JP2009038746A - Image information encoding device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image information encoding device which suppresses the impact of estimation error to the minimum, even when rate control is performed based on the estimated code amount. <P>SOLUTION: The image information encoding device includes a motion picture compressing means and an arithmetic encoding means, and includes a code amount estimating means which generates an estimated code amount by estimating the real code amount after arithmetic encoding by the arithmetic encoding means using data about the motion picture before arithmetic encoding by the arithmetic encoding means, an estimated code amount substituting means which reduces the estimated error by substituting a part or whole of the accumulated value of the estimated code amount generated by the code amount estimating means with the real code amount after arithmetic encoding, and a rate control means which sets a target code amount being a target value of the code amount output from the motion picture compressing means, based on the accumulated value of the estimated code amount substituted by the estimated code amount substituting means, wherein the motion picture compressing means compresses the motion picture based on the target code amount. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an image information encoding apparatus that encodes image information of a moving image, and more particularly to an image information encoding apparatus that outputs a quantized DCT (Discrete Cosine Transform) coefficient and the like after variable length encoding. The present invention also relates to an audio information encoding device.

  In the moving picture coding technique, the coding efficiency is improved. As a result, it is becoming possible to realize a smooth-moving videophone or to shoot a high-quality moving image on a mobile phone. With this advancement in coding technology, the H.264 / AVC, which is the international standard for the latest video compression coding technology, specifies that syntax such as DCT coefficients and motion vectors should be transmitted. As an element entropy encoding method, an encoding method with higher encoding efficiency is prepared instead of conversion by a simple table. That is, a variable-length coding method called CAVLC (Context-Adaptive Variable Length Coding) and an arithmetic coding method called CABAC (Context-Adaptive Binary Arithmetic Coding). Of these, arithmetic coding methods such as CABAC are said to be able to theoretically compress stationary signals to the limit.

  On the other hand, CABAC has the advantage of good coding efficiency as described above, but has the disadvantage that the amount of generated code is not known until the processing is completed.

  Real-time processing is difficult to achieve when hardware is mounted on CABAC that supports HD size (1920x1080) images. Therefore, a method of providing a buffer before the arithmetic coding and performing asynchronous processing can be considered. In this case, if a macroblock (hereinafter referred to as MB) having a large amount of data is concentrated at the end of one image being processed, the code amount for arithmetic coding is determined, but in the worst case, it corresponds to one image. There may be a delay.

  As described above, if the coded image rate control is performed after the code amount of arithmetic coding is determined, for example, the rate control can be performed for each picture, but the quantization step is changed for each macroblock. It is difficult to perform fine rate control such as. When the rate control for each picture is compared with the rate control for each macroblock, it is more likely that the rate fluctuation will be greater when the rate control is performed for each picture. There is a problem that there is a high possibility that image quality will be deteriorated.

Also, in conventional image coding rate control such as MPEG-2, the quantization step is determined based on the difference between the target code amount and the code amount output in real time (for example, see Patent Document 2). ). However, with H.264 that employs CABAC (arithmetic coding), it is difficult to obtain the amount of code generated by CABAC coding in real time, causing a delay, and rate control fails due to the worst bit over. For this reason, a method for ensuring real-time encoding by monitoring and limiting the amount of input data to CABAC in advance has been disclosed (see, for example, Patent Document 1).
JP 2004-135251 A JP 2004-357079 A

  However, in the above conventional example, it is necessary to limit the amount of data input to CABAC. For this reason, processing such as coarse quantization and re-encoding is indispensable, and there is a problem of causing rapid image quality degradation.

  In view of the above problems, an image encoding method in which rate control is performed based on an estimated generated code amount can be considered. In this method, since the estimated code amount is used, it is not necessary to wait for completion of the arithmetic encoding process. In addition, as described above, there is no need to limit the amount of data input to CABAC, so image quality does not deteriorate.

  However, since rate control is performed based on the estimated code amount, an error between the actual code amount and the estimated code amount is generated and accumulated. Therefore, there is a possibility that a larger code amount than the actually assignable code amount may be allocated, which causes a rate breakdown in the worst case. Also, if a smaller code amount than the actually assignable code amount is assigned, the assumed image quality may not be maintained and the image quality may deteriorate.

  Accordingly, an object of the present invention is to provide an image information encoding apparatus that minimizes the influence of an estimation error even when rate control is performed based on an estimated code amount.

  In order to solve the above-described problem, an information compression apparatus according to the present invention is an information compression apparatus including a data conversion unit and a lossless compression unit, and the lossless compression is performed based on data before lossless compression by the lossless compression unit. A code amount estimation unit that estimates an actual code amount after lossless compression by the unit and generates an estimated code amount; and an estimated code that accumulates the estimated code amount generated by the code amount estimation unit and generates a cumulative estimated code amount Amount accumulating means, actual code amount accumulating means for accumulating the actual code amount after the lossless compression to generate an accumulated actual code amount, and the accumulated estimated code amount generated by the estimated code amount accumulating means as the actual code amount Accumulated estimated code amount correcting means for correcting with the accumulated actual code amount generated by the accumulating means, and accumulating the undetermined estimated code amount for the data before the lossless compression for which the actual code amount after the lossless compression is not fixed An undetermined estimated code amount accumulating unit for generating an accumulated undetermined estimated code amount, an accumulated estimated code amount corrected by the estimated code amount correcting unit, and an accumulated undetermined estimation generated by the undetermined estimated code amount accumulating unit Rate control means for setting a target code amount that is a target value of the code amount output from the data conversion means based on the code amount, and the data conversion means converts the data based on the target code amount. By performing the conversion, the amount of code output from the data conversion means is increased or decreased. Thereby, even when the rate control is performed based on the estimated code amount, the estimated code amount is corrected with the actual code amount, and the influence of the estimation error can be minimized.

