when video contents represent complicated motion. <P>SOLUTION: On the basis of a predetermined moving vector of each block of an object frame, a temporarily estimated value MV<SB>0</SB>" of a moving vector of a block C<SB>0</SB>is determined. A representative value δ<SB>T</SB>, of the distance between the MV<SB>0</SB>" and a moving vector MV<SB>j</SB>of a circumferential block of the block C<SB>0</SB>, i.e motion complexity, is calculated. Further, motion complexity δ<SB>T-1</SB>is calculated as to a reference frame. Block search methods, corresponding to those representative values δ<SB>T</SB>and δ<SB>T-1</SB>, are selected from among various block searching methods predetermined according to the representative values δ<SB>T</SB>and δ<SB>T-1</SB>. By the block searching methods, a block search is made to determine a moving vector. Thus, the block search methods are adapted according to motion complexity to obtain high robustness, even if video contents represent complicated motion. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a motion estimation technique using block matching in video compression, and in particular, motion estimation capable of adaptively performing block matching by detecting motion of a video that is difficult to capture depending on motion vectors with high sensitivity. Regarding technology.

  Block Matching Motion Estimation (BMME) is currently the most powerful motion estimation (ME) method for both software and hardware implementations. Adopted in various international standards such as H.216 / 263/263 +.

In block matching motion estimation, a sum of absolute difference (SAD) represented by (Equation 1) is usually used as a matching criterion. Here, M represents a block size. In addition, I _C and I _R were found in the pixel value in the target block (current block), which is a block matching target in the target frame (current flame), and in the reference frame, respectively. This represents the pixel value in the rematch candidate block.

However, it is clear that BMME inherently has a very computationally intensive task. And it has been reported that the task has a calculation amount that occupies 60% to 80% of the total calculation time in the video encoder (see Non-Patent Document 1). In particular, in the case of full search block matching (FSBM), which is a brute force method, the number of search points is (2N + 1) ² with respect to a search area of +/− N pixels.

  Therefore, conventionally, as an improvement technique of ME, a two-dimensional logarithmic search method (2-D Logarithmic Search: LOCS) (see Non-Patent Document 2), a three-step search method (Three Step Search: TSS) (Non-Patent Documents 3 and 11) Many high-speed search techniques such as a four-step search method (FSS) (see Non-Patent Document 4) and a diamond search method (Diamond Search: DS) (see Non-Patent Documents 5 and 6) have been proposed. ing. Unlike FSBM, these algorithms employ a technique for reducing the number of search points by using a predefined search pattern in order to theoretically achieve a significant reduction in search time. However, this makes hardware implementation difficult. Also, if such a rigid search strategy is employed, the search strategy is not always optimal for various types of video. Therefore, for some videos, the risk that the matching quality is reduced also increases.

After that, adaptive motion estimation algorithm (Adaptive motion estimation algorithm) is used to reduce the processing cost of the computer by suppressing the distortion of the video by dynamically adapting the search strategy of BM according to the motion complexity of the video content. algorithm) was devised. Most adaptive motion estimation algorithms prepare a plurality of search patterns in advance, and select an optimal search strategy based on the result of motion type prediction for the target block. In the motion vector field adaptive search technique (MVFAST), spatial correlation is used in motion type prediction (see Non-Patent Document 7). In addition, in the predictive motion vector field adaptive search technique (PMVFAST), temporally arranged blocks are also considered for motion type prediction (see Non-Patent Document 8). A large diamond search (LDS) or a small diamond search (SDS) is selected according to the predicted motion type. Further, Non-Patent Document 9 introduces a more sophisticated motion type prediction algorithm.
JP 2003-324743 A JP 2003-274416 A JP 2002-152760 A JP 2002-135784 A PM Kuhn, G. Diebel, S. Hermann, A. Keil, H. Mooshofer, A. Kaup, et al. "Complexity and PSNR-comparison of several fast motion estimation algorithms for MPEG4," in Proceeding of Applications of Digital Image Processing ; XXI, San Diego. SPIE, vol. 3460, pp. 486-499, July, 1998. JR Jain and AK Jain, "Displacement measurement and its application in interframe image coding," IEEE Trans. Commun., Vol. 29, pp. 1799-1808, Dec. 1981. T. Koga, K. Iinuma, A. Hirano, Y. Iijima, and T. Ishiguro, "Motion-compensated interframe coding for video conferencing," in Proc. Nat. Telecommunications Conf., Pp.G 5.3.1-G 5.3 .5., Nov./Dec. 1981. LM Po, and WC Ma "A novel Four-Step Search Algorithm for fast blockmatching," IEEE Trans. Circuits Syst. Video Technol., Vol. 6, Jun. 1996. G. Cote, M. Gallant, and F. Kossentini, "Efficient motion vector estimation and coding for H.263-based very low bit rate video compression," ITU-T SG 16, Q15-A-45, June 1997. JY Tham, S. Ranganath, M. Ranganath, and AA Kassim, "A novel unrestricted center-biased diamond search algorithm for block motion estimation," IEEE Trans. On Circuits & Systems for Video Technology, vol.8, pp.369- 377, Aug. 1998. PI Hosur and KK Ma, "Motion Vector Field Adaptive Fast Motion Estimation," Second International Conference on Information, Communications and Signal Proeessing (IClCS '99). Singapore, 7-lO, Dec. 1999. AM Tourapisl, OC Au2, and ML Liou "Predictive Motion Vector Field Adaptive Search Technique (PMVFAST)-Enhancing Block Based Motion Estimation," in Proc.SPIE, Visual Commuaications and Image Processing, vol.431O, pp. 883-892, Dec . 2000. JH Lim, and HW Choi, "Adaptive motion estimation algorithm using spatial and temporal correlation," Pacific Rim (PACRIM) Conference on Communications, Computers and signal Processing, vol.2, pp.26-28 Aug. 2001. T. Sappasitwong, and S. Aramvith, "Adaptive asymmetric diamond search algorithm for block-based motion estimation," International Conference on Digital Signal Processing, vol.2, pp.1-3, July .2002. Yoshinori Sakai and Toshiyuki Yoshida, “Human Communication Engineering Series Video Information Coding”, 1st Edition, Ohm Corporation, December 20, 2001, pp.121-141.

  The above-described adaptive motion estimation algorithm is expected to obtain better performance than the previous algorithm in terms of calculation speed and matching quality. However, the adaptive motion estimation algorithm described above employs a plurality of fast search methods that are considered to be not ideal when implemented in hardware. Furthermore, the problem of error propagation between frames is also serious.

