JP2007318617A - Image encoder and image encoding program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To encode original image data after an encoded picture without preliminarily reading it while substantially holding image quality in quantization control for determining quantization steps. <P>SOLUTION: The image encoder has: a picture judgment part which detects whether a picture to be coded is any of I, P, B pictures; and a target code amount determination part which determines a target code amount, based on the generated code amount of the I picture encoded in the past when it is judged that the picture is the I picture by the picture judgment part and determines the target code amount of the encoded picture from predicted error information of the encoded picture when the picture is judged as the P or B picture. Then, the image encoder has an activity operation part which detects complexity of design of a block in a screen and a quantization control part which determines a quantization parameter in the picture from the determined target code amount and the activity information. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an image encoding device and an image encoding program for efficiently encoding image data.

  Conventionally, as an image encoding technique for efficiently encoding image data, there is an encoding method represented by MPEG (Moving Picture Experts Group). FIG. 1 shows a configuration of an image encoding device in a general MPEG standard.

  The quantization control unit shown in FIG. 1 controls the allocated data amount (target code amount) for each picture in order to transmit the encoded data with a predetermined data amount. For example, in the MPEG encoding method, there are three types of pictures (I, P, B pictures) having different properties, and therefore the target code amount is determined according to the type of each picture. That is, when an I or P picture is used as a reference picture for a P or B picture, a larger target code amount for the I or P picture is assigned than for a B picture.

  As the quantization control method as described above, for example, as in Patent Document 1, when determining the quantization parameter of an encoded picture, the original image data for 1 GOP including the encoded picture is pre-read and intra-coded. There is known a quantization control method that performs prediction processing and inter prediction error processing, and assigns a code amount to each picture according to the magnitude of prediction error obtained by intra prediction processing and inter prediction error processing. Also, as in Non-Patent Document 2, a quantization control method is known that determines the target code amount of a coded picture by paying attention to the correlation between the prediction error information of the coded picture and the generated code amount.

JP 2006-25077 A MPEG-2, Test Model 5 (TM5), Doc.ISO / IEC JTC1 / SC29 / WG11 / N0400, Test Model Editing Committee, Apr.1993 “H.264 Rate Control Method Considering Image Properties and Buffer Constraints”, IEICE Transactions, D-II, Vol.J88-D-II, 2005, p.1114-1125

  Hereinafter, the operation in the quantization control unit will be briefly described by taking the MPEG-2 Test Model 5 method used in the MPEG-2 standard as an example. Detailed operation description will be described later.

  In MPEG-2 Test Model 5, Step 1 performs code amount allocation to each picture, Step 2 determines quantization parameters for each block in the picture using a virtual buffer, and considers visual characteristics in the picture. Thus, the three-step process of step 3 for weighting the quantization parameter of each block is performed (see Non-Patent Document 1).

  By the way, in the MPEG-2 Test Model 5 method, the processing in step 2 and the processing in step 3 may operate in conflict. For example, when a large amount of code is generated in the encoded macroblock in step 2, the quantization step increases in order to suppress the generated code amount in the macroblock to be encoded next. At that time, when the activity of the macroblock to be encoded next is small, there is a problem that the quantization step is reduced and the control of step 2 and step 3 is offset.

  In addition, in the method of Patent Document 1, it is necessary to pre-read image data for 1 GOP including an encoded picture and to perform intra prediction or inter prediction error processing, so that processing time is significantly increased. In particular, when moving image data having a large size such as an HD (High Definition) image is encoded, there is a problem that it is difficult to perform encoding processing in real time.

  Also in the method of Non-Patent Document 2, in order to determine the target code amount of the encoded picture after prefetching the image data of the encoded picture and calculating the prediction error of all the macroblocks in the encoded picture, There is a problem that it is not suitable for pipeline processing in which encoding processing is performed for each macroblock.

  Accordingly, an object of the present invention is to provide an image coding that eliminates such a problem and can perform quantization control without prefetching the original image data after the coded picture while substantially maintaining the image quality. An apparatus and an encoding method are provided.

