JP2009225163A - Video encoder and video encoding method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To curtail a computational complexity required to decide a prediction mode of encoding. <P>SOLUTION: This video encoder comprises: an intra-prediction image generating part 41 for searching a prediction image in an intra-prediction for decision based on an encode cost; an inter-prediction image generating part 42 for layering search processes of the prediction image in an inter-prediction in response to search accuracy, successively switching the search accuracy from the low accuracy to the high accuracy for each layer, and searching and determining the prediction image in the inter-prediction for decision based on the encode cost for each layer; and a comparing part 46 for comparing the encode cost of the prediction image decided by the intra-prediction image generating part with the encode cost of the prediction image decided by the inter-prediction image generating part for each layer, deciding either prediction mode of the intra-prediction or inter-prediction for each layer based on a comparison result or reserving the decision, and stopping the search processes of the subsequent layers in the inter-prediction image generating part when the prediction mode is decided to a mode for intra-prediction. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a moving picture coding apparatus and a moving picture coding method that employ predictive coding and arithmetic coding.

  In recent years, digitization of image processing has become widespread. In order to reduce the amount of codes, digitized image data is recorded in H.264 format. In many cases, transmission and recording are performed after encoding using a moving image encoding method such as H.264 and AVC standards.

  H. In the H.264 standard and the like, predictive encoding of a plurality of block shapes is employed in a plurality of prediction modes, and an optimal mode is selected from a large number of prediction modes for each image block, and predictive encoding is performed. The optimum prediction mode differs depending on the property of the image block, and the encoding efficiency can be improved by selecting an appropriate mode.

  A rate-distortion optimization method based on Lagrange's undetermined multiplier method is known as a method for selecting an optimum mode from a plurality of prediction modes. In this rate-distortion optimization method, the encoding cost is calculated from the encoding distortion D (the square error between the current image and the reproduced image) and the generated code amount when encoding is performed in each prediction mode. The encoding efficiency in image encoding means how much encoding can be performed with a small amount of code and with little deterioration in image quality. That is, the encoding efficiency increases as the encoding cost decreases. Therefore, the encoding efficiency can be maximized by selecting, for each image block, a prediction mode that minimizes the encoding cost for a given quantization parameter.

  H. In H.264, intra prediction for reducing the code amount by intra prediction and inter prediction for reducing the code amount by inter prediction are performed. Intra prediction and inter prediction include various modes, and the amount of calculation for determining the mode with the minimum coding cost is extremely large. In particular, H.C. In the inter prediction in H.264, it is possible to select a prediction mode of 1/4 pixel accuracy as the reference image search accuracy, and enormous calculation is necessary for calculating the coding cost.

  Therefore, conventionally, a technique for reducing the amount of calculation for calculating the coding cost necessary for mode determination has been adopted. For example, in inter prediction, there is a technique for hierarchizing each search accuracy calculation. In this technique, a reference image block candidate is obtained by setting a predetermined search range with the lowest search accuracy, that is, two-pixel accuracy. Then, a relatively narrow search range is newly set centering on the obtained candidate reference image block, and the next search accuracy, that is, a one-pixel accuracy search is performed within this search range. Thereafter, similarly, the search range of the next search accuracy is set around the candidate block obtained with each search accuracy. Since the search range can be narrowed, it is possible to reduce the amount of calculation required for searching for reference image block candidates. Thus, finally, a candidate for a reference image block is obtained with 1/4 pixel accuracy, and the encoding cost is calculated. The encoding mode is determined by comparing the minimum encoding cost in the inter prediction thus obtained and the minimum encoding cost in the intra prediction.

  However, even if hierarchization is performed, a search with high search accuracy, for example, 1/4 pixel accuracy, requires a huge amount of calculation, and the calculation amount as a whole cannot be reduced sufficiently.

Patent Document 1 discloses a technique for selecting either inter prediction or intra prediction based on the similarity calculated from an encoding target image and a reference image. However, in the proposal of Patent Document 1, switching between inter prediction and intra prediction is performed based only on similarity, and there is a possibility that an error occurs in the determination of the optimal prediction mode. In addition, when inter prediction is selected, there is a problem that the calculation amount cannot be reduced because high-precision search processing is performed.
JP 2006-20217 A

  An object of the present invention is to provide a moving picture coding apparatus and a moving picture coding method capable of determining an optimum prediction mode with a small amount of calculation and performing highly efficient coding.