  Accordingly, an object of the present invention is to provide a motion estimation technique capable of accurately predicting motion types and obtaining high robustness even when the video content shows complex motion. It is in.

  First, the basic concept of the motion estimation method according to the present invention will be described, and then the configuration and operation of the present invention will be described.

[1] Observation Although there are a wide variety of videos, all of them can consider a video sequence as one set of frames. Movement within one frame is generally complex. However, it is possible to cut a frame into parts that can be mapped to several typical motion types. This is the basis for motion type prediction. As a result of observing various images, the present inventor has obtained two conclusions summarized as follows.

(A) The definition of motion type is based on motion complexity.
In recent research (see Non-Patent Documents 9 and 10), motion types are defined based on motion complexity. “Motion complexity” refers to a correlation between motion vectors that are predictive variables.

  For example, in the frame illustrated in FIG. 1, uniform motion is detected in the upper part of the frame. Uniform motion, no matter how fast the motion is, can be converted into a fixed region by subtracting the global motion vector offset, so there is no difficulty.

(B) Motion vectors often do not accurately represent true motion.
Although motion vectors are useful for minimizing prediction errors in video coding, there is a “trap” in that motion vectors do not necessarily describe the true motion of the video. That is, the motion vector obtained by block matching is not actually a vector representing “motion between frames”, but rather should be interpreted as “a vector that can most eliminate redundancy between frames”. For example, a uniform texture has poor visual discrimination. Therefore, in FIG. 1, uniform motion is not captured as a motion vector in the lower part of the frame (that is, the part corresponding to the ground).

[2] Local Motion Type Prediction Based on the above observation, in motion estimation according to the present invention, motion type prediction is based on analysis of motion complexity around a target block (block that is a target of block matching). Done. By using the distribution of motion vectors as the basis of motion complexity analysis, it is possible to avoid falling into the “trap” when the motion vector fails to capture the true motion.

  Several blocks located in the vicinity of the target block are selected as blocks that form a context for performing block matching. FIG. 2 is an example of a context used to perform block matching.

The context T is a context composed of blocks in the target frame, and is a context composed of three blocks C ₁ , C ₂ , and C ₃ that are adjacent to the target block C ₀ . Here, the “first adjacency” refers to a block in which any side or vertex is in contact with any side or vertex of the target block.

The block matching of the target frame is performed along the repeated scanning in the horizontal direction from the upper left block to the lower right block of the frame along the scanning direction of the raster scan. Therefore, in FIG. 2, the blocks C ₁ , C ₂ , and C ₃ of the context T are blocks whose motion vectors have already been determined. For the reference frame, motion vectors have already been determined for all blocks.

The context T-1 is a context composed of blocks in the reference frame, and the block C ₀ ′ in the reference frame corresponding to the position of the target block C ₀ and the block C ₁ located in the first adjacency thereof. This is a context composed of “˜C ₈ ”.

In motion estimation according to the present invention, unlike a conventionally proposed motion estimation method (see Non-Patent Documents 9 and 10), in order to derive a spatial correlation in block matching, several blocks from the target frame and the reference frame are used. Is extracted and used for motion estimation of the target block C ₀ . The generalized adjacent block is used as a sample for extracting local motion information included in two consecutive frames. The analysis of the motion complexity is performed based on the following two evaluations for the context T-1 and the context T.

Here, MV _i ′ (i = 0, 1,..., 8) represents a motion vector of the block C _i ′ in the context T-1. MV _i (i = 1, 2, 3) represents the motion vector of the block C _i ′ in the context T. δ _T-1 and δ _T represent local motion complexity in the reference frame and the target frame, respectively. These are defined as functions of motion vectors used as predictor variables.

In the present invention, as functions defining local motion complexity, the motion vectors MV ₀ and MV ₀ ′ of the target block C ₀ and the block C ₀ ′ corresponding thereto and other functions in the contexts T and T−1 are used. A representative value of the distance between the motion vectors MV _i and MV _i ′ of the blocks C _i and C _i ′ is used. That is, local motion complexity δ _T−1 and δ _T are defined by (Equation 4) and (Equation 5). In particular, the context T-1 has a symmetric structure, and changes in motion characteristics from all directions can be captured, so (Equation 4) is a sensitive characteristic value (indicator).

Here, Rep _T-1 and Rep _T are representative value functions, respectively. As the representative value function, a maximum value, a mode value, a median value, an average value, a sum, or the like can be used. ‖ / ‖ Represents the norm of the vector. Also, the motion vector MV _0, since no been determined at this time, in the operation of the [delta] _T, the temporary estimate value _MV 0 instead of the motion vector _{MV 0 "= MV 0" (} MV 1, MV 2, MV 3 ). Various methods are conceivable for the arithmetic expression of MV ₀ ″.

More specifically, (Equation 6) and (Equation 7) can be used as the local motion complexity δ _T−1 and δ _T.

The motion types TYPE _T-1 and TYPE _T of the contexts T-1 and _T are classified according to the local motion complexity δ _T-1 and δ _T of the contexts T-1 and T as described above. The movement type classification can be determined by threshold determination as shown in (Equation 8) and (Equation 9).

Here, Th ₁ and Th ₂ represent threshold values for classifying motion types. By selecting the threshold values Th ₁ and Th ₂ to appropriate values, the degree of influence of temporal and spatial prediction variables can be adjusted.

In the next step, in consideration of temporal coherence, the most likely motion type is predicted for the target block C ₀ based on the TYPE _T-1 and TYPE _T. The motion type prediction of the target block C ₀ can be performed according to (Table 1).

By using the motion type of the current block C _0, it is possible to perform more reliable prediction of the motion vector MV ₀ of a target block C _0. For example, in the case where a uniform texture portion is moving at high speed, even if all other motion vectors other than MV ₆ ′ do not capture motion, as long as MV ₆ ′ captures motion, the above motion The type prediction makes it possible to detect that the symmetric block C ₀ is in intense motion. On the other hand, such other motion estimation methods cannot perform such motion detection.

[3] Adaptive block matching Using the motion type obtained by the above estimation, the present invention performs motion estimation by the following adaptive block matching.

(1) Step 1
After performing the above-mentioned local motion type prediction, four search center candidates are determined by (Equation 10). Each vector V _i = (v _xi , v _yi ) is a vector representing an offset from the center position of the block C ₀ ′ corresponding to the target block C ₀ to the point serving as the center of the search.

The vectors V ₃ and V ₄ represent representative values of motion vectors of blocks belonging to the upper right half and the lower left half of the block C ₀ ′ in the reference frame, respectively. Here, the values of (Equation 11) are used as the vectors V ₃ and V ₄ .