  In order to solve the above problems, an image encoding apparatus according to the present invention includes a picture determination unit that detects whether a picture to be encoded is an I, P, or B picture, and the picture determination unit includes a picture that is an I picture. If it is determined, the target code amount is determined based on the generated code amount of the I picture encoded in the past, and if it is determined as the P or B picture, the prediction error information of the encoded picture And a target code amount determination unit for determining a target code amount of the encoded picture. Then, an activity calculation unit for detecting the complexity of the picture of the block in the screen is provided, and a quantization control unit for determining a quantization parameter in the picture from the determined target code amount and activity information.

  In the image coding method of the present invention, when the correlation between a coded picture such as a scene change and a coded picture is low, the prediction error information of the coded picture and the generated code amount of the coded picture are estimated. It is preferable to control to correct the equation.

  According to the present invention, in the quantization control for determining the quantization step in the moving picture coding system such as the MPEG standard, the image quality after the coded picture is substantially maintained without prefetching the original image data after the coded picture. It becomes possible to encode.

  FIG. 1 shows a configuration of an image encoding device in a general MPEG standard. In the figure, reference numeral 101 denotes an original image memory for storing input images. Reference numeral 102 denotes an adder that takes the difference between the original image data and predicted image data generated from data encoded in the past. Reference numeral 103 denotes an orthogonal transformation processing unit that transforms the difference data calculated by the adder 102 into the frequency domain, and reference numeral 104 denotes a quantization unit that quantizes the data orthogonally transformed in 103.

  Reference numeral 105 denotes an encoding unit that encodes the data quantized in 104. Reference numeral 106 denotes a buffer that is stored in order to transmit the data encoded in 105. Reference numeral 107 denotes an inverse quantization unit that inversely quantizes the data quantized in 104. Reference numeral 108 denotes an inverse orthogonal transform unit that inversely transforms the data inversely quantized at 107. Reference numeral 109 denotes an adder that adds past picture data stored in the frame memory to the data inversely orthogonally transformed in 108.

  Reference numeral 110 denotes a frame memory for storing the decoded image data calculated by the adder 109, and 111 detects an approximate portion of the original image data and the decoded image data, and is generated by detecting the motion. This is a motion detection / compensation unit that reads out the decoded image data closest to the original image data from the frame memory based on the motion vector thus generated, and generates a predicted image. Reference numeral 113 denotes an activity calculation unit that performs an activity calculation that divides an original image into blocks and determines a feature amount of the block. A quantization control unit 112 controls the quantization step and determines the target code amount and the quality of the encoded image quality. Reference numeral 114 denotes an intra prediction unit that generates a prediction image from pixels around the coding block.

  Next, processing of each step in MPEG-2 Test Model 5 will be described.

  In step 1, first, parameters Xi, Xp, and Xb indicating the amount of generated code of a coded picture and the complexity of the screen are obtained by the following equations.

Here, Si, Sp, and Sb are generated code amounts when I, P, and B pictures are encoded, and Qi, Qp, and Qb are average values of quantization parameters of I, P, and B pictures, respectively. In addition, as the initial values of the parameters Xi, Xp, and Xb, values represented by the following expressions are used.

Here, bit_rate is a target bit rate.

  Next, ratios Kp and Kb between the quantization parameter of the I picture and the quantization parameters of the P and B pictures are determined, and the target code amount of each picture is determined. The values shown in the following equations are used for the values of Kp and Kb, respectively.

Assuming that the target code amount for each I, P, and B picture in a GOP (Group of Picture) is Ti, Tp, and Tb, they are respectively expressed by the following equations.

Here, Np and Nb are the numbers of P and B pictures that have not been encoded in the GOP, and frame_rate is the frame rate. R is a code amount assigned to the GOP and is updated by the following equation.

Further, the following equation is used when encoding the first picture in the GOP.

N is the number of pictures in the GOP. The initial value of R at the beginning of the sequence is 0.

  In step 2, in order to match the target code amount and the generated code amount for each picture obtained in step 1 above, a macroblock (MB: picture) is based on three types of virtual buffer capacities set independently for each picture type. The quantization parameter for each block) is determined. The virtual buffer occupancy of I, P, and B pictures is obtained by the following equation.

