JP2015179944A - Method and program for determining intra-prediction direction - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce an amount of arithmetic operation in intra-prediction.SOLUTION: This method for determining an intra-prediction direction has :a step for selecting an encoding object pixel group and a reference pixel group respectively from reference pixels based on a relative positional relationship between the encoding object pixel of an encoding object block and the reference pixels on a memory, and selecting two or more directions based on a relative positional relationship between the encoding object pixel group and the reference pixel group on the memory; a step for computing costs for two or more selected directions and selecting a direction to minimize the cost; and a step for determining which correlation is higher as to correlation between the direction to minimize the cost and each intra-prediction direction and correlation between the direction which is not selected and each intra-prediction direction, computing a cost as to intra-prediction directions in which correlation with the direction to minimize the cost is determined to be relatively high, and selecting the intra-prediction direction to minimize the cost among them.

Description

  The present invention relates to an intra prediction direction determination method and an intra prediction direction determination program.

  H. HEVC formulated as the next video encoding standard system of H.264 is H.264. It is a coding method that realizes coding efficiency exceeding H.264. However, since the processing is complicated and the amount of calculation is enormous, it is necessary to significantly reduce the coding calculation cost in order to actually use the product. Have a problem. In particular, in intra prediction, the choices of the prediction mode are increasing, and it is necessary to determine an appropriate prediction mode at a higher speed. Here, with reference to FIG. 18, a standard configuration of an apparatus for encoding video will be described.

  FIG. 18 is a block diagram showing a standard configuration of an apparatus for encoding video. An encoding apparatus receives an input image to be encoded, divides a picture of the input image into blocks, encodes each block, and outputs the bit stream as an encoded stream. For this encoding, the intra prediction unit 1 outputs a prediction signal subjected to intra prediction. The motion compensation unit 2 outputs a motion compensation prediction signal. The subtraction unit 3 obtains a difference between the input image and the prediction signal that is the output of the intra prediction unit 1 or the motion compensation 2 and outputs it as a prediction residual signal. The DCT / quantization unit 4 performs orthogonal transform such as discrete cosine transform on the prediction residual signal, quantizes the transform coefficient, and outputs the quantized transform coefficient. The entropy encoding processing unit 5 performs entropy encoding on the quantized transform coefficient and outputs it as an encoded stream.

  On the other hand, the quantized transform coefficient is also input to the inverse quantization / inverse DCT unit 6, where inverse quantization and inverse orthogonal transform are performed, and a prediction residual signal is output. The adder 7 adds the prediction residual signal and the prediction signal output from the intra prediction unit 1 or the motion compensation unit 2 to generate a decoded video signal of the encoding target block. The decoded video signal is output to the loop filter 8 for use as a reference image in the motion compensation unit 2. The loop filter 8 performs a filtering process for reducing coding distortion, and outputs the image after the filtering process to the motion compensation unit 2 as a decoded video signal. The motion vector prediction unit 9 receives the input image and the output of the loop filter, predicts a motion vector, and outputs the result to the motion compensation unit 2 and the entropy encoding unit 5. The encoding control unit 10 inputs an input image and controls processing operations of the intra prediction unit 1, the DCT / quantization unit 4, and the entropy encoding unit 5.

  As described above, in video coding, a prediction method is determined in units of blocks, and there are roughly two prediction methods. The first is a prediction method using only one-screen encoded information called intra prediction, and the second is a prediction method using multiple-screen encoded information such as inter prediction (motion compensation) and skip. Hereinafter, only intra prediction related to the present invention will be described.

  Intra prediction is prediction using encoded pixels adjacent to the block to be encoded, and there are 35 modes for the selection of prediction modes. FIG. 19 is an explanatory diagram illustrating an intra prediction mode. Of these, the 33 mode refers to surrounding encoded pixels and predicts that the pixels continue in a certain direction. For example, when the prediction mode number is 26, the prediction is in the vertical direction, and the encoded pixels adjacent immediately above the encoding target block are copied one by one in the vertical direction, and the difference is encoded. On the other hand, when prediction in an oblique direction is performed such that the prediction mode number is 25, there is a possibility that adjacent encoded pixels do not exist depending on the angle. In this case, H.C. In H.264 and HEVC, filter processing is performed on encoded pixels to generate pixels at inter-pixel positions and perform interpolation. This process is called an intra smoothing process.

  In the intra prediction direction determination, the cost of each prediction direction is usually calculated, and the one with the lowest cost is determined as the optimum intra prediction mode. Therefore, in order to reduce the amount of calculation in determining the intra prediction direction, it is necessary to determine the size of the cost of each prediction mode with a smaller amount of calculation.

Next, a conventional method for performing intra prediction at high speed will be described. First, SSE (Streaming SIMD Extensions) will be described. As an instruction that enables high-speed processing of the CPU, there is a multimedia extension instruction set called SSE installed in a microprocessor. To this, a function for executing a plurality of floating point number operations simultaneously and an instruction frequently used for multimedia processing are added. Speeding up using this command is effective in processing data arranged within a certain range on the memory. FIG. 20 is an explanatory diagram illustrating an example of a memory configuration. As shown in FIG. 20, a case where data A _i and B _i (i = 0, 1, 2) are sequentially stored in the memory A and the memory B will be described.

When obtaining the absolute value S _i of the difference between the data of A _i and B _i ,
S ₀ = abs (A ₀ −B ₀ )
S ₁ = abs (A ₁ −B ₁ )
_{S 2 = abs (A 2 -B} 2)
Can be obtained.
In the above, the calculation processing of S ₀ , S ₁ , S ₂ is performed in order. In this case, the number of processes is counted as 3.

上記では、Ｓ_０，Ｓ_１，Ｓ_２の演算処理を一括して行う。この場合、処理回数は１回としてカウントされる。 On the other hand, in SSE, it carries out as follows.

In the above, the calculation processing of S ₀ , S ₁ , S ₂ is performed at once. In this case, the number of processes is counted as one.

  In this way, by using SSE, when the same processing is performed on data arranged in a certain range on the memory, it is possible to process them collectively. However, speeding up by using SSE has a problem that data to be handled must exist within a certain range on the memory. This is due to the following two characteristics in SSE.