  The image information encoding apparatus according to the present invention is an image information encoding apparatus including a moving image compression unit and an arithmetic encoding unit, and uses data related to a moving image before arithmetic encoding by the arithmetic encoding unit. A code amount estimation unit that estimates an actual code amount after arithmetic coding by the arithmetic coding unit and generates an estimated code amount; and a part of a cumulative value of the estimated code amount generated by the code amount estimation unit Alternatively, the estimated code amount replacing means for reducing the estimation error by replacing the entire code amount with the actual code amount after arithmetic coding, and the moving image based on the accumulated value of the estimated code amount replaced by the estimated code amount replacing means. Rate control means for setting a target code amount that is a target value of the code amount output from the compression means, and the moving image compression means compresses the moving image based on the target code amount. To. Thereby, even when rate control is performed based on the estimated code amount, the estimated code amount is replaced with the actual code amount, and the influence of the estimation error can be minimized.

  Note that the present invention can be realized not only as such an image information encoding device, but also as a context adaptive binary arithmetic encoding device provided in such an image information encoding device. The present invention can be realized as an image information encoding method having steps as characteristic means included in a simple image information encoding apparatus, or as a program for causing a computer to execute these steps. Needless to say, such a program can be distributed via a recording medium such as a CD-ROM or a transmission medium such as the Internet.

  As described above, according to the present invention, even when rate control is performed based on the estimated code amount, the estimated code amount is corrected with the actual code amount, and the influence of the estimation error is minimized. be able to. As a result, the code amount after arithmetic coding necessary for the rate control can be estimated before the CABAC processing is completed, and it is possible to realize the rate control with high accuracy without delay.

(Embodiment)
FIG. 1 is a block diagram showing a configuration of an image information encoding apparatus according to the present embodiment. The image information encoding apparatus according to the present embodiment predicts the code amount generated in the CABAC processing unit based on one of the residual signal, DCT coefficient, quantization coefficient, and binary data for each macroblock, An image information encoding device that controls an encoding rate by adjusting a quantization step in a quantization unit based on the prediction result, and includes a prediction processing unit 100, a DCT conversion unit 101, a quantization unit 102, A CABAC processing unit 103, a buffer 104, a code amount estimation unit 105, and a rate control determination unit 106 are provided. In the figure, the residual signal from the prediction processing unit 100, the DCT coefficient from the DCT conversion unit 101, the quantization coefficient from the quantization unit 102, and the binary data from the CABAC processing unit 103 are all encoded. Although it is input to the quantity estimation unit 105, any of these may actually be used. In the following, a case where the code amount is predicted based on the quantization coefficient will be described first. The prediction processing unit 100 detects a picture type and a slice type of a macroblock (hereinafter referred to as “MB”) that is an image area to be subjected to image information encoding processing, and detects the detected picture type (I, P, B) and the slice type (I, P, B) are output to the code amount estimation unit 105. Also, the prediction processing unit 100 encodes a residual signal that is a difference value between a predicted image generated by in-plane prediction or inter-plane prediction and an original image, and an encoding condition such as a prediction method (hereinafter referred to as “prediction information”). ) Is output. An example of the prediction information is the MB encoding mode, the macroblock type, the encoding block pattern, and the quantization parameter difference from the previous MB. DCT transform section 101 outputs a coefficient obtained by performing discrete cosine transform on the residual signal that is the output of prediction processing section 100. The quantization unit 102 outputs a coefficient obtained by quantizing the DCT coefficient that is the output of the DCT conversion unit 101. The CABAC processing unit 103 binarizes the quantization coefficient and the prediction information that are output from the quantization unit, performs arithmetic coding, and outputs the bit stream to the buffer 104. In the code amount estimation unit 105, the quantization coefficient that is the output of the quantization unit and the prediction information that is the output of the prediction processing unit 100 are created based on the actual measurement values of the generated code amounts of the respective arithmetic coding results. The code amount output from the CABAC processing unit 103 is estimated by converting into the estimated code amount using the “table”. That is, the table for estimating the code amount of the prediction information is created based on the actual measurement value obtained by encoding the prediction information separately from the table for estimating the code amount of the quantization coefficient. The code amount estimation unit 105 holds a plurality of conversion tables indicating the estimated code amount corresponding to the quantization coefficient in accordance with the picture type or slice type of the encoding target MB. The code amount estimation unit 105 selects and refers to the optimum conversion table stored therein according to the picture type or slice type input from the prediction processing unit 100. The rate control determination unit 106 determines subsequent encoding conditions based on the estimated code amount output from the code amount estimation unit 105. The encoding condition for performing the rate control is, for example, the value of the quantization step in the quantization unit 102. The rate control determination unit 106 returns the determined encoding condition to the quantization unit 102 and uses it as the subsequent MB encoding condition.

  FIG. 2 is a block diagram showing a more detailed configuration of the CABAC processing unit 103 shown in FIG. The CABAC processing unit 103 is a processing unit that binarizes the quantization coefficient input from the quantization unit 102 and arithmetically encodes binary data obtained by binarization, and includes a binarization processing unit 200, An arithmetic coding unit 201 and a context calculation unit 203 are provided.