A video encoding apparatus according to an aspect of the present invention provides an intra-predicted image generation unit that searches for and determines a predicted image in intra prediction based on a coding cost, and search processing for a predicted image in inter prediction with search accuracy. The inter prediction image generation unit for sequentially searching for and determining the prediction image in the inter prediction based on the coding cost for each layer, by layering according to the layer, sequentially switching the search accuracy from low accuracy to high accuracy for each layer, and The encoding cost of the prediction image determined by the intra prediction image generation unit and the encoding cost of the prediction image determined by the inter prediction image generation unit are compared for each layer, and intra prediction is performed for each layer based on the comparison result. If the prediction mode is determined as the intra prediction mode or the prediction mode is determined as the intra prediction mode or the inter prediction mode is determined. Which is characterized by comprising a comparator unit stopping the search process following layer in the terpolymer predicted image generation unit,
A moving image encoding method according to an aspect of the present invention searches for and determines a prediction image in intra prediction based on an encoding cost, hierarchizes search processing for a prediction image in inter prediction according to search accuracy, and performs search The accuracy is sequentially switched from low accuracy to high accuracy for each layer, and the prediction image in the inter prediction is searched and determined based on the encoding cost for each layer, and the encoding cost of the prediction image in the intra prediction and the inter prediction are determined. The coding cost of the prediction image in the prediction is compared for each layer, and the prediction mode of either intra prediction or inter prediction is determined for each layer based on the comparison result, or the determination is suspended, and the prediction mode is set. When the intra prediction mode is determined, the subsequent search process in the inter prediction is stopped.

  According to the present invention, it is possible to determine an optimal prediction mode with a small amount of calculation and perform highly efficient encoding.

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

  FIG. 1 is a block diagram showing a prediction signal generation unit employed in a video encoding apparatus according to an embodiment of the present invention. FIG. 2 is a block diagram showing the overall configuration of the moving picture coding apparatus.

  In FIG. 2, an image signal is input to the encoding unit 11. This image signal is supplied to the subtracter 12 and the prediction signal generation unit 23 of the encoding unit 11. The subtracter 12 receives a predicted image signal from an MB prediction mode selection unit 24 described later, and outputs a difference value (prediction error) between the input image signal and the predicted image signal. The prediction error is obtained in units of blocks having a size corresponding to the prediction mode.

  The prediction error from the subtracter 12 is given to the orthogonal transform unit 13. The orthogonal transform unit 13 performs orthogonal transform on the input prediction error and transforms it into a spatial frequency component in units of blocks. Thereby, spatial correlation components can be reduced. The transform coefficient from the orthogonal transform unit 13 is given to the quantization unit 14. The quantization unit 14 quantizes the input transform coefficient, thereby reducing the redundancy of the signal of the block and outputting it.

  The quantized output from the quantizing unit 14 is supplied to the CABAC encoding unit 15. The CABAC encoding unit 15 includes a binarizing unit 16 and an arithmetic encoding unit 17. The binarization unit 16 receives a syntax element as a quantization output and also receives prediction mode information from the MB prediction mode selection unit 24, and converts the syntax element into a binary symbol string of variable length. . The arithmetic encoding unit 17 arithmetically encodes the binary symbol sequence from the binarizing unit 16 according to the occurrence probability of the symbol. In this way, the CABAC encoding unit 15 outputs the encoded output with the code amount further reduced to the output buffer 18.

  In addition, information regarding a prediction method such as a quantization coefficient output from the MB prediction mode selection unit 24 is given to the variable length encoding unit 25. The variable length coding unit 25 performs variable length coding on the input information and outputs it to the output buffer 18. The output buffer 18 outputs the output of the CABAC encoder 15 and the output of the variable length encoder 25 as encoded data at a predetermined bit rate.

  The encoding control unit 31 performs control of the entire encoding process in addition to rate control for controlling the generated code amount by feedback control of generated code amount, quantization characteristic control, and the like, control of prediction mode determination processing, and the like.

  The output of the quantization unit 14 is also supplied to the inverse quantization unit 19 in order to generate a predicted image signal. The inverse quantization unit 19, the inverse orthogonal transform unit 20, the adder 21 and the reference image memory 22 constitute a local decoder. The inverse quantization unit 19 inversely quantizes the input quantized output, returns the data before the quantization (transform coefficient), and outputs it to the inverse orthogonal transform unit 20. The inverse orthogonal transform unit 20 obtains a prediction error before orthogonal transform by performing inverse orthogonal transform on the input transform coefficient. This prediction error is given to the adder 21.