  The first is that the data that can be stored in the high-speed instruction register at a time is limited to a certain range in the memory. In each register for high-speed instructions, since the memory capacity that can be read at a time is determined, it is not possible to simultaneously read data stored in a range exceeding that.

  Second, it takes a relatively long processing time to store data in the high-speed instruction register. The speed-up instruction register has a limited capacity as described above, but it takes a long time to store the memory data because of its large length. This processing time is the same even if all or only one of the data within the range is actually used for the calculation. For this reason, especially when data is scattered in various places on the memory, the time for storing data in the register increases, and therefore there is a possibility that the effect of the high-speed processing in the subsequent SSE processing cannot be obtained. There is.

Due to the above two points, when using SSE, the data to be calculated needs to be within a certain range on the memory. It should be noted that even for a plurality of combinations of data (A _i , B _{i in the} above example), if each data is arranged within a certain range on the memory (A ₀ , A ₁ , A _{2 in the} above example) ), The same can be said. In the above, S ₀ , S ₁ and S ₂ can be calculated at one time by one calculation. In this case, even if only A _i is arranged in a certain range of the memory, if the arrangement of B _i is not arranged in the certain range of the memory, the data of B _i is read one by one as described in the above problem. Since the necessity arises, it is considered that the speeding up by SSE may not be effective.

  It is conceivable to increase the speed of intra prediction by using the SSE described above for calculating each cost of intra prediction. However, in general, since the arrangement of the pixels in the block to be encoded and the encoded pixels used for prediction on each memory is different, the speed cannot be sufficiently increased only by using SSE for cost calculation. Hereinafter, the high-speed calculation refers to a high-speed calculation method that can be used in program implementation, such as SIMD.

  Next, the intra fast search method will be described. In prediction mode determination of intra prediction, it is usually necessary to calculate a prediction image for each prediction mode, and the amount of calculation for determining the prediction mode increases as the number of prediction modes increases. As a technique for reducing the amount of computation of intra prediction, there is a technique for reducing the amount of computation by narrowing down prediction mode candidates in advance using encoding information of peripheral pixels or encoding target pixels as follows (for example, Patent Document 1).

JP 2009-89267 A

  By the way, in the cost calculation in the intra prediction direction, it is necessary to calculate errors in all 33 pixels in the vertical and horizontal directions for all the pixels in the encoding target block. However, it is possible to perform high-speed calculation by SSE in a form corresponding to each direction. not appropriate. The reason will be described. In general, the pixels in the encoding target block are arranged in the memory one by one in the horizontal direction on the entire normal screen. Since the high-speed instruction register reads data along the memory array, pixel data can be taken into the same register only along one horizontal row.

  However, in the intra prediction direction narrowing down, calculations in 33 vertical and horizontal diagonal directions occur, and in the error calculation process, there is a possibility that data groups for performing the same calculation process are captured in different registers. In this way, the situation in which the same arithmetic processing can be executed on the pixels in the same register is limited, and there are few processes that can execute the high-speed instruction for the amount of calculation necessary for loading into the high-speed instruction register. Therefore, it is considered that it is difficult to reduce the calculation amount as the entire encoding process.

  On the other hand, according to Patent Document 1, since there is no particular rule for the method at the stage of narrowing down the prediction direction and the prediction direction that becomes a candidate for narrowing down, the amount of calculation depends on the selection method and the narrowing down method of candidate prediction directions. There is a possibility that the increase or the encoding efficiency is deteriorated.

  Further, by combining SSE and the technique of Patent Document 1, a prediction direction in which high-speed calculation is difficult to execute is selected based on the order of arrangement of data on the memory, and is narrowed down from the search direction when narrowing down the prediction mode of Patent Document 1. A method is conceivable. This is because prediction modes can be narrowed down by high-speed computation, so it is thought that faster intra prediction can be performed. However, if the direction that is difficult to perform high-speed computation is simply excluded from the search direction, coding efficiency may deteriorate. There is a problem that there is.

  The present invention has been made in view of such circumstances, and an object thereof is to provide an intra prediction direction determination method and an intra prediction direction determination program capable of reducing the amount of calculation in intra prediction.

  The present invention is an intra prediction direction determination method for determining an intra prediction direction when performing intra prediction of a block to be encoded using reference pixels that are pixels of adjacent encoded blocks, and includes: An encoding target pixel group and a reference pixel from a reference pixel that is a pixel of the encoded block adjacent to the encoding target block based on a relative positional relationship between the encoding target pixel and the reference pixel on a memory. 1st intra prediction direction selection which selects a group and selects two or more directions from all the intra prediction directions based on the relative positional relationship on the memory of this encoding object pixel group and this reference pixel group A second intra prediction direction selection step of calculating a cost for the two or more selected directions and selecting a direction with the lowest cost; The correlation between the correlation between the direction in which the cost is minimum and each intra prediction direction and the correlation between the direction not selected in the second intra prediction direction selection step and each intra prediction direction is relative. The cost is calculated for the intra prediction direction determined to have a relatively high correlation with the direction in which the cost is minimized, and the intra prediction direction in which the cost is minimized is selected. And a third intra prediction direction selection step.

  The present invention is an intra prediction direction determination method for determining an intra prediction direction when performing intra prediction of a block to be encoded using reference pixels that are pixels of adjacent encoded blocks, and includes: Based on the relative positional relationship between the encoding target pixel and the reference pixel on the memory, the encoding target pixel group and the reference pixel are respectively referred to from the reference pixel that is a pixel of the encoding target block and the adjacent encoded block. A group is selected, and based on the relative positional relationship between the encoding target pixel group and the reference pixel group on the memory, all the intra prediction directions and two or more directions from a predetermined direction that is not the intra prediction direction are selected. A first intra prediction direction selection step to be selected, a cost for the two or more selected directions, and a direction in which the cost is minimized. Between the intra prediction direction selection step, the correlation between the direction in which the cost is minimized and each intra prediction direction, and the direction not selected in the second intra prediction direction selection step and each intra prediction direction. It is determined whether the correlation is relatively high, and the cost is calculated for the intra prediction direction in which the correlation with the direction in which the cost is minimized is determined to be relatively high. And a third intra prediction direction selection step for selecting a minimum intra prediction direction.