  The binarization processing unit 200 converts the quantization coefficient, which is the output of the quantization unit 102, into binary data according to a predetermined conversion law. In accordance with the input of the binary data, the arithmetic encoding unit 201 performs arithmetic encoding using the occurrence probability of 0/1 output from the CONTEXT calculation unit 203 and outputs the result to the buffer 104. The CONTEXT calculation unit 203 stores therein a table indicating the occurrence probability of 0/1, and outputs the occurrence probability of 0/1 of the next bit to the arithmetic encoding unit 201 with reference to the table. If the bit value is not as predicted, the CONTEXT calculation unit 203 updates the occurrence probability of 0/1 in the internal table.

  FIG. 3 is a graph showing an example of a rate control method performed by the rate control determination unit 106 shown in FIG. The rate control determination unit 106 accumulates, for example, (1) a constant target generated code amount for each MB (for example, 3 Mbit / MB) and (2) an estimated code amount for each MB in the same picture up to the current MB. The upper and lower thresholds of the difference between the accumulated estimated code amount obtained in this way and the accumulated target generated code amount up to the current MB, and (3) the accumulated estimated code amount up to the current MB is compared with the accumulated target generated code amount When the threshold value is increased or decreased beyond the threshold value, it is stored how much the quantization step is increased or decreased. Then, the estimated code amount obtained from the code amount estimating unit 105 is totaled every 1 MB, and when the total estimated code amount increases from the accumulated target generated code amount beyond the threshold, the increase amount of the quantization step is calculated. Output to 102. As a result, the quantization unit 102 quantizes the DCT coefficient in a larger value quantization step, so that the quantization coefficient becomes a smaller value and the generated code amount is reduced.

  FIG. 4 is a flowchart showing a rate control procedure of each processing unit in the entire image information encoding apparatus. First, the quantization unit 102 sequentially performs quantization using the first MB of the picture as the target MB. The same number of luminance and color difference DCT coefficients as pixels are calculated for each MB, and the quantization unit 102 divides this DCT coefficient by the quantization step and outputs the quantization coefficient (S901). The code amount estimation unit 105 refers to the conversion table held inside using the quantization coefficient output from the quantization unit 102 as a key, and acquires the estimated code amount corresponding to the quantization coefficient (S902). . The code amount estimation unit 105 adds the acquired estimated code amount to, for example, a register called (code amount), and determines whether or not the estimated code amounts of all quantization coefficients of the target MB have been acquired ( S904). If the estimated code amounts of all the quantization coefficients of the target MB have not been acquired yet, the process returns to step S902 to acquire new quantization coefficients. When the estimated code amount is acquired for all quantization coefficients of the target MB, the estimated code amount for each quantization coefficient accumulated in the (code amount) register is added to, for example, the (accumulated code amount) register. (S905). Thereby, the cumulative estimated code amount up to the target MB is obtained. The rate control determination unit 106 calculates the target code amount up to the MB from the target code amount held for each MB. For example, if the target code amount is 3000 bits / MB, if the quantization of 5 MBs in the picture is completed, the calculation of 3000 × 5 = 15000 is performed and the target up to the target MB is calculated. The code amount is calculated. Note that the rate control determination unit 106 does not necessarily store a constant value, such as 3000 bit / MB as the target code amount, and may hold a non-constant target code amount as a table. For example, the 0th MB may be 2500 bits, the first MB may be a total of 5500 bits,..., The kth MB may be a total of 32000 bits, and so on.

  Next, the rate control determination unit 106 determines that the cumulative code amount up to the target MB including the target MB from the (accumulated code amount) calculated in step S905 and the target code amount calculated or obtained from the table is the target code amount. It is determined whether or not the difference exceeds the predetermined upper limit threshold (S906). When the accumulated code amount exceeds the target code amount and the difference exceeds the upper limit threshold, the rate control determination unit 106 increases the quantization step value (S907), and Encoding is started (S901). If the accumulated code amount exceeds the target code amount and the difference is other than the upper limit threshold, the rate control determination unit 106 further determines that the accumulated code amount is equal to or less than the target code amount and the difference It is determined whether or not exceeds a predetermined lower threshold (S908). If the accumulated code amount is less than or equal to the target code amount and the difference exceeds a predetermined lower limit threshold, the value of the quantization step is decreased (S909), and encoding of the next MB is started. (S901). In other cases, the rate control determination unit 106 starts quantization of the next MB without changing the value of the quantization step (S901). Note that, in other cases, the accumulated code amount is equal to or less than the target code amount, and the difference between the accumulated code amount and the target code amount is equal to or less than the lower limit threshold, or the accumulated code amount is equal to the target code amount. This is a case where the amount is exceeded and is not more than the upper limit threshold. In this way, by adjusting the quantization step for each MB based on the estimated code amount, the generated code amount by arithmetic coding can be made closer to the target code amount with higher accuracy. Here, the MB has been described as an encoding unit, but may be an MB line unit, a slice unit, a picture unit, or the like.

  Next, a code amount estimation method of the code amount estimation unit 105 will be described.