  The adder 21 is also supplied with the prediction image signal from the MB prediction mode selection unit 24, and restores the input image by adding the prediction error and the prediction image signal. The restored image from the adder 21 is given to the reference image memory 22 as a reference image.

  An input image signal is also input to the prediction signal generation unit 23. As will be described later, the prediction signal generation unit 23 generates a prediction image signal corresponding to each prediction mode, outputs the prediction image signal to the MB prediction mode selection unit 24, and instructs a prediction mode to be determined. An instruction signal is generated and output to the MB prediction mode selection unit 24. The MB prediction mode selection unit 24 is provided with a prediction image signal and a mode instruction signal for each prediction mode from the prediction signal generation unit 23. The MB prediction mode selection unit 24 selects a prediction image signal based on the mode instruction signal indicating the prediction mode, and outputs the prediction image signal to the subtracter 12 and the adder 21.

  FIG. 1 shows a specific configuration of the prediction signal generation unit 23 in FIG.

  The prediction signal generation unit 23 can generate a prediction image signal based on intra prediction and inter prediction. In addition, in the inter prediction, the prediction signal generation unit 23 can obtain a prediction image signal with a plurality of pixel accuracy, for example, two pixel accuracy, one pixel accuracy, ½ pixel accuracy, and ¼ pixel accuracy. In the present embodiment, as will be described later, the prediction signal generation unit 23 can reduce the amount of calculation required for determining the prediction mode.

  An input image signal is supplied to the intra predicted image generation unit 41. The intra predicted image generation unit 41 is provided with an input image signal and generates a predicted image (intra predicted image) using an image in the screen. For example, the intra predicted image generation unit 41 generates an intra predicted image using pixels adjacent to the block to be encoded. The intra predicted image generation unit 41 outputs the generated intra predicted image to the difference image generation unit 44 and also outputs it to the MB prediction mode selection unit 24. Finally, as will be described later, the intra-predicted image generation unit 41 outputs an intra-predicted image of a mode in which the lowest coding cost is obtained among the intra-prediction modes to the MB prediction mode selection unit 24. It is like that.

  The inter prediction image generation unit 42 is given a reference image signal, and generates a prediction image (inter prediction image) using an image (reference image) of another screen. In this case, the inter prediction image generation unit 42 generates an inter prediction image so that the encoding cost is minimized.

  In the present embodiment, the inter-predicted image generation unit 42 is controlled by the comparison unit 46 described later, and stratifies search processing for generating an inter-predicted image. In other words, the inter predicted image generation unit 42 performs a search sequentially by switching the search accuracy for each hierarchy from the lowest search accuracy to the higher search accuracy. The present embodiment is referred to as H.264. When applied to H.264, 2-pixel accuracy, 1-pixel accuracy, 1 / 2-pixel accuracy, and 1 / 4-pixel accuracy are employed as search accuracy.

  For example, the inter prediction image generation unit 42 sets a search range in the reference image for each layer, and within the set search range, the macro block is changed while sequentially changing the blocking position according to the search accuracy of the layer. It selects and outputs to the difference image generation part 44 as an inter estimated image. Then, the inter prediction image generation unit 42 finally receives information on the encoding cost from the comparison unit 46 described later, and selects an inter prediction image having the lowest encoding cost in the hierarchy as an MB prediction mode selection. To the unit 24.

  The difference image generation unit 44 also receives an input image signal, obtains a difference between the intra predicted image signal and the input image signal, and outputs the difference image to the encoding cost calculation unit 45. Further, the difference image generation unit 44 obtains a difference between the inter predicted image signal and the input image signal, and outputs the difference image to the encoding cost calculation unit 45.

  The encoding cost calculation unit 45 calculates the encoding cost based on the output of the difference image generation unit 44. That is, the encoding cost calculation unit 45 calculates the encoding cost of the intra predicted image from the difference between the intra predicted image and the input image. Also, the encoding cost calculation unit 45 calculates the encoding cost of the inter predicted image from the difference between the inter predicted image and the input image.

  The comparison unit 46 includes a memory 47, and performs comparison while storing the encoding cost of the input intra prediction image and the encoding cost of the inter prediction image in the memory 47. That is, the comparison unit 46 compares the encoding costs for each intra prediction image of each mode in the intra prediction, and instructs the intra prediction image generation unit 41 which mode can obtain the lowest encoding cost. In addition, the comparison unit 46 compares the encoding cost of the intra prediction image in each layer of the inter prediction, and the inter prediction image (macroblock) from which the lowest encoding cost is obtained for each layer is the intra prediction image generation unit. 42.