(2) Step 2
A vector V representing the best search center is selected from the four search center candidates. This selection is performed by calculating SAD by (Equation 12).

Here, I _T and I _T-1 represent pixel values of the target block and the reference block, respectively. The part of the operator argmin (•) represents “V _k that minimizes (•)”.

(3) Step 3
Finally, a block search is performed using the selected vector V as an offset. At this time, the motion type of the current block C ₀ obtained by the prediction, and selects a search method for determining the motion vector of the target block C _0. In this case, when the motion type of the target block C ₀ is “CHAOS”, it is predicted that the motion in the vicinity of the target block C ₀ is complicated. Accordingly, it is necessary to select a motion vector search method suitable for a wide range search. When the motion type of the target block C ₀ is “CRITICAL”, it is predicted that the motion in the vicinity of the target block C ₀ is near the boundary where the motion changes from a complex motion to a uniform motion. Accordingly, in this case, it is necessary to select a motion vector search method suitable for a medium range search. When the motion type of the target block C ₀ is “SIMPLE”, it is predicted that the motion in the vicinity of the target block C ₀ is uniform. Therefore, it is necessary to select a motion vector search method suitable for a narrow range search.

Therefore, for example, by a motion type of the current block C _0, it is possible to select a search method as shown in (Table 2).

Here, FS represents a full search method, and TSS represents a three step search method. N _simple , N _critical , and N _chaos represent search ranges. For example, in the case of N _simple represents that of searching in the area of N _simple × N _simple. Here, there is a relationship such as ( _Equation 13) among N _simple , N _critical , and N _chaos .

  In the case of 'SIMPLE', a considerable amount of useless calculation can be omitted by limiting the search area within a small window.

In the case of “CRITICAL”, the motion complexity of the target frame is lower than the motion complexity of the reference frame. Therefore, in most cases, the offset of the search center can be predicted with high reliability by (Equation 12). Therefore, the search range in block matching may be small. However, considering the “trap” described above, the search region should be larger than N _simple .

  In the case of ‘CHAOS’, since the correlation between the motion vectors is small, the search range for performing block matching needs to be large. However, the search center is shifted by the offset. Also, the size of the motion vector search area is limited to the extent that the size of the motion vector obtained by the search is within the practical range. Therefore, since the size of the search range is limited to an area of a certain size, it is considered that the calculation amount is reduced even when the motion complexity is large. The motion vector search can be performed according to a spiral pattern and an early break technique can be used.

In each case, when two or more candidates of the same SAD are found, the one closer to the position of the target block C ₀ may be adopted.

[4] Configuration and Operation of the Present Invention The first configuration of the motion estimation method according to the present invention is a motion estimation that estimates a motion vector of each block in a target frame with respect to a reference frame in a video composed of a plurality of frames. A method of searching for a block in the reference frame that has a maximum correlation with a block (hereinafter referred to as “target block”) C ₀ that is a target of motion vector estimation in the target frame. by performing in accordance with the block search method selected by the selection procedure with steps, and determines the motion vector MV ₀ of the current block C _0:
(1) First to determine a temporary estimated value MV ₀ ″ of the motion vector MV ₀ of the target block C ₀ based on the motion vector of each block of the reference frame or the target frame that has already been determined. Step;
(2) The temporary estimated value MV ₀ ″ and each block C _j (j that is in a predetermined range R ₁ around the target block C ₀ in the target frame and for which a motion vector has already been determined. A second step of calculating a representative value δ _T of the distance between the motion vector MV _j of ∈ R ₁ );
(3) The motion vector MV ₀ ′ of the block C ₀ ′ in the reference frame corresponding to the target block C ₀ and each of the reference frames within the predetermined range R ₂ around the block C ₀ ′ A third step of calculating a representative value δ _T-1 of the distance between the block C _i ′ (i∈R ₂ ) and the motion vector MV _i ′;
(4) from among the representative values [delta] _T and various block search method is determined in advance in correspondence with each value of the representative value [delta] _T-1, calculated in the second step and the third step the the fourth step of selecting the block search method corresponding to the representative values [delta] _T and the value of the representative values [delta] _T-1.

With this configuration, the degree of dispersion of the local motion vectors of the surrounding of the target block C ₀ for the temporary estimated value of the motion vector of the target block C ₀ block (hereinafter, "local motion complexity (local motion complexity)" The block search method can be determined adaptively. “Local motion complexity” refers to the degree of change of the surrounding motion vector with respect to the motion vector of the target block C ₀ . This local motion complexity is evaluated by the values of the representative values δ _T and δ _T−1 calculated in the second step and the third step.

That is, in the second and third steps, the local motion complexity centered on the target block C ₀ in the target frame and the local motion complexity centered on the block C ₀ ′ in the reference frame are calculated. Is done. Then, in the fourth step, reference is made to them, and if the local motion complexity is large, a block search method suitable for a wide block search is selected, and if the local motion complexity is small, the block search method is narrow. It becomes possible to select a block search method suitable for a block search of a range. As a result, it is possible to maintain high estimation accuracy of the motion vector when the motion is small, while suppressing the amount of computation required for the block search to a realistic amount of computation, and greatly deviate the prediction of the motion vector when the motion is large. This can be prevented.

In particular, by evaluating both the “local motion complexity” around the target block C ₀ and the “local motion complexity” around the block C ₀ ′ in the reference frame, a uniform texture is obtained. capture but as part moves at high speed, even for images having a certain movement around despite difficult conditions in which it is reflected in the motion to the motion vector of the target block C _0, the presence of motion sensitive It becomes possible. That is, accurate motion type prediction is possible, and high robustness can be obtained even when the video content shows complex motion. Since the block search method is adaptively selected according to the captured motion type, high matching quality can be obtained. Since the block search method is adaptively selected according to the captured motion, the residual prediction error can be further reduced and the video compression rate can be improved.

  Here, “block” refers to a unit of an area corresponding to one motion vector when encoding an image by motion compensation. Normally, a block of 16 × 16 pixels is used, but the size of the block is not particularly limited in the present invention.

The “block search method” refers to a method of searching for a block in a reference frame having the maximum correlation (the best match) with the target block C ₀ in the target frame. In the present invention, the “block search method” is not particularly limited. For example, a full search technique (2-D logarithmic search technique) (see Non-Patent Document 2). ), Multi-stage search methods (Three Step Search Technique, Four Step Search Technique, etc.) (see Non-Patent Documents 3, 4, and 11), motion vector field adaptive search method (Motion Vector) Field Adaptive Search Technique (see Non-Patent Document 7), Diamond Search Technique (see Non-Patent Documents 5 and 6), and the like can be used.