  The present invention is characterized in that two or more directions selected in the first intra prediction direction selection step are in a relationship in which the mutual correlation becomes low.

  In the present invention, the second intra prediction direction selection step calculates a cost for the two or more selected directions, selects a plurality of directions from the smallest cost, and selects the third intra prediction direction. The step is centered on the direction in which the cost is the smallest among the directions selected in the second intra prediction direction selection step, and the cost is the smallest in the directions selected in the second intra prediction direction selection step. Determining a predetermined range determined using a direction excluding the predetermined direction, calculating a cost for an intra prediction direction included in the predetermined range, and selecting an intra prediction direction in which the cost is minimum And

  In the present invention, in the first intra prediction direction selection step, the block size of the encoding target block in addition to the relative positional relationship between the encoding target pixel of the encoding target block and the reference pixel on the memory. Based on the above, the encoding target pixel group and the reference pixel group are selected.

  In the present invention, the first intra prediction direction selection step is a pixel of the adjacent encoded block when there is no reference pixel which is a pixel of the adjacent encoded block at a position to be referred to. A decimal pixel is generated from a reference pixel, the generated decimal pixel is included in the reference pixel group, and the reference pixel group is selected.

  In the present invention, the relative positional relationship between the encoding target pixel of the encoding target block and the reference pixel on the memory is such that the encoding target pixel is stored in the memory one by one in the horizontal direction or the vertical direction. The reference pixel which is a pixel of the adjacent encoded block is a positional relationship stored in the memory in one column in the horizontal direction on the one side or in the vertical direction on the left side, and The pixel group is the first row from the top or the first column from the left of the encoding target block, and the reference pixel group is a reference pixel one above or one left from the encoding target pixel group. It is characterized by.

  The present invention is an intra prediction direction determination program for causing a computer to execute the intra prediction direction determination method.

  According to the present invention, since it is possible to reduce the amount of calculation in intra prediction, it is possible to reduce the amount of calculation in the encoding process.

It is a flowchart which shows the basic processing operation | movement of the intra estimation part 1 by embodiment of this invention. It is a flowchart which shows the detailed operation | movement of the process of step S1 shown in FIG. It is a flowchart which shows the detailed operation | movement of the process of step S3 shown in FIG. It is a flowchart which shows the detailed operation | movement of the process of step S1 shown in FIG. It is explanatory drawing which shows the mode of intra prediction. It is a flowchart which shows the detailed operation | movement of the process of step S1 shown in FIG. It is a flowchart which shows the detailed operation | movement of the process of step S3 shown in FIG. It is a flowchart which shows the detailed operation | movement of the process of step S1 shown in FIG. It is a flowchart which shows the detailed operation | movement of the process of step S2 shown in FIG. It is explanatory drawing which shows the processing operation shown in FIG. It is a flowchart which shows the detailed operation | movement of the process of step S1 shown in FIG. It is a flowchart which shows the detailed operation | movement of the process of step S2 shown in FIG. It is explanatory drawing which shows an example of a structure of memory. It is explanatory drawing which shows an example of a structure of memory. It is explanatory drawing which shows an example of a structure of memory. It is explanatory drawing which shows the range of a prediction mode number. It is explanatory drawing which shows the range of a prediction mode number. It is a block diagram which shows the standard structure of the apparatus which encodes an image | video. It is explanatory drawing which shows the mode of intra prediction. It is explanatory drawing which shows an example of a structure of memory.

  Hereinafter, an intra prediction direction determination method according to an embodiment of the present invention will be described with reference to the drawings. Since the intra prediction direction determination method is executed in the intra prediction unit 1 in the apparatus shown in FIG. 18, the description of the apparatus configuration is omitted here. FIG. 1 is a flowchart showing the basic processing operation of the intra prediction unit 1 according to the embodiment of the present invention. With reference to FIG. 1, the basic processing operation of the intra prediction unit 1 according to the embodiment of the present invention will be described. First, the intra prediction unit 1 performs an intra prediction direction narrowing process (step S1). Subsequently, the intra prediction unit 1 performs cost estimation by performing cost calculation (step S2). Next, the intra prediction unit 1 determines whether or not all candidates have been cost-calculated, and performs cost calculation for all candidates. And the intra estimation part 1 performs the determination process of an intra prediction direction (step S3). The intra prediction direction is determined by such processing operation.

  Hereinafter, unless otherwise defined, the cost is characterized by including, as a variable, an error between an encoding target image and an encoded image (reference image) or a generated code amount such as an RD cost and / or the above-described error. It means cost.

<First Embodiment>
Next, an intra prediction direction determination method according to the first embodiment of the present invention will be described. In the first embodiment, in the intra prediction direction determination, the costs of two or more selected intra prediction directions are calculated using the encoded pixels and the encoding target pixels on the coordinates having high correlation with them. Then, the intra prediction direction candidates determined based on the intra prediction direction with the lowest cost are narrowed down, and the intra prediction direction is determined from the narrowed down intra prediction direction candidates.

  Next, with reference to FIG. 2, the detailed operation of the processing in step S1 shown in FIG. 1 according to the first embodiment will be described. FIG. 2 is a flowchart showing a detailed operation of the process of step S1 shown in FIG. First, the intra prediction unit 1 narrows down the intra prediction direction from the encoding target pixel according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel (reference pixel) and the encoding target pixel on the memory. The encoding target pixel group to be used is selected (step S111). A reference pixel group used for predictive narrowing is selected from the reference pixels according to the relative positional relationship determined from the arrangement and coordinates of the reference pixels and the encoding target pixel on the memory (step S112). In consideration of the relative positional relationship between the encoding target pixel group selected from the intra prediction direction candidates and the reference pixel group, two or more prediction directions used for narrowing down the intra prediction direction are selected (step S113). In the case of HEVC, the 35 intra prediction directions are the intra prediction direction candidates.