  FIG. 3 is an explanatory diagram for explaining hierarchical search processing in inter prediction. 3A shows the search processing with the lowest search accuracy, FIG. 3B shows the search processing with the second lowest search accuracy, and FIG. 3C shows the search processing with the third lowest search accuracy. Show.

  Assume that a macroblock (hereinafter referred to as an inter-predicted image macroblock) that is an inter-predicted image is obtained for the macroblock (processing MB) 52 in FIG. First, the inter prediction image generation unit 42 performs a search with the lowest search accuracy. In other words, the inter predicted image generation unit 42 sets a predetermined search range R1 around the macroblock of the reference image at the same position as the processing MB52 of the input image. The inter prediction image generation unit 42 sequentially selects inter prediction image macroblocks while changing the blocking position according to the search accuracy within the search range R1.

  The inter prediction image from the inter prediction image generation unit 42 is given to the difference image generation unit 44, and a difference image from the input image is obtained. The encoding cost calculation unit 45 obtains the encoding cost of the inter prediction image from the difference image. The encoding cost is sequentially supplied to the comparison unit 46. The comparison unit 46 sets a macroblock that provides the minimum coding cost within the search range R1 as a candidate C1 (x mark).

  When the search with the second lowest search accuracy is instructed from the comparison unit 46, the inter predicted image generation unit 42 sets a predetermined search range R2 around the macroblock of the candidate C1. The inter prediction image generation unit 42 sequentially selects inter prediction image macroblocks while changing the blocking position according to the second lowest search accuracy within the search range R2. The encoding cost of the inter prediction image from the inter prediction image generation unit 42 is compared by the comparison unit 46. In this way, the macroblock that provides the lowest coding cost is obtained as the candidate C2.

  Further, when a search with the third lowest search accuracy is instructed by the comparison unit 46, the inter predicted image generation unit 42 sets a predetermined search range R3 centering on the macroblock of the candidate C2. The inter prediction image generation unit 42 sequentially selects inter prediction image macroblocks while changing the blocking position according to the third lowest search accuracy within the search range R3. The inter prediction image generation unit 42 is designated by the comparison unit 46 as a macroblock that provides the lowest coding cost, and sets the macroblock as a candidate C3.

  Thereafter, similarly, the inter-predicted image generation unit 42 searches for inter-predicted images hierarchically while switching from low search accuracy to high search accuracy in accordance with an instruction from the comparison unit 46.

  In the present embodiment, the comparison unit 46 compares the encoding cost in intra prediction and the encoding cost in inter prediction for each layer of inter prediction. The comparison unit 46 sets a threshold value for each search with each search accuracy, and when searching with a predetermined search accuracy, the coding cost of the intra prediction image is smaller than the coding cost of the inter prediction image and is smaller than the predetermined cost. Outputs a mode instruction signal for setting the intra prediction mode, and causes the MB prediction mode selection unit 24 to output the intra prediction image signal from the intra prediction image generation unit 41.

  In addition, when the difference between the coding cost of the intra-predicted image and the coding cost of the inter-predicted image is within a predetermined threshold during the search with a predetermined search accuracy, the comparing unit 46 is higher than the search accuracy. An instruction is given to the inter-predicted image generation unit 42 so as to continue searching for search accuracy.

  In addition, the comparison unit 46 sets the inter prediction mode when the coding cost of the inter prediction image is smaller than the coding cost of the intra prediction image at a predetermined search accuracy and exceeds a predetermined threshold. While outputting the mode instruction signal, the inter prediction image generation unit 42 outputs the inter prediction image signal obtained at the time of the search with the predetermined search accuracy to the MB prediction mode selection unit 24. In this case, the comparison unit 46 stops the search process (inter prediction image generation process) of the inter prediction image generation unit 42 thereafter.

  In the present embodiment, the inter prediction image generation unit 42 is controlled by the comparison unit 46 to search for the inter prediction image. That is, the inter prediction image generation unit 42 performs the calculation of the inter prediction image for each search accuracy according to the instruction from the comparison unit 46, and when the mode instruction signal for determining the intra prediction mode is output from the comparison unit 46, The operation in the hierarchy is stopped.