In the fourth step, usually, when the value of the representative value [delta] _T is small, the target block C ₀ neighborhood "local motion complexity" is considered to be small, matching the "block search method" error Is selected, and when the representative value δ _T is large, the local motion complexity in the vicinity of the target block C ₀ is considered to be large. The “search method” is preferably set to select a method capable of searching a wide area at high speed, such as a multistage search method, a motion vector field adaptive search method, or a diamond search method.

  The “motion vector” refers to a motion amount that represents the direction and distance in which the video has moved between the reference frame and the target frame.

The method of determining the temporary estimated value MV ₀ ″ of the motion vector MV ₀ is not particularly limited here. Examples of the method of determining the temporary estimated value MV ₀ ″ include, for example, in a target frame in the vicinity of the target block C ₀ . If the set of those motion vectors of the blocks has already been determined and the R _3, the average value of the motion vector of the block belonging to the set R _3, median, mode, temporary estimate the weighted average value value MV ₀ Or the like.

"Predetermined range R _1" is a vicinity of the current block C _0, is defined to a range of blocks having a motion vector is highly correlated with the motion vector of the target block C _0. In the present invention, the “predetermined range R ₁ ” is not particularly limited, but is usually set to a range including the first adjacent or second adjacent of the target block C ₀ .

"Predetermined range R _2" is _'a neighborhood of the block C _0' block C ₀ is defined to a range of blocks having a motion vector is highly correlated with motion vectors of. In the present invention, the “predetermined range R ₂ ” is not particularly limited, but is usually set to a range including the first adjacency or the second adjacency of the block C ₀ ′.

  The “distance” between two motion vectors refers to the spatial separation between the two motion vectors in the motion vector space. As the “distance”, the sum of absolute differences (SAD), mean square error (MSE), mean absolute error (MAE), Euclidean distance, etc. can be used. .

The “representative value” of distance refers to a statistic that summarizes the distribution of variables. As the “distance representative value δ _T ”, a maximum value, a minimum value, a median value, an average value, a mode value, a sum, a sum of squares, a sum of squares, or the like can be used.

The second configuration of the motion estimation method according to the present invention, the in the first configuration, in the fourth step, when the value of the representative values [delta] _T is greater than a predetermined threshold value Th ₁ is block search method as, select the multi-stage search method in a given search area, wherein when the value of the representative value [delta] _T is a predetermined threshold value Th ₁ or less, as the block search method, a narrow search region than the search area of the multi-stage search method The global search method is selected.

According to this configuration, the threshold determination of the value of the representative values [delta] _T, the search method of the motion vector it is possible to switch adaptively depending on local motion complexity in the vicinity of the target block C _0. Moreover, since it is threshold value determination, it is realizable with a simple algorithm, and the circuit configuration in the case of realization by hardware is also simple.

Here, the “predetermined threshold value Th ₁ ” is a value set experimentally. Further, the “predetermined threshold Th ₁ ” does not necessarily have to be a fixed value, and may be freely set from the outside according to a request for accuracy of motion estimation.

A third configuration of the motion estimation method according to the present invention, the in the first or second configuration, in the fourth step, when the value of the representative values [delta] _T is greater than a predetermined threshold value Th ₁ is block As a search method, when a multi-stage search method in a predetermined search region S ₀ is selected and the value of the representative value δ _T is equal to or less than a predetermined threshold Th ₁ , the value of the representative value δ _T-1 is a predetermined threshold. It is larger than Th ₂ as block search method, select the full search method in the multi-stage search method of the search area S ₀ and the same or narrower search area S _1, the representative value [delta] _T-1 value Is equal to or smaller than a predetermined threshold Th ₂ , the block search method is selected as a whole area search method in the search region S ₂ narrower than the search region R 1.

According to this configuration, the search method of the motion vector in accordance with the local motion complexity of the vicinity of the threshold determination of the representative value [delta] _T and representative values [delta] _T-1 value, the target block C ₀ and the block C _{0 '} Can be switched adaptively. Moreover, since it is threshold value determination, it is realizable with a simple algorithm, and the circuit configuration in the case of realization by hardware is also simple.

Here, the size of the search areas S ₀ , S ₁ , S ₂ in each case can be set to an appropriate size according to the required calculation speed, calculation accuracy, and the like.

The “predetermined threshold values Th ₁ and Th ₂ ” are values set experimentally. Further, the “predetermined threshold values Th ₁ and Th ₂ ” do not necessarily have to be fixed values, and may be freely set from the outside in response to a request for accuracy of motion estimation.

According to a fourth configuration of the motion estimation method of the present invention, in any one of the first to third configurations, in the first step, the temporary estimated value MV ₀ ″ is the target block in the target frame. It is determined based on a motion vector MV _k of each block C _k (kεR ₃ ) that is in a predetermined range R ₃ around C _{0 and} for which a motion vector has already been determined. .

With this configuration, it is possible to determine the temporary estimated value MV ₀ ″ of the motion vector of the target block C ₀ using the motion vector determined before the target block C ₀ according to scanning.

Here, the “predetermined range R ₃ ” is not particularly limited, but is normally set to a range including the first adjacency or the second adjacency of the target block C ₀ .

“Based on the motion vector MV _k ” means, for example, determining the average value, median value, mode value, etc. of the motion vector {MV _k | ∀k∈R ₃ } as the temporary estimated value MV ₀ ″. . also, the current block C ₀ in accordance with a raster scan when the motion will determine the vector MV ₀ represents a block of lines above the current block C _0, the current block C ₀ in a row of the target block C ₀ to the left For a block, a motion vector has already been determined, so if the left adjacent block of the target block C ₀ is C _a , the upper left adjacent block is C _b , and the upper adjacent block is C _c , for example, MV _The temporary estimated value MV ₀ ″ may be determined by a method similar to the two-dimensional pixel value prediction method as ₀ ″ = MV _a + MV _c −MV _b .

A fifth configuration of the motion estimation method according to the present invention is characterized in that, in the fourth configuration, the temporary estimated value MV ₀ ″ is calculated by (Equation 14).

According to this configuration, the provisional estimated value MV ₀ | can be determined "as, by using a central value of {MV _k ∀k∈R _3}, the provisional estimated value MV ₀ by a simple algorithm".

In a sixth configuration of the motion estimation method according to the present invention, in any one of the first to fifth configurations, in the second step, the predetermined range R ₁ is a first adjacent to the target block C ₀ . It is a block and is a part or all of the four blocks whose motion vectors have already been determined.