  The processing in step S2 in the first embodiment is the same as the processing operation shown in FIG. That is, for each prediction direction selected by the processing operation illustrated in FIG. 2, cost estimation for each prediction direction is performed using the selected encoding target pixel group.

  Next, the detailed operation of step S3 shown in FIG. 1 according to the first embodiment will be described with reference to FIG. FIG. 3 is a flowchart showing a detailed operation of the process of step S3 shown in FIG. First, the intra prediction unit 1 determines the prediction direction A that minimizes the cost estimation result (step 131). Next, the intra prediction unit 1 narrows down all N intra prediction directions (N> 0) having high correlation with the prediction direction A as candidates (step S132). Then, the intra prediction unit 1 performs cost calculation on the N intra prediction directions, and determines a prediction direction that minimizes the cost (step S133).

  Here, high correlation means, for example, an area up to an intermediate position between two or more intra prediction directions selected by the processing shown in FIG. 2, and there is no narrowing candidate in an area rotated by 180 degrees. In that case, the area is also indicated.

  Thus, the correlation between the encoding target pixel and the reference pixel is high (that is, the relative positional relationship determined from the arrangement and coordinates of the encoding target pixel and the reference pixel on the memory is suitable for high-speed calculation by SIMD. Therefore, in the cost estimation at the time of narrowing down, there is a high possibility that the magnitude of the correlation in each prediction direction can be correctly compared. Moreover, since the result can be narrowed down to an appropriate prediction direction, the amount of calculation can be reduced while suppressing deterioration in encoding efficiency. Furthermore, high-speed calculation is facilitated by estimating the cost using pixels that are close in arrangement on the memory.

Second Embodiment
Next, an intra prediction direction determination method according to the second embodiment of the present invention will be described. In the second embodiment, in narrowing down the intra prediction direction, it is possible to use any direction that can be used as a prediction direction as a candidate, in addition to those that are candidates for the original intra prediction direction.

  Next, with reference to FIG. 4, the detailed operation of the processing in step S1 shown in FIG. 1 according to the second embodiment will be described. FIG. 4 is a flowchart showing the detailed operation of the process in step S1 shown in FIG. First, the intra prediction unit 1 uses the encoding target pixel to narrow down the intra prediction direction from the encoding target pixel according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel in the memory. A pixel group is selected (step S211). Next, the intra prediction unit 1 selects a reference pixel group used for prediction narrowing from the reference pixels according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel on the memory. (Step S212). Then, the intra prediction unit 1 selects two or more prediction directions used for narrowing down from the intra prediction direction or other directions (an arbitrary direction that can be used as the prediction direction) (step S213).

  Here, the other direction is a prediction direction between a direction in which the intra prediction mode is present and an adjacent direction as in A of FIG. 5 (in the case of A, a prediction direction between 21 and 22). If it is rotated 180 degrees like B, it means that it is within the range where the intra prediction mode exists (in the case of B, the prediction direction is obtained by rotating 5 by 180 degrees).

  The processing in step S2 in the second embodiment is the same as the processing operation shown in FIG. That is, for each selected prediction direction, cost estimation in each prediction direction is performed using the selected encoding target pixel group. Further, the processing in step S3 in the second embodiment is the same as the processing operation shown in FIG. That is, candidates are narrowed down from all intra prediction directions that have a high correlation with the prediction direction in which the cost estimation result is minimum, and the intra prediction direction is determined from the candidates.

  Thus, since the range of the prediction direction used for narrowing down is expanded compared with 1st Embodiment, 2nd Embodiment can anticipate narrowing down a more suitable prediction direction. On the other hand, since the calculation amount is not different from that in the first embodiment, it is possible to expect an improvement in encoding efficiency while maintaining a reduction in calculation amount.

<Third Embodiment>
Next, an intra prediction direction determination method according to the third embodiment of the present invention will be described. In the third embodiment, in the intra prediction direction narrowing process (step S1), it is possible to select two or more prediction directions such that the correlation between the prediction directions becomes low.

  Next, with reference to FIG. 6, the detailed operation of the processing in step S1 shown in FIG. 1 according to the third embodiment will be described. FIG. 6 is a flowchart showing the detailed operation of the process in step S1 shown in FIG. First, the intra prediction unit 1 uses the encoding target pixel to narrow down the intra prediction direction from the encoding target pixel according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel in the memory. A pixel group is selected (step S311). Next, the intra prediction unit 1 selects a reference pixel group used for prediction narrowing from the reference pixels according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel on the memory. (Step S312). Then, the intra prediction unit 1 selects two or more prediction directions used for narrowing down from the intra prediction direction or other directions so that the correlation between them becomes low (step S313).

  Here, when two or more prediction directions are selected so that the correlation between them becomes low, the number of narrowing down is N (N is a positive integer), and the angle between both ends of all directions that can be selected as the intra prediction direction is α ( When 0 <α), among the prediction directions actually selected, the angle β formed by those adjacent to each other becomes {(α ÷ N) −δ} <β <{(α ÷ N) + δ} ( However, δ is selected so as to be the maximum value (δ <180 degrees) between the angles between all adjacent intra prediction directions.

  The processing in step S2 in the third embodiment is the same as the processing operation shown in FIG. That is, for each selected prediction direction, cost estimation in each prediction direction is performed using the encoding target pixel group selected above. Further, the processing in step S3 in the third embodiment is the same as the processing operation shown in FIG. That is, candidates are narrowed down from all intra prediction directions that have a high correlation with the prediction direction in which the cost estimation result is minimum, and the intra prediction direction is determined from the candidates.

  Thus, when the directions used for narrowing down are close, it is considered that there is a case where the correlation is high and appropriate cost estimation cannot be performed. On the other hand, by selecting the prediction directions in the narrowing-down process so that the correlation between them becomes low by the process shown in FIG. 6, it is easier to determine the correlation at the time of cost estimation, and it is expected to narrow down the more appropriate prediction direction. it can. On the other hand, since the calculation amount is not different from that in the first and second embodiments, it is possible to expect improvement in coding efficiency while maintaining a reduction in calculation amount.