  Next, the operation of the embodiment configured as described above will be described with reference to FIGS. FIG. 4 is a flowchart for explaining prediction mode determination processing. FIG. 5 is a flowchart specifically showing the intra prediction candidate determination process in FIG. 6 to 9 show predicted image generation processing (search processing) in inter prediction, and examples using search processing with 2-pixel accuracy, 1-pixel accuracy, 1 / 2-pixel accuracy, or 1 / 4-pixel accuracy, respectively. Show. 5 to 9 show a flow when the processing of each unit of the prediction signal generation unit 23 shown in FIG. 1 is realized by software.

  The input image signal is given to the prediction signal generator 23. In step S1 of FIG. 4, the prediction signal generation unit 23 determines a prediction image candidate that minimizes the coding cost by intra prediction. That is, the input image signal is given to the intra prediction image generation unit 41 in the prediction signal generation unit 23, and a prediction image in each mode is generated. The difference image generation unit 44 obtains a difference image between the input image and the predicted image, and the encoding cost calculation unit 45 calculates the encoding cost of the difference image. The comparison unit 46 compares the coding costs of the respective modes in intra prediction, and obtains a predicted image that can obtain the minimum coding cost.

  The prediction signal generation unit 23 can be configured using a computer system, and FIG. 5 shows a process of generating an intra prediction image by the CPU. In step S11 of FIG. 5, the CPU first initializes a variable intraCost of the minimum encoding cost to a maximum value that can be taken as the encoding cost. Next, in step S12, an intra predicted image pred [x, y] in a predetermined prediction mode is generated. Next, the difference between the intra-predicted image and the input image is obtained (step S13), and the coding cost is calculated for the difference image diff [x, y].

  The encoding cost is given as the sum of the coefficient cost coeffCost that is the cost of the difference image and the prediction mode cost modeCost that is the cost when the prediction mode information is encoded. In step S14, the coefficient cost is calculated, and in step S15, the prediction mode cost is calculated. The coefficient cost coeffCost is obtained, for example, by the absolute value sum calculation formula shown in the following formula (1).

coeffCost = ΣΣabs (diff [x, y]) (1)
Note that abs () indicates the absolute value of the value in ().

  Next, in step S <b> 16, the CPU substitutes the sum of the coefficient cost value coeffCost and the prediction mode cost value modeCost into a variable tmpCost used for comparing the coding cost of each mode at the time of intra prediction. In step S <b> 17, the CPU determines the magnitude of tmpCost and the variable intraCost of the minimum encoding cost. If tmpCost is smaller than intraCost, the CPU assigns tmpCost to intraCost and updates it (step S18). Otherwise, it does not update intraCost.

  The CPU repeats steps S12 to S18 for each prediction mode. In this way, the value of the smallest coding cost of all prediction modes is stored in intraCost. In step S <b> 19, the CPU adds an adjustment parameter intraCostOffset for comparison between the encoding cost at the time of intra prediction and the encoding cost at the time of inter prediction, to intraCost. In this way, an intra prediction image that obtains the minimum coding cost intraCost and the minimum coding cost is obtained.

  Next, a candidate by inter prediction is determined. Now, the search accuracy of the inter prediction image is assumed to be 2 pixel accuracy, 1 pixel accuracy, 1/2 pixel accuracy, or 1/4 pixel accuracy. In the present embodiment, the search for inter predicted images is performed hierarchically.

  First, in step S2 of FIG. 4, a search process with the lowest search accuracy is performed. The predicted signal generation unit 23 also determines a predicted image candidate that minimizes the coding cost even in inter prediction. The inter prediction image generation unit 42 is provided with a reference image signal, selects a macroblock while changing the blocking position according to the lowest search accuracy within the search range, and sequentially generates a difference image generation unit as an inter prediction image. 44. The difference image generation unit 44 obtains a difference image between the input image and the inter prediction image, and the encoding cost calculation unit 45 calculates the encoding cost of the difference image. The comparison unit 46 compares the encoding costs of the inter-predicted images and obtains an inter-predicted image that provides the minimum encoding cost interCost0.