Here, Bj is the amount of generated code from the beginning of the picture to the jth macroblock, and is the number of macroblocks in the picture. The quantization parameter of the jth macroblock is calculated by the following equation.

Here, r is a parameter for controlling the feedback response speed, and is given by the following equation.

Note that the initial value of the virtual buffer occupation amount is given by the following equation.

  In step 3, a human visual characteristic that distortion is conspicuous in a flat part and inconspicuous in a complicated part of a pattern is used, and a variance value called activity is used for the discrimination. Then, the quantization step is changed based on the activity. As a specific process, the macroblock is divided into four, and the variance value of the luminance signal is calculated in units of 8 × 8 pixel blocks. The variance value is obtained by the calculation formula of Formula 11. Here, the variable n is the position of the 8 × 8 block in the macro block, the upper left 8 × 8 block is 1 the upper right 8 × 8 block is 2, the lower left 8 × 8 block is 3, the lower right 8 × 8 blocks is set to 4. Pn (x, y) indicates the luminance signal at the (x, y) position in the picture.

The activity is calculated by the formula shown in Formula 12.

Here, the minimum value is taken in order to make the quantization fine when only a part of the macroblock has a flat portion. Further, the weighting coefficient Nact of the quantization step is calculated by the following formula.

Here, avg_act is an average value of the activity of the picture encoded immediately before.
Subsequently, the quantization parameter QP of the coded block is calculated by multiplying the quantization parameter QPj calculated in step 2 by the weighting coefficient Nact.

  By such a method, the amount of generated code is controlled by performing rough quantization on a complicated portion of a pattern that is not easily noticeable visually, and finely quantizing on a flat area where the image is easily visually deteriorated.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
<Example 1>
FIG. 2 is a block configuration diagram showing the configuration of an embodiment of an image encoding device according to the present invention. As shown in the figure, the present image encoding apparatus is an original image memory 101, an adder 102, an orthogonal transform unit 103, a quantization unit 104, an encoding unit 105, a buffer 106, as in a conventional image encoding device. , Inverse quantization unit 107, inverse orthogonal transform unit 108, frame memory 110, motion detection / compensation unit 111, quantization control unit 112, activity calculation unit 113, intra prediction unit 114, scene change detection unit 201, picture type determination Section 202, prediction error information calculation section 203, and target code amount determination section 204.

  In the above configuration, a scene change detection unit 201 that detects a scene change, a picture type determination unit 202 that determines the type of an encoded picture, a prediction image generated based on a prediction mode selected by intra prediction or inter prediction processing Except for the prediction error information updating unit 203 that updates the difference average value of the original image and the target code amount determining unit 204 that determines the target code amount for each picture. The same. Therefore, conventionally known processing units are denoted by the same reference numerals as those in FIG. 1 and description thereof is omitted.

  Hereinafter, with reference to FIG. 3, FIG. 4, FIG. 5, and FIG. 6, the processing in the scene change detection unit 201, the picture determination unit 202, the prediction error information calculation unit 203, and the target code amount determination unit 204, which are the main parts of the present invention The present embodiment will be described focusing on the above. FIG. 3 is a flowchart showing the flow of the encoding process in this embodiment.

  First, in step S301, the correlation between the encoded picture and the picture encoded immediately before the encoded picture is checked to determine whether a scene change has occurred. In the determination of a scene change, for example, a luminance difference value between an encoded picture and a picture encoded immediately before the encoded picture is calculated. If this luminance difference value is greater than a certain threshold, a scene change has occurred. Judge that it is. If no scene change is detected in step S301, the process proceeds to step S302 for determining the target code amount of the encoded picture. Hereinafter, a method for determining the target code amount of the encoded picture will be described.

  In moving picture encoding, the amount of generated code increases because the amount of information required to encode the prediction error increases as the prediction error increases. Further, the smaller the quantization step width, the finer the quantization, and the larger the generated code amount. Hereinafter, the parameter for determining the quantization width is referred to as a quantization parameter QP. As an example, when the relationship between the quantization parameters QP1, QP2, and QP3 is QP1 <QP2 <QP3, the relationship between the prediction error and the generated code amount can be approximated as shown in FIG.