<Fourth embodiment>
Next, an intra prediction direction determination method according to the fourth embodiment of the present invention will be described. In the fourth embodiment, when the range of the prediction direction having a high correlation with the intra prediction direction is set in accordance with the selection of the intra prediction direction, and the correlation with the intra prediction direction that minimizes the cost is high with respect to the cost estimation result, The intra prediction direction is determined from a possible range.

  The processing in step S1 in the fourth embodiment is the same as the processing operation shown in FIG. In other words, the encoding target pixel group used for narrowing down the intra prediction direction is selected from the encoding target pixel according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel in the memory. Subsequently, a reference pixel group used for predictive narrowing is selected from the reference pixels according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel on the memory. Then, two or more prediction directions used for narrowing down the intra prediction direction are selected from the intra prediction direction candidates.

  The processing in step S2 in the fourth embodiment is the same as the processing operation shown in FIG. That is, for each selected prediction direction, cost estimation in each prediction direction is performed using the selected encoding target pixel group.

  Next, with reference to FIG. 7, the detailed operation of the processing in step S3 shown in FIG. 1 according to the fourth embodiment will be described. FIG. 7 is a flowchart showing a detailed operation of the process in step S3 shown in FIG. First, the intra prediction unit 1 rearranges the prediction directions in ascending order of cost according to the type of prediction direction selected in step S1 and the result of cost estimation in step S2 (step S431). Subsequently, the intra prediction unit 1 narrows down N (N> 0) intra prediction directions that are likely to be the lowest cost from the rearranged prediction directions (step S432). And the intra estimation part 1 calculates cost with respect to the narrowed down intra prediction direction, and determines the thing which becomes the minimum as an intra prediction direction (step S433).

The determination of the prediction direction here is as follows, for example. In order from the direction in which the cost of step S2 is minimized, each direction is X0, X1, X2, X3,..., And the angle formed by X0 and Xi is αi (i = 1, 2, 3,...). (Αi> 0). At this time, when all the directions Xi are within the region of the intra prediction direction,
{(Α1 × 2) ÷ 3 ± δ1} degrees in the direction of X1 {(α2) ÷ 3 ± δ2} degrees in the direction of X2 (when X1 and X2 are in the same direction as viewed from X0, in the opposite direction to X1 Configuration)
(Where δi is an angle satisfying 0 <δi <{(αi) ÷ 3})
The cost is calculated with respect to the intra prediction direction in the region satisfying either of the above, and the minimum is determined. Here, in the intra prediction direction area, for example, in HEVC intra prediction, refers to an area of 180 degrees in which each direction from directions 2 to 35 is defined, and the opposite area (in FIG. 5). (Direction B) is not within the intra prediction direction area.

  Also, if there is a third direction between the direction of the first cost and the second cost, and if either the first or second direction is out of the area, the direction is rotated 180 degrees and searched. Set the range. Furthermore, if there is a direction outside the region within the set search range, the portion rotated by 180 degrees only that portion is taken as the search range.

  In addition to this, the prediction direction may be determined by the following method. In order from the direction in which the cost of step S2 is minimized, each direction is X0, X1, X2, X3,..., And the angle formed by X0 and Xi is αi (i = 1, 2, 3,...). (Αi> 0) and the cost of Xi is Si. At this time, the average of the costs in the direction Xki (i = 1, 2,...) In the clockwise direction from X0 and in the direction Xlj (j = 1, 2,...) In the counterclockwise direction. The prediction direction is determined using the ratio of the reciprocal values. Here, clockwise refers to rotation within 180 degrees clockwise from X0, and counterclockwise refers to rotation within 180 degrees counterclockwise from X0. Here, ki means the rearranged order when Xi is rearranged in the order of k1, k2,... In the clockwise order from X0, and lj is the order of l1, l2,. , Xi is rearranged in the order counterclockwise from X0. For any i, j, ki ≠ lj.

Ｘ０から反時計回り方向にある方向のコスト和の平均Ｓｌ（Ｎｌは、反時計回り方向にあるＸｉの個数）

であるときの予測方向は、以下のようになる。
Ｘ０から時計回り方向に

Ｘ０から反時計回り方向に

但し、αｋ１，αｌ１は、Ｘ０から時計回り方向、反時計回り方向に回転させるとき、最もＸ０に近いＸｋｉ、Ｘｌｊの方向までの角度とする。 At this time, for example,
Average cost sum Sk in the clockwise direction from X0 (Nk is the number of Xi in the clockwise direction)

Average cost sum S1 in the counterclockwise direction from X0 (Nl is the number of Xi in the counterclockwise direction)

The prediction direction when is is as follows.
Clockwise from X0

Counterclockwise from X0

However, αk1 and αl1 are angles from X0 to the directions of Xki and Xlj that are closest to X0 when rotating clockwise and counterclockwise.

  In the first to third embodiments, the intra prediction direction having a high correlation with the narrowed prediction direction is set as a candidate for performing the final intra prediction. However, in the fourth embodiment, all the cost estimation results are used, The region where the cost is minimum is estimated according to the distribution of. The calculation for performing the region estimation that minimizes the cost is generated from the cost estimation result, but the increase amount is very small compared to the calculation amount in the intra prediction, which is almost the same as the first to third embodiments. Intra prediction can be performed with higher accuracy while maintaining the amount of calculation.

<Fifth Embodiment>
Next, an intra prediction direction determination method according to a fifth embodiment of the present invention will be described. In the fifth embodiment, in the intra prediction direction determination, encoded pixels and pixels within a size limited according to the block size among the encoding target pixels on the coordinates having high correlation with the encoded pixels are used. Then, the cost of two or more preselected prediction directions is estimated by high-speed calculation to narrow down intra prediction direction candidates, and the intra prediction direction is determined from the narrowed down directions.

  Next, with reference to FIG. 8, the detailed operation of the processing in step S1 shown in FIG. 1 according to the fifth embodiment will be described. FIG. 8 is a flowchart showing the detailed operation of the process in step S1 shown in FIG. First, the intra prediction unit 1 determines whether the encoding target pixel is based on the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel on the memory and the block size to which the encoding target pixel belongs. The encoding target pixel group used for narrowing down the intra prediction direction is selected (step S511). Subsequently, the intra prediction unit 1 predicts from the reference pixel according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel on the memory and the block size to which the encoding target pixel belongs. A reference pixel group used for narrowing down is selected (step S512). Then, the intra prediction unit 1 selects two or more prediction directions to be used for narrowing down the intra prediction direction from the candidates for the intra prediction direction (step S513).