Thus, the detection of local motion complexity surrounding the current block C _0, by using the first neighbor block to the current block C _0, the influence of the motion vector of the block away from the current block C ₀ The local motion complexity around the target block C ₀ can be detected without receiving the signal. Therefore, it is possible to prevent a wide area search method from being selected in vain when selecting a block search method.

Seventh structure of the motion estimation method according to the present invention, in any one of configurations of the first to fifth, in the second step, the representative value [delta] _T is a feature that is calculated by (equation 15) To do.

Thus, by using the sum of absolute differences of the motion vector as a representative value [delta] _T (SAD), the calculation of the representative value [delta] _T becomes possible to realize by well-expressed, and simple algorithm local motion complexity .

According to an eighth configuration of the motion estimation method of the present invention, in any one of the first to seventh configurations, in the third step, the predetermined range R ₂ is adjacent to the block C ₀ ′. It is a range of 8 blocks.

In this way, by evaluating the local motion complexity with reference to the eight blocks adjacent to the block C ₀ ′, it is possible to capture the change in the motion characteristics from all directions. Therefore, the representative value δT _-1 is a sensitive characteristic value that captures a change in the characteristic of motion.

In a ninth configuration of the motion estimation method according to the present invention, in any one of the first to seventh configurations, in the third step, the representative value δT _-1 is calculated by (Equation 16). It is characterized by that.

  In this way, by using the maximum value, it is possible to increase the sensitivity to changes in motion characteristics and accurately capture changes in motion characteristics that occur in an arbitrary direction.

A first configuration of a motion estimation apparatus according to the present invention is a motion estimation apparatus that estimates a motion vector of each block in a target frame with respect to a reference frame in a video composed of a plurality of frames, the motion estimation apparatus including: A motion vector storage means for storing a motion vector for each block of the target frame; and a block in the target frame that is a target for motion vector estimation based on the motion vector stored in the motion vector storage means (hereinafter, referred to as "target block".) and the motion vector provisional estimation means for calculating a provisional estimate MV 0 _"of the motion vector MV ₀ of C _0, blocks in the current block C ₀ around in a predetermined range R ₁ And the motion vector of each block C _j (j∈R ₁ ) whose motion vector has already been determined. Torr MV _j and, a first representative value calculating means for calculating a representative value [delta] _T of the distance between the temporary estimated value MV 0 _", the block C ₀ in the reference frame corresponding to the current block C ₀ and 'motion vector MV _0' and between the motion vector MV _i '' of the blocks _{C i} of the reference frame in the predetermined range _{R 2} of the periphery of 'the block _{C 0 (i∈R 2)} Second representative value calculation means for calculating a representative value δ _T-1 of the distance of the distance, and various block search methods determined in advance corresponding to the representative value δ _T and the representative value δ _T-1 from among the search method selecting means for selecting the first representative value block search method corresponding to said calculated representative value [delta] _T and the value of the representative values [delta] _T-1 by calculating means, the search method selection According to the search method selected by the means. A motion vector determining unit that searches for a block in the reference frame having the maximum correlation with the lock C ₀ , determines the motion vector MV _0, and stores the motion vector MV ₀ in the motion vector storage unit; And

According to this configuration, the motion vector temporary estimation unit calculates the temporary estimation value MV ₀ ″ of the motion vector MV ₀ of the target block C ₀ based on the motion vector stored in the motion vector storage unit. The representative value calculating means calculates a representative value δ _T of the distance between the temporary estimated value MV ₀ ″ and each motion vector MV _j . The second representative value calculating means calculates a representative value δ _T-1 of the distance between the motion vector MV ₀ ′ and each motion vector MV _j ′. Then, the search method selection means selects the block search method corresponding to the value of the representative values [delta] _T and representative values [delta] _T-1. Then, the motion vector determination means searches for a block in the reference frame according to the selected search method, determines a motion vector MV ₀ and stores it in the motion vector storage means. By repeating the above operation, the motion vector of each block in the target frame is sequentially determined.

In this case, the search method of the motion vector is to accommodate both centered around the current block C ₀ "local motion complexity" and the current block C _{0 '} "local motion complexity" Selected. As a result, it is possible to maintain high estimation accuracy of the motion vector when the motion is small, while suppressing the amount of computation required for the block search to a realistic amount of computation, and greatly deviate the prediction of the motion vector when the motion is large. This can be prevented.

Further, even when the temporary estimation value of the motion vector of the target block C ₀ does not capture the true motion, the “local motion complexity” around both the target block C ₀ and the target block C ₀ ′ is evaluated. by, than in the case of focusing only on the current block C _0, easily capture the presence of motion in more sensitive. Since the block search method is adaptively selected according to the captured movement, the residual prediction error can be further reduced and the video compression rate can be improved.

The second configuration of a motion estimation apparatus according to the present invention, in the first configuration, the search method selection unit, when the value of the representative values [delta] _T is greater than a predetermined threshold value Th ₁ is block search as a method to select a multi-stage search method in a given search area, wherein when the value of the representative value [delta] _T is a predetermined threshold value Th ₁ or less, as the block search method, narrower search than the search area of the multi-stage search method It is characterized by selecting a whole area search method in a region.

A third configuration of a motion estimation apparatus according to the present invention, in the configuration of the first or 2, wherein the search method selection means, in the case the value of the representative values [delta] _T is greater than a predetermined threshold value Th ₁ is As the block search method, a multi-stage search method in a predetermined search region S ₀ is selected, and when the representative value δ _T is equal to or less than a predetermined threshold Th ₁ , the representative value δ _T-1 is a predetermined value. is larger than the threshold Th ₂ is a block search method, the select full search method in a multi-stage search method of the search area S ₀ and the same or narrower search area S _1, the representative value [delta] _T-1 of value in the case of predetermined threshold value Th ₂ or less, as the block search method, and selects the full search method in a narrow search region S ₂ than the search region R1.

According to a fourth configuration of the motion estimation apparatus of the present invention, in any one of the first to third configurations, the motion vector provisional estimation means is a predetermined range around the target block C ₀ in the target frame. The temporary estimated value MV ₀ ″ is determined based on the motion vector MV _k of each block C _k (k∈R ₃ ) that is a block in R ₃ and for which a motion vector has already been determined. .

The fifth configuration of the motion estimation apparatus according to the present invention is characterized in that, in the fourth configuration, the temporary estimated value MV ₀ ″ is calculated by (Equation 17).

Sixth configuration of a motion estimation apparatus according to the present invention, in any one of configurations of the first to fifth, wherein the predetermined range R ₁ is adjacent to the vertical and horizontal and oblique directions to the current block C _0, And it is a part or all of the four blocks whose motion vectors have already been determined.