  FIG. 5 is a conversion table showing the estimated code amount generated in the CABAC processing unit corresponding to each quantization coefficient value. The code amount estimation unit 105 estimates the code amount after CABAC processing based on the quantization coefficient output from the quantization unit 102. For the estimation, a conversion table as shown in FIG. 5 is used. The code amount estimation unit 105 acquires an estimated code amount corresponding to the input quantization coefficient value with reference to the conversion table. According to the conversion table of FIG. 5, for example, when the quantization coefficient value is 1, the estimated code amount is 1.2 bits, and when the quantization coefficient value is 2, the estimated code amount is 1.4 bits. The reason why the estimated code amount is not an integer but a real number is that it is created from an average value of actually measured code amounts. In this way, by referring to the conversion table, the estimated code amount is uniquely determined from each quantization coefficient value. The code amount estimation unit 105 stores such a conversion table in an internal memory, and refers to the conversion table based on the input quantization coefficient value. The conversion table may be provided outside the code amount estimation unit 104. Further, the conversion table does not need to be one, but may be selected from a plurality of tables and referred to. For example, the table to be referenced may be switched between when the image to be encoded is predicted in-plane and when the inter-frame prediction is performed, and the estimated code amount corresponding to the quantization step may be acquired. Furthermore, the conversion table may be switched depending on the MB encoding type. In addition, switching may be performed for each type of picture type, slice type, or prediction information. Furthermore, the conversion table may be a static table using a fixed value. The conversion table may be a dynamic table that can be updated.

  FIG. 6 is a diagram showing a more specific example of the conversion table shown in FIG. In the figure, i in abs_level [i] indicates a coefficient number from 1 to 384. FIG. 6A is a table showing estimated bit values when the quantized values take values from 1 to 14. In this table, the quantization value and the estimated code amount have a one-to-one correspondence. In this table, it is possible to obtain an estimated code amount of quantization values 1 to 14. For example, when the quantized coefficient abs_level is 1, the estimated code amount is 1.56 bits, and when the quantized coefficient abs_level is 8, the estimated code amount is 5.20 bits. FIG. 6B is a table showing an estimated bit value when the quantized value is 15 or more. In this table, quantized values of 15 to 60, 61 to 250, and 251 or more are respectively set to the same estimated code amount. By referring to this table, the estimated bit value when the quantized value takes a value of 15 or more can be uniquely determined. For example, when the quantized coefficient abs_level is 125, the estimated code amount is 0.82 bits, and when the quantized coefficient abs_level is 61, the estimated code amount is 0.82 bits. As described above, the code amount estimation unit 105 can obtain the estimated code amount corresponding to all values of 384 quantization coefficients (luminance Y256, color difference Cb64, color difference Cr64) by referring to this table.

  The conversion tables shown in FIGS. 6A and 6B are used not only when the estimated code amount is obtained based on the quantization coefficient but also when the estimated code amount is obtained based on binary data. Can do. The reason is that the binary data is uniquely determined from the quantization coefficient value. For example, when the quantization coefficient Coeff_abs_level_minus1 = 10, binary data is uniquely obtained as binary data = 1111 1111 110.

  Here, only the quantized coefficients are mentioned, but prediction information that is subject to CABAC processing other than the quantized coefficients may be the target of the code amount estimating unit 105.

  Further, the data used for the estimation of the code amount estimation unit 105 is not only the quantization coefficient but also the binary data value or the binary data amount generated by the binarization processing unit 200, the DCT coefficient, the output of the prediction processing unit 100 It is also possible to generate a conversion table as a residual signal that is the difference between the predicted image and the input image.

  FIG. 7 is a diagram illustrating an example of a conversion table indicating the estimated code amount by arithmetic coding corresponding to the DCT coefficient value. When the code amount is estimated based on the DCT coefficient, the estimated code amount is changed by the subsequent quantization. Therefore, the quantization variable QP (value defined by H.264 is a value from 0 to 51. A conversion table is created for each (obtain), and each time the quantization variable QP is determined for each MB, the conversion table is switched and used. FIG. 7A is a diagram illustrating an example of a conversion table created for each quantization variable QP. FIG. 7B is a diagram illustrating a flow of data and a procedure when rate control is performed by estimating a code amount using a DCT coefficient. The code amount estimation unit 105 receives the currently quantized MB quantization variable QP from the quantization unit 102, and selects the conversion table shown in FIG. 7A based on the received quantization variable QP. . The code amount estimation unit 105 estimates the arithmetic code amount of each DCT coefficient by referring to the table selected for each DCT coefficient, and aggregates the estimated code amount for all the DCT coefficients in the current MB, thereby estimating the current MB. The code amount is calculated. As shown in the graph of FIG. 3, the code amount estimation unit 105 calculates the accumulated code amount from the first MB to the current MB and outputs it to the rate control determination unit 106. The rate control determination unit 106 compares the accumulated code amount input from the code amount estimation unit 105 with the accumulated target generated code amount from the first MB to the current MB, and determines the quantization variable QP of the next MB.

  In the above embodiment, the method for estimating the code amount from the DCT coefficient before quantization has been described with reference to FIG. 7, but the method for estimating the code amount from the residual signal before performing frequency conversion will be described below. To do. FIG. 8 is a table for estimating the residual amount signal of each pixel in the MB and the code amount of the MB from the residual signal. FIG. 8A is a diagram illustrating an example of the absolute difference value sum (SAD) of each MB in a picture and its residual signal. Note that the absolute difference value sum or the sum of squared differences (SSD) is generally used for the prediction error. Here, the absolute difference value sum will be described. FIG. 8B is a table showing the estimated code amount of the MB corresponding to the absolute difference value sum (SAD) for each MB. Since the residual signal is also quantized at a later stage, a conversion table is created for each quantization variable QP. Further, the residual signal represents a signal component in space, whereas the DCT coefficient represents those frequency components. Therefore, the prediction error value for each pixel does not correspond one-to-one with the quantization coefficient or binary data. For this reason, the code amount estimation unit 105 totals the absolute difference value sum for each MB, refers to the conversion table in FIG. 8B using the total sum of absolute difference values, and obtains the estimated code amount for the MB. .