  FIG. 6 shows a two-pixel precision search process by the CPU. In step S21 in FIG. 6, the CPU first initializes the variable coding cost interCost0 of the minimum coding cost to the maximum value that can be taken as the coding cost. Next, in step S22, an inter prediction image is generated. That is, the CPU searches for a motion vector with 2-pixel accuracy within a predetermined search range, and determines a motion vector candidate MV candidate Mv0. Next, the CPU generates an inter predicted image pred [x, y] from the motion vector candidates with 2-pixel accuracy. Next, the difference between the inter predicted image and the input image is obtained (step S23), and the coding cost is calculated for the difference image diff [x, y].

  The encoding cost in the inter prediction is a coefficient cost coeffCost that is a cost of the difference image diff [x, y] and an MV cost mvCost that is a cost in the case of encoding motion vector information indicating the blocking position of the inter prediction image. Is given as the sum of In step S24, the coefficient cost is calculated, and in step S25, the MV cost is calculated.

  Next, in step S26, the sum of the coefficient cost value coeffCost and the MV cost value mvCost is substituted into a variable tmpCost used for comparison of the coding cost of each mode during inter prediction. In step S27, it is determined whether tmpCost is the minimum encoding cost variable interCost0. If tmpCost is smaller than interCost0, tmpCost is substituted and updated for interCost0 (step S28). Otherwise, interCost0 is not updated.

  Steps S22 to S28 are repeated to obtain the minimum encoding cost while changing the blocking position with two-pixel accuracy within the search range. Thus, interCost0 stores the smallest coding cost value within the search range. In step S29, the adjustment parameter interCostOffset for comparing the coding cost at the time of intra prediction and the coding cost at the time of inter prediction is added to interCost0. Thus, when the search accuracy is two-pixel accuracy, an inter prediction image that obtains the minimum coding cost and the minimum coding cost is obtained.

  In this embodiment, as will be described later, since a predetermined threshold is used when comparing the encoding cost of intra prediction and the encoding cost of inter prediction, adjustment can be made by appropriately setting this threshold. The addition of the parameter interCostOffset may be omitted.

  Next, in step S3 in FIG. 4, the CPU determines whether or not the search process in step S2 is the search process with the highest search accuracy. If it is not the search process with the highest search accuracy, the process proceeds to step S4, and it is determined whether or not the encoding cost (intra cost) in intra prediction is sufficiently small. For example, as shown in the following equation (2), the CPU sets a predetermined threshold thIntra0 from the encoding cost intraCost in intra prediction based on the encoding cost interCost in inter prediction (in this case, interCost0 obtained with two-pixel accuracy). It is determined whether or not it is too small.

intraCost <interCost0-thIntra0 (2)
If the above equation (2) is not satisfied, it is determined in the next step S6 whether or not the inter cost is sufficiently small. For example, as shown in the following equation (3), the CPU determines whether or not the coding cost interCost in inter prediction is smaller than the coding cost intraCost in intra prediction (in this case, interCost0) exceeding a predetermined threshold thInter0. Determine.

interCost0 <intraCost-thInter0 (3)
If it is determined that the above equation (3) is satisfied, that is, if it is determined that the inter prediction mode can provide a sufficiently lower encoding cost than the intra prediction mode, the CPU proceeds to step S7. In the same manner as in the prior art, the search processing is sequentially performed from the next highest search accuracy to the highest search accuracy.

  FIG. 7 shows search processing with one pixel accuracy by the CPU. In the search process with one pixel accuracy, as shown in FIG. 3, the search range is set to a range including candidates in the search process with two pixel accuracy. Therefore, the coding cost of intra prediction obtained with one pixel accuracy is always a value equal to or lower than the coding cost of intra prediction obtained with two pixel accuracy. Therefore, in step S31 of FIG. 7, the CPU initializes the variable interCost1 of the minimum encoding cost in the search process with one pixel accuracy to the interCost0 that is the maximum value that can be taken as the encoding cost.

  Next, in step S32, an inter prediction image is generated. That is, the CPU searches for a motion vector with one pixel accuracy within a predetermined search range, determines a motion vector candidate, and generates an inter-predicted image pred [x, y] from the determined motion vector candidate. The subsequent steps S33 to S39 are the same as the two-pixel precision search process of FIG. Thus, in step S29, the minimum encoding cost interCost1 in the search process with one pixel accuracy is obtained.

  Next, the CPU performs a search process with 1/2 pixel accuracy shown in FIG. In the search process with 1/2 pixel accuracy, in step S41, the CPU initially sets a variable interCost2 of the minimum encoding cost in the search process with 1/2 pixel accuracy to interCost1, which is the maximum value that can be taken as the encoding cost. Turn into.