Seventh configuration of a motion estimation apparatus according to the present invention, in any one of configurations of the first to fifth, wherein the first representative value calculating means, calculating a representative value [delta] _T by (equation 18) It is characterized by.

An eighth configuration of the motion estimation device according to the present invention is the configuration according to any one of the first to seventh configurations, wherein the predetermined range R ₂ is adjacent to the block C ₀ ′ vertically and horizontally and obliquely. It is a range of 8 blocks.

According to a ninth configuration of the motion estimation apparatus of the present invention, in any one of the first to seventh configurations, the second representative value calculation means calculates a representative value δ _T-1 by (Equation 19). It is characterized by doing.

  A program according to the present invention causes a computer to execute the motion vector estimation method having any one of the first to ninth configurations.

  As described above, according to the present invention, when the local motion complexity is large, a block search method suitable for a wide range block search is selected, and when the local motion complexity is small, a narrow range is selected. It is possible to select a block search method suitable for the block search. As a result, it is possible to maintain high estimation accuracy of the motion vector when the motion is small, while suppressing the amount of computation required for the block search to a realistic amount of computation, and greatly deviate the prediction of the motion vector when the motion is large. This can be prevented.

Further, even when the temporary estimation value of the motion vector of the target block C ₀ fails to capture the true motion, the “local motion complexity” around the target block C ₀ can be evaluated to evaluate the target than in the case of focusing only on the block C _0, easily capture the presence of motion in more sensitive. That is, accurate motion type prediction is possible, and high robustness can be obtained even when the video content shows complex motion. Since the block search method is adaptively selected according to the captured movement, the residual prediction error can be further reduced and the video compression rate can be improved.

  The best mode for carrying out the present invention will be described below with reference to the drawings.

  FIG. 3 is a diagram illustrating the overall configuration of a general video encoding device 1 (see Non-Patent Document 11). Since the video encoding device 1 of FIG. 3 is generally widely used, the description here is limited to the outline description.

  The video encoding device 1 includes a frame memory 2, a difference unit 3, a discrete cosine transformer (DCT) 4, a quantizer 5, an entropy encoder 6, a local decoder 7, a motion estimation device 8, and a motion compensator 9. It is configured.

The frame memory 2 is a memory that temporarily stores a video frame input from an image sensor or the like. The difference unit 3 outputs a difference image Δf _T obtained by subtracting the predicted frame f _T ′ output from the motion compensator 9 from the target frame f _T output from the frame memory 2. The DCT 4 performs a discrete cosine transform on the difference image Δf _T. The quantizer 5 quantizes the DCT coefficient output by the DCT 4. The entropy encoder 6 entropy-encodes the quantized DCT coefficient output from the quantizer 5 and the motion vector MV output from the motion estimation device 8 and outputs the result as a bit string of encoded video.

The local decoder 7 restores the DCT coefficient output from the quantizer 5 back to the original frame image. Then, the restored frame image is temporarily stored, and the reference frame f _T-1 ″ before the target frame f _T is output. The motion estimation device 8 uses the reference frame f _T-1 ″ and the target frame f. _The motion vector MV is estimated and output using _T. Motion compensator 9 uses the reference frame f _{T-1 "and} the motion vector MV, _'to generate the output. The prediction frame f _T' predicted frame f _T is a prediction value of the target frame f _T is the It is used when the difference image Δf _T is generated in the difference unit 3.

Here, when generating the motion vector MV in the motion estimation device 8 and in the motion compensator 9, the frame f _T−1 obtained from the input video as it is is not used as a reference frame, but is restored by the local decoder 7. The reason why the frame f _T-1 ″ is used is to prevent drift caused by accumulation of quantization errors in the decoded image on the video decoding side.

The local decoder 7 includes an inverse quantizer 10, an inverse discrete cosine transformer (inverse DCT) 11, an adder 12, and a frame memory 13. The DCT coefficient output from the quantizer 5 is input to the inverse quantizer 10 and subjected to inverse quantization, and then subjected to inverse discrete cosine return in the inverse DCT 11 to restore the difference image Δf _T ″. In the unit 12, the difference image Δf _T ″ and the predicted frame f _T ′ output from the motion compensator 9 are added, and the restored target frame f _T ″ is stored in the frame memory 13. These configurations are used. This is the same as a general video decoder.

  Next, the motion estimation device 8 according to the present invention will be described. FIG. 4 is a block diagram illustrating the configuration of the motion estimation apparatus 8 according to the first embodiment of the invention.

  The motion estimation device 8 includes a motion vector storage unit 20, a motion vector temporary estimation unit 21, representative value calculation units 22, 23, a search method selection unit 24, a motion vector determination unit 25, and a search center determination unit 26.

The motion vector storage means 20 is a memory for storing motion vectors MV ′ and MV for each block of the reference frame f _T−1 ”and the target frame f _T.

Motion vector temporary estimator 21, a motion based on the motion vector MV that is stored in the vector storage means 20, subject to the block (target block) motion C ₀ of a by motion vector estimation block in the target frame f _T A temporary estimated value MV ₀ ″ of the vector MV ₀ is calculated.

The representative value calculator 22 is a block within the first adjacent range R ₁ of the target block C ₀ and the motion vector MV _{j of} each block C _j (j∈R ₁ ) whose motion vector has already been determined. calculates a representative value [delta] _T of the distance between the temporary estimated value MV 0 _".

Representative value calculating means 23, the 'motion vector MV _0' of the block _{C 0} in the reference frame _{f T-1} ", corresponding to the target block _{C 0,} in a first range of adjacent _{R 2} blocks _{C 0} ' A representative value δ _T-1 of the distance between each block C _i ′ (iεR ₂ ) and the motion vector MV _i ′ in the reference frame f _T-1 ″ is calculated.

Search method selecting means 24, from various block search method is determined in advance in correspondence with the representative value [delta] _T and the value of the representative values [delta] _T-1, calculated by the representative value calculating means 22 Representative selecting a block search method corresponding to the value [delta] _T and the value of the representative values [delta] _T-1.

  The search center determining unit 26 determines an offset of the search center when the motion vector determining unit 25 performs a search.

The motion vector determination unit 25 has the reference frame f _T having the maximum correlation with the target block C ₀ around the search center determined by the search center determination unit 26 according to the search method selected by the search method selection unit 24. ₋₁ ″ is searched for a block, a motion vector MV ₀ is determined, and stored in the motion vector storage means 20.