  Next, a method for creating a dynamic conversion table will be described.

  The first method is described. FIG. 9 is a diagram illustrating a method of updating the conversion table with reference to the code amount actually generated by arithmetic coding. In this case, the code amount estimation unit 105 includes, for example, a conversion table generation unit.

  The code amount estimation unit 105 stores the quantization coefficient output from the quantization unit 102 in the internal memory. Next, the conversion table generation unit measures the code amount from the code encoded by the CABAC processing unit 103 and stores it in the internal memory. Specifically, the code amount is measured by counting whether the code output from the CABAC processing unit 103 is 0, 1 or undetermined. When the code to be output is undecided, the conversion table generation unit measures the code amount using the value of the counter Bits_Outstanding that is provided in the CABAC processing unit 103 and counts which digit code value has not been determined. .

  Next, the conversion table generation unit temporarily enters the quantization value in the internal memory and the actually measured code amount in the existing conversion table. The conversion table generation unit updates the existing conversion table with a value obtained by averaging the code amounts corresponding to the same quantized value entered in the conversion table. The update of the conversion table may be performed for every quantization coefficient or only a part of the quantization coefficients.

  The second method is described.

  Since the CABAC processing unit 103 shown in FIG. 2 performs section division using the “predetermined occurrence probability of 0/1” output from the CONTEXT calculation unit 203 according to the input, the estimated code amount Changes according to the occurrence probability of 0/1. The code amount estimation unit 105 measures the actual 0/1 occurrence probability of the 0/1 series of binary data and compares it with a predetermined 0/1 occurrence probability. The code amount estimation unit 105 updates the conversion table in accordance with the difference between the actual occurrence probability and a predetermined occurrence probability. For example, when the measured value of the occurrence probability of the dominant symbol is significantly lower than the CABAC table, the code amount estimation unit 105 (specifically, the measured value of the occurrence probability of the dominant symbol is previously compared with the CABAC table. When the value falls below a predetermined threshold value), the conversion table is updated. For example, when the two occurrence probabilities are equal, the table is not updated. When there is a difference between the two occurrence probabilities, the compression rate by arithmetic coding changes, so the value of the estimated code amount in the conversion table is changed. As for the change amount, the difference between the two occurrence probabilities and the difference between the compression rates can be stored in a memory, and the change amount can be determined using the difference. It is also possible to update the table by estimating the occurrence probability based on the 0/1 pattern. For example, when 1 continues from a certain number, or when 0 continues, the actual occurrence probability varies from the previous occurrence probability calculated by the CONTEXT calculation unit 203. When such a pattern is detected, The conversion table may be updated. That is, the code amount estimation unit 105 analyzes the occurrence pattern of binary data 0 and 1, and when a certain number of 0s or 1s continues, the compression rate by arithmetic coding becomes high. The conversion table is updated so as to reduce the estimated code amount for the same value of any of the quantization coefficient, binary data, DCT coefficient, and residual signal. When a certain number of 0s and 1s alternately appear, the estimated code amount of the conversion table is updated so as to increase with respect to the same value.

  The conversion table may be updated when the predetermined occurrence probability of 0/1 varies, or the conversion table when the actual occurrence probability and the predetermined occurrence probability vary. May be updated, or a combination thereof may be used.

  Further, the predetermined occurrence probability of 0/1 corresponds to pStateIdx in the entropy coding CABAC of the video compression coding standard H.264, and the conversion table may be updated accordingly. . Also, instead of updating the conversion table, two types of conversion tables for high compression rate and low compression rate are stored from the beginning, and either one is selected and used according to the value of pStateIdx It is good. That is, at the start of MB arithmetic coding, the CONTEXT calculation unit 203 calculates the occurrence probability of binary data. This occurrence probability is given by the occurrence probability table number pStateIdx (0-63). When the pStateIdx is larger than a predetermined threshold, the code amount estimation unit 105 determines that the compression rate is high and determines a high compression rate conversion table (that is, quantization coefficient, binary data, DCT coefficient, and remaining). For the same value of any of the difference signals, the estimated code amount is reduced as compared with the standard conversion table). Note that the standard conversion table refers to a conversion table when the estimated code amount is referred to using only one conversion table, instead of switching between the high compression rate and the low compression rate. . Conversely, when the value of pStateIdx is smaller than the threshold value, it is determined that the compression rate is low, and the low compression rate conversion table (that is, the quantization coefficient, binary data, DCT coefficient, and residual signal) For any one of the same values, the estimated code amount is increased over the standard conversion table). Specifically, when the pStateIdx value is 50 and MB arithmetic coding is started, when the pStateIdx value becomes 40 in the middle, the code amount estimation unit 105 uses this as a trigger to convert to a low compression rate conversion table. The code amount may be estimated by switching to.

  The timing for updating the conversion table or switching between the conversion table for high compression ratio and the conversion table for low compression ratio may be every time pStateIdx is updated, or every fixed number of times pStateIdx is updated. (For example, every 5th update of pStateIdx). Further, the conversion table may be updated or the high compression rate conversion table and the low compression rate conversion table may be switched in accordance with the moving average of pStateIdx. For example, every time pStateIdx is updated three times, an average value for three times is obtained, and when the obtained average value is larger than a predetermined threshold, a conversion table for high compression ratio is used, or the estimated code amount is updated to a small value To do. On the contrary, when the obtained average value is smaller than a predetermined threshold value, a low compression rate conversion table may be used, or the estimated code amount may be updated to a large value. Note that the method of obtaining the moving average need not be every three updates of pStateIdx, and may be any number of times.