  Based on such characteristics of the image encoding process, the relationship between the prediction error of the picture encoded before the encoded picture and the generated code amount is statistically obtained in advance and formulated as the following equation.

Here, T is a generated code amount of a predicted encoded picture, SAD is a picture prediction error average value encoded in the past, and a and b are variables determined by quantization parameters. For the variables a and b corresponding to each quantization parameter, a correspondence table as shown in FIG. 15 is created and stored in the memory in the target code amount determination unit. In addition, although Formula 15 was represented by the linear expression, you may represent it by the polynomial more than a quadratic expression. At that time, the coefficient increases in proportion to the order. Then, the value calculated using Equation 15 is set as the target code amount for each picture.

  Here, when the past coded picture using the prediction error information is different from the coded picture, the correlation between the prediction error average value of the previously coded picture and the code amount generated in the coded picture is low. Therefore, it may be difficult to determine the target code amount using Equation 15. Therefore, for example, as shown in FIG. 5, when the encoded picture is an I picture (I1), if the target code amount is Ti, the generated code amount of the already encoded I picture (I0) and the average of the QP The target code amount Ti is obtained from the value Xi obtained by multiplying the values as follows:

Here, ave (QP) is an average value of quantization parameters of a picture (P1) encoded immediately before the I picture (I1) to be encoded. Further, a is a weighting coefficient for correcting the generated code amount of the I picture.

  On the other hand, when the encoded picture is a P picture (P2), the picture encoded before the P picture (P2) encoding is the I picture (I1). In this case, before the I picture (I1) encoding The target code amount is determined by Equation 15 using the prediction error average value of the picture (P1) encoded in. Similarly, when the picture to be encoded is a B picture (B2), the target code amount is calculated by Equation 15 using the prediction error average value of the B picture (B1) encoded before the B picture (B2) encoding. decide. The target code amount can also be used for determining the target code amount of the entire next GOP. For example, a target code amount of a picture encoded in the past is stored in a memory for each picture type. Then, the number of each picture is calculated according to the configuration of the picture type in the next GOP (I, B, B, P, B, B, P...), And the target code amount is added to the next GOP. The overall target code amount can be determined.

  Note that, when a plurality of prediction modes are used as in the H.264 encoding method, the prediction error has the highest encoding efficiency among the prediction modes selected by intra prediction or inter prediction processing. The sum of absolute differences between the predicted image generated in the prediction mode and the original image is used. The prediction error may be a value obtained by weighting the sum of absolute differences. In addition, for the prediction error, various parameters can be widely applied as necessary, for example, when a sum of squares of prediction errors is applied instead of the sum of absolute differences between the prediction image and the original image. In addition, as the prediction error average value, the prediction error average value calculated every other macro block may be used instead of the prediction error average of all the macroblocks in the picture. In addition, the prediction error average value may be switched according to the situation by obtaining the prediction error average value of the intra prediction image used when generating the intra prediction image and the prediction error average value generated by the inter prediction process, respectively. .

  After determining the target code amount of the encoded picture by the above method, the activity calculation unit 113 calculates the activity of the encoded macroblock (S303). As an activity calculation method, for example, it is obtained using Formulas 11 and 12. The activity may use the absolute difference sum of luminance values instead of the variance value of luminance values.

  Next, a quantization parameter for each macroblock is determined based on the target code amount determined in step S302 and the activity information calculated in step S303 (S304). In determining the quantization parameter for each macroblock, first, the reference value QPinit of the quantization parameter of the encoded picture is obtained from the target code amount determined in step 302. A method for calculating the reference value Qpinit of the quantization parameter will be described.