The search method selection unit 24 includes motion type determination units 31 and 32 and a search method selection unit 33. Motion type determining means 31, the value of the representative values [delta] _T to the representative value calculating unit 22 outputs determined threshold, to determine the local motion type TYPE ₁ in the vicinity of the target block C _0. The motion type determining unit 32 determines the threshold value of the value of the representative value δ _T-1 output from the representative value calculating unit 22, and determines the local motion type TYPE ₂ near the target block C ₀ ′. Then, the search method selection unit 33 selects a search method according to the local motion types TYPE ₁ and TYPE ₂ .

  The motion estimation method of the motion estimation apparatus 8 according to the present embodiment configured as described above will be described below. The estimation of the motion vector is performed from the upper right block of the target frame toward the lower left block along the scanning line of the raster scan. In addition, for the block located at the end of the target frame, the motion vector is estimated using the whole area search method. Hereinafter, a motion estimation operation for a target block other than the block located at the end of the target frame will be described.

FIG. 5 is a flowchart showing the motion estimation method according to the first embodiment of the present invention. First, the motion vector temporary estimation means 21 reads the motion vectors MV ₁ , MV ₂ , and MV ₃ of each block of the context T shown in FIG. 2 from the motion vector storage means 20. Then, a temporary estimated value MV ₀ ″ of the motion vector of the target block C ₀ is calculated by (Equation 20) (S1).

Next, the representative value calculation means 22 reads the motion vectors MV ₁ , MV ₂ , MV ₃ of each block of the context T shown in FIG. 2 from the motion vector storage means 20. Then, the local motion complexity in the vicinity of the target block C ₀ in the target frame f _T is calculated based on the previously calculated temporary estimated value MV ₀ ″ of the motion vector and the motion vectors MV ₁ , MV ₂ , and MV _3. Here, as a local motion complexity, a representative value δ _T of the distance between the motion vector MV _{j of} each block C _j (jεR ₁ ) and the temporary estimated value MV ₀ ″. Is used. Accordingly, the representative value calculating unit 22 calculates a representative value [delta] _T is a local motion complexity of the current block C ₀ near by (Expression 21).

On the other hand, the representative value calculation means 23 reads the motion vectors MV ₀ ′ to MV ₈ ′ of each block of the context T-1 shown in FIG. Based on these motion vectors MV ₀ ′ to MV ₈ ′, the local motion complexity in the vicinity of the block C ₀ ′ in the reference frame f _T−1 ″ is calculated (S3). Here, the local motion complexity is calculated. as the motion complexity, and 'motion vector MV _i of (i∈R _2)' each block _{C i,} using a representative value _{[delta] T-1} of the distance between the 'motion vector MV _0' of the block _{C 0.} Thus The representative value calculation means 23 calculates a representative value δ _T-1 that is a local motion complexity in the vicinity of the target block C ₀ by (Equation 22).

Next, the search method selection means 24, calculates from various block search method is determined in advance in correspondence with the representative value [delta] _T and the value of the representative values [delta] _T-1, by the representative value calculating means 22, 23 representative value [delta] _T and block search method corresponding to the value of the representative value [delta] _T-1 selecting (S4). Specifically, first, the motion type decision unit 31, based on the representative value [delta] _T, determining the motion type TYPE _T according equation (9). Further, the motion type determination means 32 determines the motion type TYPE _T-1 according to (Equation 8) based on the representative value δT _-1 . The search method selection means 33 determines the motion type of the target block C ₀ according to (Table 1) based on the motion types TYPE _T and TYPE _T-1 . Then, based on the motion type of the target block C _0, the search method is selected according to (Table 2). In this way, the search strategy in block matching is adaptively determined according to the local motion complexity.

  Next, the search center determining means 26 determines the search center offset V when the motion vector determining means 25 performs the search according to (Equation 10) to (Equation 12) (S5).

Next, the motion vector determination unit 25 searches for a block in the reference frame f _T-1 ″ having the maximum correlation with the target block C ₀ according to the search method selected by the search method selection unit 24 (S6). At this time, the block search is performed around the position translated from the position of the block C ₀ ″ by the offset V determined in step S5. Thus, the motion vector MV ₀ of a target block _{C 0} is determined.

Finally, the motion vector determination unit 25 stores the motion vector MV ₀ in the motion vector storage unit 20 (S7), and ends the motion vector estimation operation of the target block C ₀ .

Such motion vector estimation operation, along a scan line of the raster scan by sequentially be performed while moving the current block C _0, it is possible to perform the motion vector estimation target frame f _T.

  Note that the motion estimation apparatus of the present embodiment can adaptively change the search strategy depending on the nature of the video, so that even when the video content shows complex motion, high robustness can be obtained, and the hardware It can be implemented as well as software. It can also be configured as a computer program and realized by a general-purpose computer.

It is a figure showing the motion vector map in the frame obtained from the test image | video (CIF). It is an example of the context used in order to perform block matching. It is a figure showing the whole structure of a general video coding apparatus. It is a block diagram showing the structure of the motion estimation apparatus 8 which concerns on Example 1 of this invention. It is a flowchart showing the motion estimation method which concerns on Example 1 of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Video coding apparatus 2 Frame memory 3 Differentiator 4 Discrete cosine transformer (DCT)
5 Quantizer 6 Entropy Encoder 7 Local Decoder 8 Motion Estimation Device 9 Motion Compensator 10 Inverse Quantizer 11 Inverse Discrete Cosine Transformer (Inverse DCT)
DESCRIPTION OF SYMBOLS 12 Adder 13 Frame memory 20 Motion vector storage means 21 Motion vector temporary estimation means 22, 23 Representative value calculation means 24 Search method selection means 25 Motion vector determination means 26 Search center determination means 31, 32 Motion type determination means 33 Search method selection means

Claims (19)