  Furthermore, although the conversion table is used here, the estimated code amount may be determined by one or a plurality of functions. In other words, the code amount estimation unit 105 uses the one or more functions representing the relationship between one of the quantized coefficient, binary data, DCT coefficient, and residual signal and the estimated code amount to calculate the estimated code. The amount may be calculated.

  In the above embodiment, the coding condition is described as the quantization step. However, for example, the type of MB prediction processing in the prediction processing unit 100 may be set to the skip mode. The skip mode is a mode in which only a flag meaning to copy image data at a predetermined position is sent to the MB without calculating a residual signal based on prediction. Since this skip mode encodes only 1-bit data per 1 MB, it is effective in reducing the code amount when the accumulated estimated code amount is very large. In this case, the rate control determination unit 106 feeds back the encoding condition determined to be the skip mode to the prediction processing unit 100.

  Conversely, if the cumulative estimated code amount estimated by the code amount estimation unit 105 is much smaller than the target generated code amount, the rate control determination unit 106 sets the subsequent MB encoding conditions. I_PCM data may be used. The I_PCM data is raw data of each MB before motion prediction and in-plane prediction are performed, and includes a luminance signal and two color difference signals. In other words, setting the encoding condition to I_PCM data means that the raw data of the picture is output as it is without performing motion prediction, in-plane prediction, or encoding. In this way, when the accumulated estimated code amount is less than the target generated code amount, it is possible to output image data that does not impair image quality because the data amount is large but encoding is not performed.

  FIG. 10 is a block diagram showing a detailed configuration of a main part of the present invention. Specifically, detailed configurations of the moving image compression unit 400, the CABAC processing unit 103, the code amount estimation unit 105, the estimated code amount replacement unit 35, and the rate control determination unit 106 are illustrated. The moving image compression unit 400 is an example of the data conversion unit according to claim 1 or the moving image compression unit according to claim 4. The CABAC processing unit 103 exemplifies a configuration in which the arithmetic coding unit 12 is included, but a variable length coding unit 02 (not shown) may be employed instead of the arithmetic coding unit 12. . The arithmetic coding unit 12 and the variable length coding unit 02 are both examples of the lossless compression means according to claim 1.

  The bit amount of the encoded sequence subjected to variable length encoding by the CABAC processing unit 103 is input to the cumulative code amount calculation unit 30. The cumulative code amount calculation unit 30 cumulatively adds the bit amount of the encoded sequence. At this time, it is recorded simultaneously from which encoding unit to which encoding unit the bit amount accumulated in the storage area (not shown) is. The coding unit here refers to a picture unit, a slice unit, a macroblock line unit, a macroblock unit, or the like. The bit amount may be managed and recorded together with an identifier for identifying a coding unit. A picture number indicating the coding order can be used as a picture identifier, and a macroblock address can be used as a macroblock identifier.

  In addition, since the variable length coding unit 02 operates asynchronously with respect to the input in circuit implementation, the variable length coding unit 30 notifies the cumulative code amount calculation unit 30 that the predetermined coding unit variable length coding has been completed. Input the bit amount of the encoded stream. At this time, which coding unit is the code amount is notified at the same time.

  Similarly, the code amount estimated by the code amount estimation unit 105 (corresponding to the code amount estimation unit 105 in FIG. 1) is cumulatively added by the cumulative estimated code amount calculation unit 31. This cumulatively added value is called a cumulative code amount. At this time, it is recorded simultaneously from which encoding unit to which encoding unit the code amount accumulated in the storage area (not shown) is. Furthermore, as will be described later, it may be recorded here which encoding unit of variable-length encoding is completed. The bit amount may be managed and recorded together with an identifier for identifying a coding unit.

  Further, the cumulative target code amount calculation unit 32 cumulatively adds the target code amount that is the target value of the code amount. At this time, the target code amount accumulated in a storage area (not shown) is recorded simultaneously from which encoding unit to which encoding unit. The bit amount may be managed and recorded together with an identifier for identifying a coding unit.

  Next, the cumulative value of the estimated code amount calculated from the predetermined coding unit to the predetermined coding unit calculated by the cumulative estimated code amount calculation unit 31, and the predetermined encoding calculated by the cumulative target code amount calculation unit 32 A difference value (hereinafter, accumulated difference value) from the accumulated value of the target code amount from the unit to a predetermined coding unit is output to the target code amount calculation unit 33. Here, for convenience, the accumulated difference value is a value obtained by subtracting the accumulated value of the estimated code amount from the accumulated value of the target code amount. In the target code amount calculation unit 33, cumulative difference values are cumulatively added. This cumulatively added value is called a surplus code amount. The target code amount calculation unit 33 determines the target code amount of the next coding unit based on the surplus code amount and the accumulated difference value. As an example of the determination method, there is a method in which the target code amount calculation unit 33 calculates the average code amount obtained from the bit rate set at the time of encoding.

  When the accumulated difference value is positive, that is, when there is a margin in the code amount, the target code amount is allocated more than the average target code amount. Conversely, when negative, the target code amount is assigned less than the average code amount. Further, it is possible to control the allocation amount according to the image feature amount of the input image calculated by the image feature amount calculation unit (not shown), for example, the adjacent pixel difference indicating the flatness of the image. Here, the adjacent pixel difference is a pixel in a specific coding unit, and indicates a sum obtained by adding the absolute differences of pixel values of adjacent pixels. As a result, the absolute value sum or the square sum of the adjacent pixel differences when the surplus code amount, which is a case where there is a sufficient code amount that can be allocated, is larger than a predetermined threshold and when the flatness of the image in the coding region is high If it is smaller than the predetermined threshold, setting a target code amount that is larger than the average code amount makes it possible to preserve a subtle change in the pixel value of the flat portion and maintain the fineness of the image.