  First, as shown in FIG. 6, the intersection T between the target code amount (BIT) determined in step S302 and the prediction error average value (AVG_SAD) of the picture encoded before encoding the encoded picture is obtained. . Next, from the correspondence table storing the relationship between the quantization parameter and a and b, the straight line represented by Equation 15 is obtained, and the quantization parameters QP1 and QP2 corresponding to the two straight lines sandwiching the intersection T are searched. A value obtained by internally dividing the quantization parameters QP1 and QP2 at the position of T is set as a quantization parameter reference value QPinit. Here, the variable D shown in FIG. 6 is the interval between the quantization parameters stored in the correspondence table for associating the quantization parameter with the coefficients a and b. If the value of the stored quantization parameter is 5, D = 5. After determining the quantization parameter reference value Qpinit in this way, the quantization parameter for each macroblock is determined based on the activity information calculated in step S303. Specifically, the quantization parameter weighting coefficient Nact is calculated by Expression 13 and Expression 14, and the quantization parameter reference value Qpinit is weighted with Nact to determine the quantization parameter of the encoded macroblock.

  Next, after the quantization parameter of the encoded macroblock is determined in step S304, intra prediction processing is performed by the intra prediction unit 114 using the quantization parameter in the same manner as in the conventional encoding device. A series of coding processes in which inter prediction processing is performed by the motion compensation unit 111, orthogonal transformation is performed on the prediction error by the orthogonal transformation unit 103, quantization is performed by the quantization unit 104, and coding is performed by the coding unit 105. (S305).

  Next, the prediction error information updating unit 203 calculates an average value of prediction errors from the first macroblock of the picture to the encoded macroblock (S306). The processes of S303 to S306 are repeated until all macroblocks in the picture are completed. Then, after the above processing is completed for all macroblocks in the picture, the process returns to step S301.

  Next, a case where a scene change is detected in step 301 will be described. When a scene change is detected, it is expected that the correlation between the coded picture and the past coded picture will be low. Therefore, when the encoded picture is a P or B picture, many intra prediction modes are selected. Therefore, in such a case, the target code amount is determined using Equation 16. Then, when a picture in which the difference between the target code amount and the actual generated code amount exceeds the threshold TH continues in the encoded picture after the scene change, the values of the coefficients a and b shown in FIG. And the relational expression of Equation 15 is corrected (S309). The processing from S301 to S309 is repeated until all macroblocks in the picture are completed.

In this embodiment, since the prediction error average value of the previously encoded picture of the encoded picture is used, the target code amount and the quantization parameter are determined without prefetching the image data of the encoded picture. be able to. In addition, after determining the target code amount of each picture, in order to move to the quantization parameter determination process for each macroblock corresponding to the size of the activity, the steps 2 and 3 are performed as in the MPEG-2 Test Model 5 method. It does not happen that control is offset.
<Example 2>
In the present embodiment, an image coding apparatus capable of stably coding by avoiding buffer failure by setting an upper limit value and a lower limit value of the code amount generated for each picture will be described. The image coding apparatus of the present embodiment has the same configuration as that shown in FIG.

  In the quantization control process, it is necessary to generate encoded data so that a virtual model of the buffer state of the decoder is not broken (overflow or underflow). FIG. 7 shows a conceptual diagram of the buffer. The horizontal axis is the elapsed time, and the vertical axis is the buffer occupancy, and it is necessary to determine the quantization parameter so that this virtual buffer does not fail.

  FIG. 8 is a flowchart showing the flow of the encoding process in this embodiment. First, in step S801, the state of buffer occupancy immediately before encoding is confirmed. In general, it is desirable to control the buffer occupancy so that the buffer level decreases in an I frame with a large amount of information and then recovers in the subsequent P or B frame. Therefore, for example, as shown in FIG. 7, the quantization parameter is determined so that the buffer occupation amount changes between the overflow limit (Overflow Limit) and the underflow limit (Underflow Limit).

Next, in steps S802 and S804, an upper limit value and a lower limit value of the target code amount of the encoded picture and the allowable generated code amount are determined. Hereinafter, a method for determining the upper limit value and lower limit value of the target code amount and the allowable generated code amount will be described.
At the time of I picture encoding, the upper limit value and the lower limit value of the target code amount and the allowable generated code amount are determined so that the remaining buffer capacity after the I picture encoding does not fall below the minimum buffer level as much as possible.