  The processing in step S2 in the fifth embodiment is the same as the processing operation shown in FIG. That is, for each selected prediction direction, cost estimation in each prediction direction is performed using the selected encoding target pixel group. Also, the processing in step S3 in the fifth embodiment is the same as the processing operation shown in FIG. That is, candidates are narrowed down from all intra prediction directions that have a high correlation with the prediction direction in which the cost estimation result is minimum, and the intra prediction direction is determined from the candidates.

  Thus, by limiting the number of pixels used for narrowing down based on the block size to which the encoding target pixel belongs, the upper limit of the calculation amount can be set while suppressing the deterioration of the encoding efficiency.

<Sixth Embodiment>
Next, an intra prediction direction determination method according to the sixth embodiment of the present invention will be described. In the sixth embodiment, when calculating the cost of two or more preselected prediction directions in determining the intra prediction direction, if the reference position is between pixels and there is no pixel at that position, decimals are calculated from the surrounding pixels. Generate pixels and narrow down intra prediction direction candidates. For the fractional pixel generation here, in addition to the intra-smoothing process used in the encoding method, any method that does not require an enormous amount of computation, such as generation by upsampling or quoting surrounding pixels as they are, can be used. Various methods may be used.

  The processing in step S1 in the sixth embodiment is the same as the processing operation shown in FIG. In other words, the encoding target pixel group used for narrowing down the intra prediction direction is selected from the encoding target pixel according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel in the memory. Subsequently, a reference pixel group used for predictive narrowing is selected from the reference pixels according to the relative positional relationship determined from the arrangement and coordinates of the encoded pixel and the encoding target pixel on the memory. Then, two or more prediction directions used for narrowing down the intra prediction direction are selected from the intra prediction direction candidates.

  Next, with reference to FIG. 9, the detailed operation of the process in step S2 shown in FIG. 1 according to the sixth embodiment will be described. FIG. 9 is a flowchart showing the detailed operation of the process in step S2 shown in FIG. First, the intra prediction unit 1 determines whether or not the reference coordinates in the prediction direction B are between pixels (step S621). That is, for the selected prediction direction, it is determined whether the reference position is between pixels as shown in FIG. If the result of this determination is that the reference position is between pixels and no pixel exists at that position, a decimal position pixel is generated from the surrounding pixels (step S622). On the other hand, if the reference position is not between pixels, the process of step S622 is skipped. Then, cost estimation in each prediction direction is performed using the set encoding target pixel group (step S623).

  The processing in step S3 in the sixth embodiment is the same as the processing operation shown in FIG. That is, candidates are narrowed down from all intra prediction directions that have a high correlation with the prediction direction in which the cost estimation result is minimum, and the intra prediction direction is determined from the candidates.

  In this way, when decimal position pixels are required in the intra prediction direction used for each narrowing down, by generating the decimal position pixels from the surrounding pixels while suppressing the calculation amount, it is possible to improve the coding efficiency while suppressing an increase in the calculation amount. Can be improved.

<Seventh embodiment>
Next, an intra prediction direction determination method according to a seventh embodiment of the present invention will be described. In the seventh embodiment, in the intra prediction direction narrowing-down process, the cost is obtained and compared based on a sum of differences such as an absolute value or a square error between the encoding target pixel 1 line adjacent to the reference pixel and the reference pixel. Here, one line refers to a line of pixels arranged in a certain range on the memory.

  Next, with reference to FIG. 11, the detailed operation of the processing in step S1 shown in FIG. 1 according to the seventh embodiment will be described. FIG. 11 is a flowchart showing a detailed operation of the process in step S1 shown in FIG. First, the intra prediction unit 1 selects the upper first stage (first left column) of the encoding block as an encoding target pixel group (step S711). Subsequently, the intra prediction unit 1 selects the pixel on one line (or one line left) as a reference pixel group used for prediction narrowing (step S712). And the intra estimation part 1 selects two or more prediction directions used for narrowing down from the candidate of an intra prediction direction (step S713).

  Next, with reference to FIG. 12, the detailed operation of the process in step S2 shown in FIG. 1 according to the seventh embodiment will be described. FIG. 12 is a flowchart showing a detailed operation of the process in step S2 shown in FIG. First, the intra prediction unit 1 uses the set encoding target pixel group for the selected prediction direction to calculate the cost of each prediction direction as a difference sum such as an absolute value or a square error between the encoding target pixel and the reference pixel. (Step S721).

  The processing in step S3 in the seventh embodiment is the same as the processing operation shown in FIG. That is, candidates are narrowed down from all intra prediction directions that have a high correlation with the prediction direction in which the cost estimation result is minimum, and the intra prediction direction is determined from the candidates.

  As described above, it is possible to execute by high-speed calculation while minimizing the amount of calculation required for narrowing down the intra prediction direction, and it is possible to suppress deterioration in encoding efficiency by using adjacent pixels as reference pixels.

<Eighth Embodiment>
Next, an intra prediction direction determination method according to the eighth embodiment of the present invention will be described. The eighth embodiment is a technique for fast prediction direction determination for HEVC intra prediction. In the present embodiment, as reference pixels, pixels on a line on one row of encoding target pixels stored in the memory as decoding pixels are used, and the encoding target pixels are stored in the memory one by one in the horizontal direction. (See the upper diagram in FIG. 13).

First, an adjacent pixel 1 line A _k (k = 0, 1,..., W + 1) and pixels B _j (j = 1,..., W) from the top of the encoding target pixel to the first stage are used. Of the prediction directions (prediction modes) of intra prediction, the absolute value sum (or the sum of squares, etc.) of errors between pixels is calculated as a cost by SSE for the directions 18 and 34 as shown in the following calculation formula. (See the lower diagram in FIG. 13). Here, w represents the number of pixels in the horizontal direction of the block on which intra prediction is performed.