  Next, the quantization parameter calculation unit 34 included in the moving image compression unit 400 determines a quantization step that is one piece of encoding control information from the target code amount determined by the target code amount calculation unit 33. Here, the quantization step is determined, but what is determined may be a parameter (encoding control information) for increasing or decreasing the code amount. For example, it may be a quantization matrix that changes the quantization step for quantization according to the coefficient position. In addition, when the surplus code amount is negative, in order to limit the code amount, not only the quantization step may be increased, but also control may be performed so that inter prediction is easily selected. For example, in the determination process between the inter prediction and the intra prediction of the encoding means, an offset is set in the evaluation function so that the intra prediction is easily selected.

  In addition, as an example of a method for calculating a quantization step, a target value can be calculated based on a ratio between a quantization step of a previously encoded coding unit (past quantization step) and a code amount actually generated (past code amount). There is a method for determining a quantization step for a code amount.

Specifically, the quantization step is determined so as to satisfy the following equation.
Past code amount: Past quantization step = Target code amount: Quantization step

  Based on the calculated quantization step, the quantization unit 102 performs quantization.

  In addition to the above, the estimated code amount replacing unit 35 replaces a part or all of the estimated code amount calculated by the accumulated estimated code amount calculating unit 31 with the accumulated code amount calculated by the accumulated code amount calculating unit 30. After the coding unit for which the CABAC coding has been completed is updated, the accumulated value of the estimated code amount from the coding unit for which the previous replacement has been completed to the coding unit for which the current CABAC coding has been completed is actually accumulated. Replace with code amount. It is simultaneously recorded up to which encoding unit the replacement is completed in a storage area (not shown). By replacing a part of the accumulated value of the estimated code amount with the actual accumulated code amount, propagation of the estimation error is minimized. The replaced cumulative estimated code amount is output to rate control determination section 106.

  The accumulated estimated code amount calculating unit 31 and the estimated code amount replacing unit 35 shown here correspond to the undetermined estimated code amount accumulating unit, the estimated code amount accumulating unit, and the accumulated estimated code amount correcting unit.

  FIG. 11 shows a specific example of replacement.

  Each block shown in the figure corresponds to one register in a storage area (not shown). The type of information stored (code amount, estimated code amount and corresponding picture number) is described in each block. In the figure, it is shown that CABAC encoding has been completed for picture numbers 0 to 2. It also indicates that the estimated code amount calculation processing has been completed for picture numbers 0 to 4. Here, a replacement process is performed every time CABAC encoding is completed for three pictures.

  The picture position where the replacement before “replacement” is completed is the position of picture number 0 in the initial state. Here, it indicates that no replacement has occurred. Next, the picture position where the replacement of “after replacement” is completed is updated to picture number 3. In the above manner, the replacement process is performed every time CABAC encoding is completed for three pictures. The target code amount calculation unit 33, through a storage area (not shown), to which picture the encoding is completed (encoding completion picture position), to which picture the estimated code amount is completed (estimation completion picture position) ) And to which picture the replacement is completed (replacement completed picture position). Although various completed picture positions are managed via the storage area, it goes without saying that the same effect can be obtained even if each block directly adopts a configuration that notifies the target code amount calculation unit 33 of the completion notification. Yes.

  As described above, according to the present invention, even when rate control is performed based on the estimated code amount, the estimated code amount is corrected with the actual code amount, and the influence of the estimation error is minimized. be able to. As a result, the code amount after arithmetic coding necessary for the rate control can be estimated before the CABAC processing is completed, and it is possible to realize the rate control with high accuracy without delay.

  Here, a method of replacing the estimated code amount with the code amount has been described, but the present invention is not limited to such simple replacement. That is, the estimated code amount may be corrected by a method such as multiplying the estimated code amount or the code amount by a coefficient corresponding to the ratio between the estimated code amount and the code amount.

  Needless to say, if the encoding unit in which the CABAC encoding is completed and the estimated encoding unit are the same, the code amount may be directly output to the rate control without replacement.

  Each functional block in FIGS. 1 and 10 is typically realized as an LSI which is an integrated circuit. These may be individually made into one chip, or may be made into one chip so as to include a part or all of them (for example, functional blocks other than the memory may be made into one chip).

  The name used here is LSI, but it may also be called IC, system LSI, super LSI, or ultra LSI depending on the degree of integration.

  Further, the method of circuit integration is not limited to LSI's, and implementation using dedicated circuitry or general purpose processors is also possible. An FPGA (Field Programmable Gate Array) that can be programmed after manufacturing the LSI or a reconfigurable processor that can reconfigure the connection and setting of the circuit cells inside the LSI may be used.

  Further, if integrated circuit technology comes out to replace LSI's as a result of the advancement of semiconductor technology or a derivative other technology, it is naturally also possible to carry out function block integration using this technology. Biotechnology can be applied.

  In addition, among the functional blocks, only the means for storing the data to be encoded or decoded may be configured separately instead of being integrated into one chip.

  In addition, although image coding has been described here as an example, it goes without saying that the present invention can be applied to audio signal coding and other signal coding.

  The image information encoding apparatus according to the present invention is useful as an image information encoding apparatus provided in a personal computer, a PDA, a digital broadcast station, a mobile phone, and the like having a communication function (see FIG. 12).