  Next, in step S42, an inter prediction image is generated. That is, the CPU searches for a motion vector with a 1/2 pixel accuracy within a predetermined search range, determines a motion vector candidate, and generates an inter predicted image pred [x, y] from the determined motion vector candidate. The subsequent steps S43 to S49 are the same as the two-pixel precision search process of FIG. Thus, in step S49, the minimum encoding cost interCost2 in the search process with 1/2 pixel accuracy is obtained.

  Next, the CPU performs a search process with 1/4 pixel accuracy shown in FIG. In the 1/4 pixel precision search process, in step S51, the CPU initially sets a variable coding cost interCost3 of the minimum coding cost in the 1/4 pixel precision search process to interCost2, which is the maximum value that can be taken as the coding cost. Turn into.

  Next, in step S52, an inter prediction image is generated. That is, the CPU searches for a motion vector with 1/4 pixel accuracy within a predetermined search range, determines a motion vector candidate, and generates an inter predicted image pred [x, y] from the determined motion vector candidate. The subsequent processing of steps S53 to S59 is the same as the two-pixel precision search processing of FIG. Thus, in step S59, the minimum encoding cost interCost3 in the search process with 1/4 pixel accuracy is obtained.

  In the next step S8, the CPU sets an inter prediction mode. That is, when the process of step S7 is performed, in the inter prediction, the process is performed with all the search precisions from the lowest search precision to the highest search precision.

  On the other hand, if it is determined in step S4 that the above equation (2) is satisfied and the process proceeds to step S5, the process with all search accuracy from the minimum search accuracy to the maximum search accuracy is performed. Not done. For example, if it is determined that the coding cost obtained in the search process with the minimum search accuracy for inter prediction is sufficiently small, the CPU shifts the process from step S4 to step S5 to set the intra prediction mode. To do. That is, in this case, only the process with the lowest search accuracy is performed as the inter prediction search process.

  In step S4, if it is determined that the above equation (2) is not satisfied, and further, if it is determined in step S6 that the above equation (3) is not satisfied, either the intra prediction mode or the inter prediction mode is selected. In step S9, the search processing with the next highest search accuracy is performed without determining whether to select the mode.

  For example, when steps S4 and S6 are determined based on the coding cost obtained by the search process with two pixel accuracy and the process proceeds to step S9, the search process with one pixel accuracy is performed at step S9. . Then, determinations in steps S4 and S6 are made based on the encoding cost obtained by the search process with one pixel accuracy.

  For example, in steps S4 and S6, it is determined whether or not the following expressions (4) and (5) are satisfied. In addition, CPU resets the threshold value used for step S4, S6 for every hierarchy. Since the encoding cost of inter prediction becomes smaller as the search accuracy becomes higher, it is better to reduce the threshold according to the search accuracy.

intraCost <interCost1-thIntra1 (4)
interCost1 <intraCost-thInter1 (5)
When the above equation (5) is satisfied, the next step S7 is executed, and when the above equation (4) is satisfied, the next step S5 is executed. When step S5 is executed, the intra prediction mode is determined only by the search process with the 2-pixel accuracy and the 1-pixel accuracy of inter prediction.

  Similarly, when the search process with 1/2 pixel accuracy is performed in step S9, the determinations of steps S4 and S6 are made based on the coding cost obtained by the search process with 1/2 pixel accuracy. In this case, the CPU determines whether or not the encoding cost satisfies the following expressions (6) and (7) in steps S4 and S6.

intraCost <interCost2-thIntra2 (6)
interCost2 <intraCost-thInter2 (7)
When the above equation (7) is satisfied, the next step S7 is executed, and when the above equation (6) is satisfied, the next step S5 is executed. When step S5 is executed, the intra prediction mode is determined by the search process with the inter-prediction 2-pixel accuracy to 1 / 2-pixel accuracy.

  When the 1/4 pixel precision search process is performed in step S9, the CPU shifts the process from step S3 to step S10, and the encoding cost and the intra obtained by the 1/4 pixel precision search process. Compare with cost. In this case, the CPU determines in step S10 whether, for example, the encoding cost satisfies the following expression (8).

interCost3> intraCost (8)
If the expression (8) is satisfied, the intra prediction mode is set in step S5. If the expression (8) is not satisfied, the inter prediction mode is set in step S8.