A _k and B _j are a reference pixel group used for narrowing down the intra prediction direction and an encoding target pixel group used for narrowing down the intra prediction direction, and the prediction directions 18 and 34 are prediction directions used for narrowing down the intra prediction direction. . Here, the prediction directions (prediction modes) 18 and 34 can be accelerated in the data arrangement on the memory as described above. Further, both of the vertical and horizontal prediction modes 10 and 26 which are generally easy to select are in the middle positions. For this reason, even if either 18 or 34 is selected, it can be expected that the risk that the coding efficiency is deteriorated can be suppressed even if the selection is wrong by having 10 and 26 as prediction candidates.

  Next, the direction that minimizes the cost is set as the optimum prediction direction. Subsequently, processing is performed for each determined optimum prediction direction. When the optimal prediction direction is 18, the cost of intra prediction is calculated for prediction mode numbers 0, 1, 10 to 26 (corresponding to all intra prediction direction candidates that are highly correlated with the minimum prediction direction). The minimum intra cost is set as the optimal intra prediction direction.

  On the other hand, when the optimal prediction direction is 34, for prediction mode numbers 0, 1, 2 to 9, and 27 to 34 (corresponding to all intra prediction direction candidates that have a high correlation with the minimum prediction direction). The cost of intra prediction is calculated, and the one with the lowest cost is set as the optimal intra prediction direction.

<Ninth Embodiment>
Next, an intra prediction direction determination method according to a ninth embodiment of the present invention will be described. The ninth embodiment is a technique for determining a high-speed prediction direction for HEVC intra prediction. In the present embodiment, it is assumed that adjacent pixels stored in one column on the memory are used as the reference pixels for intra smoothing, and the encoding target pixels are stored in the memory one by one in the horizontal direction (see FIG. 14 See top diagram).

First, the adjacent pixels of one line _{A k (k = -1,0,1, ···} , w + 2) and the coding from the top of the target pixel in the first stage pixel _{B j (j = 1, ···} , w) For the prediction direction (prediction mode) of intra prediction, −6 (6 is assumed to be a direction rotated by 180 degrees), 14 and 26 directions, the error between pixels by SSE as in the following calculation formula Is calculated as a cost (see the lower diagram of FIG. 14). Here, w represents the number of pixels in the horizontal direction of the block on which intra prediction is performed.

A _k and B _j are a reference pixel group used for narrowing down the intra prediction direction and a coding target pixel group used for narrowing down the intra prediction direction, and prediction directions of −6, 14 and 26 are used for narrowing down the intra prediction direction. Direction. Here, when the number of prediction directions (prediction modes) increases to three, the amount of calculation increases correspondingly. However, the number of candidates for prediction directions (prediction modes) to be narrowed down in later processing is reduced, and can be appropriately narrowed down. The possibility increases. Further, the prediction directions of −6, 14 and 26 can be accelerated in the data arrangement on the memory as described above.

  In the present embodiment, when the memory is stored in the horizontal direction one column at a time, the cost calculation in the vertical direction can be easily performed by the high-speed calculation, but the cost calculation in the horizontal direction can be processed even by using the high-speed calculation. It is difficult to increase the speed. Therefore, when selecting -6, which is closer to the horizontal direction and capable of high-speed calculation and -6, which is a line target with respect to the vertical direction, as the prediction direction, it always includes either the horizontal or vertical direction as a prediction candidate. Can be expected to reduce the risk of worsening.

  Next, the direction that minimizes the cost is set as the optimum prediction direction. Subsequently, processing is performed for each determined optimum prediction direction. When the optimal prediction direction is −6, intra prediction is performed for prediction mode numbers 0, 1, 2, 10 and 32 to 34 (corresponding to all intra prediction direction candidates that are highly correlated with the minimum prediction direction). The cost of prediction is calculated, and the one with the lowest cost is set as the optimal intra prediction direction.

  In addition, when the optimal prediction direction is 14, the cost of intra prediction for prediction mode numbers 0, 1, 10 to 20 (corresponding to all intra prediction direction candidates having a high correlation with the minimum prediction direction). Calculation is performed, and the one with the lowest cost is set as the optimum intra prediction direction.

  In the case where the optimal prediction direction is 26, the cost of intra prediction for prediction mode numbers 0, 20, 20 to 32 (corresponding to all intra prediction direction candidates having a high correlation with the minimum prediction direction). Calculation is performed, and the one with the lowest cost is set as the optimum intra prediction direction.

<Tenth Embodiment>
Next, an intra prediction direction determination method according to the tenth embodiment of the present invention will be described. The tenth embodiment is a technique for fast prediction direction determination for HEVC intra prediction. In the present embodiment, it is assumed that adjacent pixels stored in one column on the memory are used as the reference pixels for intra smoothing, and the encoding target pixels are stored in the memory one by one in the horizontal direction (see FIG. 15 (see upper figure).

First, the adjacent pixel 1 line A _k (k = −1, 1,..., W + 2) and the first-stage pixel B _j (j = 1,..., W) from the top of the encoding target pixel are used. In the prediction direction (prediction mode) of intra prediction, for the directions of −6 (6 is a direction rotated by 180 degrees), 14 and 26, the absolute error between pixels by SSE as in the following calculation formula A sum of values (or a sum of squares or the like) is calculated as a cost (see the lower diagram of FIG. 15). Here, w represents the number of pixels in the horizontal direction of the block on which intra prediction is performed.

A _k and B _j are a reference pixel group used for narrowing down the intra prediction direction and a coding target pixel group used for narrowing down the intra prediction direction, and prediction directions of −6, 14 and 26 are used for narrowing down the intra prediction direction. Direction.

Next, the prediction mode number of the search range is determined according to the following conditions 1 to 3 for each calculated cost.
(Condition 1) The search range includes eight directions in the direction from the first cost to the second cost, and two directions in the third direction.
(Condition 2) If there is a third direction between the direction of the first cost and the second cost, and if either the first or second direction is outside the intra prediction direction region, The search range is set by rotating the current direction 180 degrees.
(Condition 3) If there is a direction outside the region within the search range set above, the direction rotated by 180 degrees in that direction is combined to be used as the search range.