It is a block diagram which shows the structure of this Embodiment. It is a block diagram which shows the more detailed structure of the CABAC process part shown in FIG. It is a graph which shows an example of the rate control method by the rate control determination part shown in FIG. It is a flowchart which shows the procedure of the rate control of each process part in the image information coding apparatus of this Embodiment. It is a conversion table which shows the estimated code amount which generate | occur | produces in a CABAC process part corresponding to each quantization coefficient value. FIG. 6 is a diagram showing a more specific example of the conversion table shown in FIG. 5. (A) is a table which shows the estimated bit value when a quantization value takes the value of 1-14. (B) is a table showing an estimated bit value when the quantized value is 15 or more. It is a figure which shows an example of the conversion table which shows the estimated code amount by arithmetic coding corresponding to the value of a DCT coefficient. (A) is a figure which shows an example of the conversion table produced for every quantization variable QP. (B) is a figure which shows the flow of the data and procedure in the case of estimating code amount using a DCT coefficient and performing rate control. It is a table for estimating a residual signal of each pixel in the MB and a code amount of the MB from the residual signal. (A) is a figure which shows an example of the absolute difference value sum of each MB in a picture, and its residual signal. (B) is a table showing the estimated code amount of the MB corresponding to the sum of absolute difference values for each MB. It is a figure explaining the method of updating a conversion table with reference to the code amount actually generated by arithmetic coding. It is a block diagram which shows the detailed structure of a rate control determination part. It is a figure explaining replacement of the code amount of a target code amount calculation part. It is a general-view figure of an example of the image information coding apparatus which concerns on this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Prediction processing part 101 DCT conversion part 102 Quantization part 103 CABAC processing part 104 Buffer 105 Code amount estimation part 106 Rate control determination part 200 Binary processing part 201 Arithmetic coding part 203 CONTEXT calculation part 400 Moving picture compression part 11 2 Numeralization unit 12 Arithmetic coding unit 30 Cumulative code amount calculation unit 31 Cumulative estimated code amount calculation unit 32 Cumulative target code amount calculation unit 33 Target code amount calculation unit 34 Quantization parameter calculation unit 35 Estimated code amount replacement unit

Claims (6)

An information compression apparatus comprising data conversion means and lossless compression means,
A code amount estimation unit that estimates an actual code amount after lossless compression by the lossless compression unit and generates an estimated code amount based on data before the lossless compression unit;
An estimated code amount accumulating means for accumulating the estimated code amount generated by the code amount estimating means to generate an accumulated estimated code amount;
An actual code amount accumulating means for accumulating the actual code amount after the lossless compression to generate an accumulated actual code amount;
A cumulative estimated code amount correcting unit that corrects the cumulative estimated code amount generated by the estimated code amount accumulating unit with the accumulated actual code amount generated by the actual code amount accumulating unit;
Undetermined estimated code amount accumulating means for accumulating the undetermined estimated code amount for the data before the lossless compression in which the actual code amount after the lossless compression is not determined,
Based on the accumulated estimated code amount corrected by the estimated code amount correcting unit and the accumulated undetermined estimated code amount generated by the undetermined estimated code amount accumulating unit, a code amount target output from the data conversion unit Rate control means for setting a target code amount that is a value,
The data compression unit increases or decreases a code amount output from the data conversion unit by converting data based on the target code amount. The information compression apparatus according to claim 1, wherein the lossless compression unit performs variable-length encoding of data. The information compression apparatus according to claim 1, wherein the lossless compression unit arithmetically encodes data. An image information encoding device comprising a moving image compression means and an arithmetic encoding means,
Code amount estimation means for estimating an actual code amount after arithmetic coding by the arithmetic coding means and generating an estimated code amount using data relating to a moving image before arithmetic coding by the arithmetic coding means;
Estimated code amount replacement means for reducing the estimation error by replacing part or all of the accumulated value of the estimated code amount generated by the code amount estimation means with the actual code amount after the arithmetic coding;
A rate control unit that sets a target code amount that is a target value of a code amount output from the moving image compression unit based on a cumulative value of the estimated code amount replaced by the estimated code amount replacement unit;
The moving image compression unit compresses a moving image based on the target code amount. An image information encoding method including a moving image compression step and an arithmetic encoding step,
A code amount estimation step for generating an estimated code amount by estimating an actual code amount after arithmetic coding in the arithmetic coding step using data on a moving image before arithmetic coding in the arithmetic coding step;
An estimated code amount replacement step for reducing an estimation error by replacing a part or all of the accumulated value of the estimated code amount generated in the code amount estimation step with the actual code amount after the arithmetic coding;
A rate control step of setting a target code amount that is a target value of the code amount output in the moving image compression step based on the accumulated value of the estimated code amount replaced in the estimated code amount replacement step,
An image information encoding method, wherein the moving image is compressed based on the target code amount in the moving image compression step. An integrated circuit comprising moving image compression means and arithmetic coding means,
Code amount estimation means for estimating an actual code amount after arithmetic coding by the arithmetic coding means and generating an estimated code amount using data relating to a moving image before arithmetic coding by the arithmetic coding means;
Estimated code amount replacement means for reducing the estimation error by replacing part or all of the accumulated value of the estimated code amount generated by the code amount estimation means with the actual code amount after the arithmetic coding;
A rate control unit that sets a target code amount that is a target value of a code amount output from the moving image compression unit based on a cumulative value of the estimated code amount replaced by the estimated code amount replacement unit;
The integrated circuit, wherein the moving image compression means compresses a moving image based on the target code amount.

Images