  FIG. 9 shows a method for determining the target code amount of the I picture and the upper limit value and lower limit value of the allowable generated code amount. B is the buffer occupancy immediately before encoding the I picture. The target code amount Ti for the I picture is calculated using Equation 16 from the value Xi obtained by multiplying the generated code amount of the I picture encoded in the past and the average value of QP, as in the first embodiment. Next, when the buffer occupation amount confirmed in step S801 is located near the minimum buffer level (MinLevel), the target code amount is determined as follows. First, when B-Ti does not fall below the minimum buffer level (MinLevel) ((a) in FIG. 9), the target code amount Ti is not changed, and the upper limit value of the generated generated code amount is set to a × Ti, Let b × Ti be the lower limit of the amount of generated code allowed. Here, the coefficients a and b have values of a> 1, 0 <b <1. On the other hand, when B-Ti is lower than the minimum buffer level ((b) in FIG. 9), B-MinLevel is set as the upper limit value of the generated code amount, c × (B-MinLevel) is the target code amount, d × Let (B-MinLevel) be the lower limit value of the generated code amount that is allowed. Here, the coefficients c and d are 1> c> d> 0.

Next, a method for determining the target code amount at the time of P and B picture encoding and the upper limit value and lower limit value of the allowable generated code amount will be described.
When encoding P and B pictures, the target code amount for P and B pictures, the upper limit value of the allowable generated code amount, and the allowable generated code amount so that the generated code amount of P and B pictures does not fall below the underflow limit. Determine the lower limit.

  FIG. 10 shows a method for determining the target code amount of P and B pictures, the upper limit value of the allowable generated code amount, and the lower limit value of the allowable generated code amount. B is the buffer occupancy immediately before encoding the P and B pictures. For P and B pictures, since the correlation between the prediction error average value of a picture encoded before encoding the encoded picture and the generated code amount of the encoded picture is high, the value calculated using Equation 15 is the target. The code amount is T.

  Next, when the buffer occupation amount confirmed in step S801 is located near the underflow limit boundary, the target code amount is determined as follows. First, when BT does not fall below the buffer underflow limit (UFL) ((a) in FIG. 10), the target T is not changed, and the upper limit of the generated code amount to be allowed is allowed to be a × T. The lower limit value of the generated code amount is b × T. Here, the coefficients a and b have values of a> 1, 0 <b <1. On the other hand, when BT falls below the buffer underflow limit (UFL) ((b) in Fig. 10), (B-UFL) is set as the upper limit of the amount of generated code, and c x (B-UFL) is The target code amount, d × (B-UFL), is set as the lower limit value of the generated code amount. Here, the coefficients c and d have values of 1> c> d> 0.

  Next, processing when the buffer occupation amount B confirmed in step S801 is located near the overflow limit will be described. FIG. 11 shows a method for determining the target code amount of each picture in the vicinity of the overflow limit and the upper limit value and lower limit value of the allowable generated code amount. As the target code amount T of each picture, a value calculated using Equation 15 or Equation 16 is used. If BT does not exceed the overflow limit (OFL), the target code amount T is not changed, the upper limit value of the allowable generated code amount is a × T, and the lower limit value of the allowable generated code amount is b × T. ((A) in FIG. 11). Here, the coefficients a and b are a> 1, 0 <b <1. On the other hand, when BT exceeds the overflow limit (OFL) ((b) in FIG. 11), B-OFL is set as the lower limit value of the generated code amount, e × (B-UFL) is the target code amount, and f × Let (B-UFL) be the upper limit of the amount of generated code that is allowed. Here, the coefficients e and f are 1 <e <f.

  After determining the target code amount of the encoded picture and the upper limit value and lower limit value of the allowable code amount by the method described above, the reference value of the quantization parameter of the encoded picture is determined from the target code amount (S803). The minimum value and the maximum value of the quantization parameter of the coded picture are determined from the upper limit value and lower limit value of the generated code amount (S805). The method for determining the reference value, maximum value, and minimum value of the quantization parameter of the coded picture can be obtained by the same process as the method for obtaining the reference value QPinit from the target code amount described in the first embodiment.