  When the intra prediction mode is set in step S5, the CPU stops the subsequent search process. In FIG. 1, the search process can be stopped by the comparison unit 46 controlling the inter predicted image generation unit 42. That is, when the intra prediction mode is selected in the comparison process in step S4, the search process after the hierarchy is stopped, and the calculation for that level can be omitted. That is, it is possible to reduce the amount of calculation in inter prediction.

  As described above, in the present embodiment, the inter prediction search process is hierarchized for each search accuracy, and the mode determination process is performed for each hierarchy based on the intra prediction and the inter prediction encoding cost. When the determination is made, the search process for the next hierarchy is stopped. Thereby, the calculation amount in the inter prediction can be reduced, and the power consumption can be reduced. In addition, since the amount of calculation for inter prediction is reduced, the time required for mode determination can be shortened.

  In addition, this invention is not limited to said each embodiment, A various deformation | transformation is possible. For example, the configuration of the encoding device and the encoding method are not particularly limited.

The block diagram which shows the prediction signal production | generation part employ | adopted for the moving image encoder which concerns on embodiment of this invention. The block diagram which shows the whole structure of a moving image encoder. Explanatory drawing for demonstrating the hierarchical search process in inter prediction. FIG. 4 is a flowchart for explaining prediction mode determination processing. FIG. 5 is a flowchart specifically showing intra prediction candidate determination processing in FIG. 4. The flowchart which shows the prediction image generation process of 2 pixel precision in inter prediction. The flowchart which shows the prediction image generation process of 1 pixel precision in inter prediction. The flowchart which shows the prediction image generation process of the 1/2 pixel precision in inter prediction. The flowchart which shows the prediction image generation process of the 1/4 pixel precision in inter prediction.

Explanation of symbols

    DESCRIPTION OF SYMBOLS 23 ... Prediction signal generation part, 41 ... Intra prediction image generation part, 42 ... Inter prediction image generation part, 44 ... Difference image generation part, 45 ... Coding cost calculation part, 46 ... Comparison part.

Claims (5)

An intra-prediction image generation unit that searches and determines a prediction image in intra prediction based on the coding cost;
Search processing for predicted images in inter prediction is hierarchized according to search accuracy, search accuracy is sequentially switched from low to high accuracy for each layer, and prediction images in inter prediction are searched based on coding cost for each layer An inter-prediction image generation unit to be determined,
The encoding cost of the prediction image determined by the intra-prediction image generation unit and the encoding cost of the prediction image determined by the inter-prediction image generation unit are compared for each hierarchy, and the intra cost is calculated for each hierarchy based on the comparison result. When the prediction mode of prediction or inter prediction is determined or suspended, and the prediction mode is determined to be the intra prediction mode, the search process of the subsequent layers in the inter prediction image generation unit is stopped. A moving picture coding apparatus comprising: a comparison unit. The comparison unit includes:
When the encoding cost of the prediction image determined by the intra prediction image generation unit is smaller than the encoding cost of the prediction image determined by the inter prediction image generation unit by exceeding a predetermined threshold, the prediction mode is changed to the intra prediction mode. Decide
When the encoding cost of the prediction image determined by the inter prediction image generation unit is smaller than the encoding cost of the prediction image determined by the intra prediction image generation unit exceeding a predetermined threshold, the prediction mode is set to the inter prediction mode. Decide
When the difference between the encoding cost of the prediction image determined by the intra prediction image generation unit and the encoding cost of the prediction image determined by the inter prediction image generation unit is within a predetermined threshold, the prediction mode is determined. The video encoding apparatus according to claim 1, wherein the video encoding apparatus holds the video.   The moving image encoding apparatus according to claim 1 or 2, wherein the inter predicted image generation unit sets a search range including a predicted image determined in the immediately preceding hierarchy as a search range of a predicted image. .   The moving image encoding apparatus according to claim 2, wherein the comparison unit resets the predetermined threshold for each of the hierarchies. Search and determine the prediction image in intra prediction based on the coding cost,
Search processing for predicted images in inter prediction is hierarchized according to search accuracy, search accuracy is sequentially switched from low to high accuracy for each layer, and prediction images in inter prediction are searched based on coding cost for each layer Decide
Comparing the encoding cost of the prediction image in the intra prediction and the encoding cost of the prediction image in the inter prediction for each layer,
Based on the comparison result, determine any prediction mode of intra prediction and inter prediction for each tier or hold the decision,
A moving picture coding method characterized by stopping subsequent search processing in the inter prediction when the prediction mode is determined to be an intra prediction mode.

Images