  Specifically, it is as shown in FIG. A range of prediction mode numbers selected in A to F shown in FIG. 16 is shown in FIG.

  As described above, the degradation of coding efficiency is suppressed by narrowing down the intra prediction direction for calculating the cost based on the relative positional relationship between the reference pixel that is the pixel of the coded block adjacent to the coding target block. However, the amount of calculation can be reduced.

  You may make it implement | achieve the intra prediction direction determination method in embodiment mentioned above with a computer. In that case, a program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system and executed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Furthermore, the “computer-readable recording medium” dynamically holds a program for a short time like a communication line when transmitting a program via a network such as the Internet or a communication line such as a telephone line. In this case, a volatile memory inside a computer system serving as a server or a client in that case may be included and a program held for a certain period of time. Further, the program may be for realizing a part of the functions described above, and may be a program capable of realizing the functions described above in combination with a program already recorded in the computer system. It may be realized using hardware such as PLD (Programmable Logic Device) or FPGA (Field Programmable Gate Array).

  As mentioned above, although embodiment of this invention has been described with reference to drawings, the said embodiment is only the illustration of this invention, and it is clear that this invention is not limited to the said embodiment. is there. Therefore, additions, omissions, substitutions, and other modifications of the components may be made without departing from the technical idea and scope of the present invention.

  The present invention can be applied to applications where it is essential to reduce the amount of calculation in intra prediction of video coding.

  DESCRIPTION OF SYMBOLS 1 ... Intra prediction part, 2 ... Motion compensation part, 3 ... Subtraction part, 4 ... DCT / quantization part, 5 ... Entropy encoding part, 6 ... Inverse quantization / Inverse DCT unit, 7 ... adder, 8 ... loop filter, 9 ... motion vector predictor, 10 ... encoding control unit

Claims (8)

An intra-prediction direction determination method for determining an intra-prediction direction when performing intra-prediction on an encoding target block using a reference pixel that is a pixel of an adjacent encoded block,
Encoding is performed from reference pixels that are pixels of the encoded block adjacent to the encoding target block based on a relative positional relationship between the encoding target pixel of the encoding target block and the reference pixel on the memory. A target pixel group and a reference pixel group are selected, and two or more directions are selected from all intra prediction directions based on a relative positional relationship between the encoding target pixel group and the reference pixel group on a memory. 1 intra prediction direction selection step;
Calculating a cost for two or more selected directions, and selecting a direction in which the cost is minimized;
Correlation with respect to any of the correlation between the direction in which the cost is minimum and each intra prediction direction and the correlation between the direction not selected in the second intra prediction direction selection step and each intra prediction direction Determine whether the cost is relatively high, calculate the cost for the intra prediction direction determined to have a relatively high correlation with the direction in which the cost is minimum, and select the intra prediction direction in which the cost is minimum And a third intra prediction direction selection step. An intra prediction direction determination method comprising: An intra-prediction direction determination method for determining an intra-prediction direction when performing intra-prediction on an encoding target block using a reference pixel that is a pixel of an adjacent encoded block,
Based on the relative positional relationship between the encoding target pixel of the encoding target block and the reference pixel on the memory, encoding is performed from the reference pixel that is a pixel of the encoding target block and the adjacent encoded block, respectively. Select the target pixel group and the reference pixel group, and based on the relative positional relationship between the encoding target pixel group and the reference pixel group on the memory, all the intra prediction directions and the predetermined directions that are not the intra prediction directions A first intra prediction direction selection step of selecting two or more directions;
Calculating a cost for two or more selected directions, and selecting a direction in which the cost is minimized;
Correlation with respect to any of the correlation between the direction in which the cost is minimum and each intra prediction direction and the correlation between the direction not selected in the second intra prediction direction selection step and each intra prediction direction Determine whether the cost is relatively high, calculate the cost for the intra prediction direction determined to have a relatively high correlation with the direction in which the cost is minimum, and select the intra prediction direction in which the cost is minimum And a third intra prediction direction selection step that performs the intra prediction direction determination method.   3. The intra prediction direction determination method according to claim 1, wherein two or more directions selected in the first intra prediction direction selection step are in a relationship in which mutual correlation becomes low. The second intra prediction direction selection step calculates a cost for the two or more selected directions, and selects a plurality of directions from the smallest cost,
The third intra prediction direction selection step is selected in the second intra prediction direction selection step centering on the direction with the lowest cost among the directions selected in the second intra prediction direction selection step. A predetermined range determined using a direction excluding the direction with the lowest cost among the determined directions, the cost is calculated for the intra prediction directions included in the predetermined range, and the intra with the lowest cost among them is calculated. The intra prediction direction determination method according to claim 1, wherein a prediction direction is selected.   The first intra prediction direction selection step is based on the block size of the encoding target block in addition to the relative positional relationship of the encoding target pixel of the encoding target block and the reference pixel on the memory, The intra prediction direction determination method according to any one of claims 1 to 4, wherein an encoding target pixel group and the reference pixel group are selected.   When there is no reference pixel that is a pixel of the adjacent encoded block at a position to be referred to, the first intra prediction direction selection step calculates a decimal number from the reference pixel that is a pixel of the adjacent encoded block. 6. The intra prediction direction determination method according to claim 1, wherein a pixel is generated, the generated decimal pixel is included in the reference pixel group, and the reference pixel group is selected. The relative positional relationship between the encoding target pixel of the encoding target block and the reference pixel on the memory is such that the encoding target pixels are stored in the memory one by one in the horizontal direction or the vertical direction, and are adjacent to each other. A reference pixel that is a pixel of an encoded block is a positional relationship that is stored in the memory in one column in the horizontal direction on one of them or in the vertical direction on the left one of them.
The encoding target pixel group is the first row from the top or the first column from the left of the encoding target block,
The intra prediction direction determination method according to claim 1, wherein the reference pixel group is a reference pixel that is one pixel above or one pixel left of the encoding target pixel group.   The intra prediction direction determination program for making a computer perform the intra prediction direction determination method of any one of Claim 1 to 7.

Images