A motion estimation method for estimating a motion vector of each block in a target frame with respect to a reference frame in a moving image composed of a plurality of frames, the block being a target of motion vector estimation in the target frame (hereinafter, “ It is referred to as “target block”.) By searching for the block in the reference frame having the maximum correlation with C ₀ according to the block search method selected by the selection procedure having the following steps, the target block C _{0 is obtained.} A motion estimation method characterized by determining a motion vector MV ₀ of:
A first step of determining a temporary estimated value MV ₀ ″ of the motion vector MV ₀ of the target block C ₀ based on a motion vector of each block of the reference frame or the target frame that has already been determined;
The temporary estimated value MV ₀ ″ and each block C _j (j∈R ₁ ) that is within a predetermined range R ₁ around the target block C ₀ in the target frame and for which a motion vector has already been determined. A second step of calculating a representative value δ _T of the distance from the motion vector MV _j
Wherein the current block C _{0 'motion} vector MV _0' of the block C ₀ in the reference frame corresponding to each block C _i of the reference frame in the predetermined range R ₂ of the periphery of said block C _{0 '} A third step of calculating a representative value δ _T-1 of a distance between '(iεR ₂ ) and the motion vector MV _i ';
From various block search method is determined in advance in correspondence with the representative value [delta] _T and the value of the representative value [delta] _T-1, wherein the representative value calculated in the second step and the third step [delta] _A fourth step of selecting a block search method corresponding to _T and the value of the representative value δT ₋₁ . In the fourth step,
If the value of the representative values [delta] _T is greater than a predetermined threshold value Th ₁ is a block search method, select the multi-stage search method in a given search area,
Wherein the value of the representative value [delta] _T is in the case of predetermined threshold value Th ₁ or less, where the block search method, and selects the full search method in a narrow search region than the search area of the multi-stage search method Item 2. The motion estimation method according to Item 1. In the fourth step,
In the case the value of the representative values [delta] _T is greater than a predetermined threshold value Th ₁ is a block search method, select the multi-stage search method in a given search area S _0,
In the case the value of the representative values [delta] _T is a predetermined threshold value Th ₁ or less,
If the value of the representative values [delta] _T-1 is greater than a predetermined threshold Th ₂ is a block search method, the entire search method in the multi-stage search method of the search area S ₀ and the same or narrower search area S ₁ Select
Wherein when the value of the representative value [delta] _T-1 is in a predetermined threshold value Th ₂ or less, as the block search method, and selects the full search method in a narrow search region S ₂ than the search region R1 The motion estimation method according to claim 1 or 2. In the first step, the temporary estimated value MV ₀ ″ is a block within a predetermined range R ₃ around the target block C ₀ in the target frame, and each block C for which a motion vector has already been determined. The motion estimation method according to claim 1, wherein the motion estimation method is determined based on a motion vector MV _k of _k (k∈R ₃ ). The motion estimation method according to claim 4, wherein the temporary estimated value MV ₀ ″ is calculated by (Equation 1).

In the second step, the predetermined range R ₁ is a block adjacent to the target block C ₀ and a part or all of the four blocks whose motion vectors have already been determined. 6. The motion estimation method according to claim 1, wherein the motion estimation method is provided. In a second step, the representative value [delta] _T, the motion estimation method of any one of claims 1 to 5, characterized in that is calculated by (Equation 2).

The motion estimation method according to claim 1, wherein, in the third step, the predetermined range R ₂ is a range of eight blocks adjacent to the block C ₀ ′. . The motion estimation method according to claim 1, wherein, in the third step, the representative value δT ₋₁ is calculated by (Equation 3).

A motion estimation device that estimates a motion vector of each block in a target frame with respect to a reference frame in a moving image composed of a plurality of frames,
Motion vector storage means for storing a motion vector for each block of the reference frame and the target frame;
Based on the motion vector stored in the motion vector storage means, a block in the target frame that is a target of motion vector estimation (hereinafter referred to as “target block”) C ₀ motion vector MV ₀ Motion vector temporary estimation means for calculating a temporary estimated value MV ₀ ″;
A motion vector MV _{j of} each block C _j (jεR ₁ ) that is in a predetermined range R ₁ around the target block C ₀ and for which a motion vector has already been determined, and the temporary estimated value MV ₀ a first representative value calculating means for calculating a representative value [delta] _T of the distance between the "
Wherein the current block C _{0 'motion} vector MV _0' of the block C ₀ in the reference frame corresponding to each block C _i of the reference frame in the predetermined range R ₂ of the periphery of said block C _{0 '} Second representative value calculating means for calculating a representative value δ _T-1 of a distance between the motion vector MV _i 'of (iεR ₂ );
From various block search method is determined in advance in correspondence with the representative value [delta] _T and the value of the representative value [delta] _T-1, calculated by the first and the second representative value calculating means and the a search method selection means for selecting a block search method corresponding to the representative values [delta] _T and the value of the representative values [delta] _T-1,
According to the search method selected by the search method selection means, the block in the reference frame having the maximum correlation with the target block C ₀ is searched, the motion vector MV ₀ is determined, and the motion vector storage means Motion vector determination means to be stored in
A motion estimation apparatus comprising: The search method selection means includes
If the value of the representative values [delta] _T is greater than a predetermined threshold value Th ₁ is a block search method, select the multi-stage search method in a given search area,
Wherein the value of the representative value [delta] _T is in the case of predetermined threshold value Th ₁ or less, where the block search method, and selects the full search method in a narrow search region than the search area of the multi-stage search method Item 11. The motion estimation apparatus according to Item 10. The search method selection means includes
In the case the value of the representative values [delta] _T is greater than a predetermined threshold value Th ₁ is a block search method, select the multi-stage search method in a given search area S _0,
In the case the value of the representative values [delta] _T is a predetermined threshold value Th ₁ or less,
If the value of the representative values [delta] _T-1 is greater than a predetermined threshold Th ₂ is a block search method, the entire search method in the multi-stage search method of the search area S ₀ and the same or narrower search area S ₁ Select
Wherein when the value of the representative value [delta] _T-1 is in a predetermined threshold value Th ₂ or less, as the block search method, and selects the full search method in a narrow search region S ₂ than the search region R1 The motion estimation apparatus according to claim 10 or 11. The motion vector temporary estimation means is a block C _k (kεR ₃ ) that is in a predetermined range R ₃ around the target block C ₀ in the target frame and for which a motion vector has already been determined. The motion estimation apparatus according to claim 10, wherein the temporary estimated value MV ₀ ″ is determined based on a motion vector MV _k of the first motion vector MV _k . 14. The motion estimation apparatus according to claim 13, wherein the temporary estimated value MV ₀ ″ is calculated by (Equation 4).

The predetermined range R ₁ is a part or all of the four blocks that are adjacent to the target block C ₀ in the up / down / left / right and diagonal directions and whose motion vectors have already been determined. The motion estimation device according to claim 10. It said first representative value computation means (5) by the motion estimation apparatus as claimed in claim 10 to 14, wherein the calculating a representative value [delta] _T.

The motion estimation apparatus according to claim 10, wherein the predetermined range R ₂ is a range of eight blocks adjacent to the block C ₀ ′ in the up / down / left / right and diagonal directions. . The motion estimation method according to any one of claims 10 to 16, wherein the second representative value calculation means calculates a representative value δT _{-1 according} to (Equation 6).

A program for causing a computer to execute the method according to any one of claims 1 to 9.

Images