Based on the reference value, maximum value, and minimum value of the quantization parameter in the picture determined as described above, the quantization parameter for each macroblock is determined. (S806)
In the determination of the quantization parameter for each macroblock, the quantization parameter reference value QPinit is weighted according to the size of the activity in the same manner as in the first embodiment, and the QP for each macroblock is determined. That is, the quantization parameter for each macroblock is calculated by applying the processing of Equations 11 to 14. At this time, if the calculated quantization parameter QP of the macroblock exceeds the maximum value QPmax of the quantization parameter in the picture obtained in step S805, QP = QPmax, and if below the minimum value QPmin in the picture QP = QPmin. This embodiment makes it possible to perform stable quantization control while avoiding buffer failure.
<Example 3>
In this embodiment, a method for determining the quantization parameter of each macroblock will be described. This embodiment has the same configuration as that of the encoding apparatus described in the above embodiment. First, the target code amount and the upper limit value and lower limit value of the allowable generated code amount are obtained by the method described in steps S802 and S804 of the second embodiment (S1201). Next, by performing the same processing as steps S803 and S805, the quantization parameter reference value Qpinit, the maximum value QPmax, and the minimum value QPmin of the encoded picture are obtained. Next, based on the activity information calculated by the activity calculation unit 113, the characteristics of the activity and the quantization parameter are determined (S1202).

  Specifically, as shown in FIG. 13, the maximum value Act_max of the activity of the encoded picture before encoding the encoded picture with respect to the maximum value Qpmax of the quantization parameter is set as the minimum value QPmin of the quantization parameter. Is associated with the minimum value Act_min of the activity of the encoded picture before encoding the encoded picture. Next, the quantization parameter QP is determined according to the activity size of the encoded macroblock calculated by the activity calculation unit 113 (S1203). That is, as shown in FIG. 13, when the activity value ACT is in the region A, QP = QPmin, when it is in the region C, QP = QPmax, and when it is in the region B, the following equation is used: Ask.

After determining the quantization parameter QP, the activity maximum value and the activity minimum value calculated by the activity calculation unit 113 are updated (S1205). The processes in steps S1202 to S1205 are performed on all macroblocks in the picture. Note that, as the maximum value and the minimum value of the activity, values obtained by filtering the activity value may be used.

  The activity and QP characteristics determined in step 1202 may be the characteristics shown in FIG. That is, in the quantized parameter reference value QPinit, maximum value QPmax, and minimum value QPmin of the coded picture, the activity average value Act_avg of the picture before coding the coded picture with respect to the quantization parameter reference value Qpinit is maximized. The activity maximum value ACT_max of the picture before coding the coded picture is associated with the value QPmax, and the activity minimum value ACT_min of the picture before coding the coded picture is associated with the minimum value QPmin.

  Then, the quantization parameter QP is determined according to the activity size of the encoded macroblock calculated by the activity calculation unit 113. That is, as shown in FIG. 14, when the activity value ACT is in the region A, QP = QPmin, when it is in the region D, QP = QPmax, and when it is in the region B, Ask.

If it is in region C, it is obtained using the following equation.

As a feature of the distribution of activity in most natural images, most activity values are concentrated in a low area. According to the present embodiment, by changing the sensitivity of the quantization parameter to the fluctuation of the activity value between the high activity region and the low activity region, the quantization parameter that reflects the difference in the activity is reflected in the activity values of all regions. Settings can be made. Further, by adjusting the position of the point Q in FIG. 14, the amount of generated code for each picture can be increased or decreased. That is, as Q moves to the upper left, QP increases as a whole and the amount of generated code is suppressed. Accordingly, it is possible to adjust the code amount by feeding back the difference between the target code amount and the generated code amount for each picture, and to improve the followability of the generated code amount to the target bit rate.
<Example 4>
In this embodiment, the computer can be operated by creating a program that records the step procedure for executing the encoding process in the above embodiment. Note that a program that executes such encoding processing can be downloaded and used by a user via a network such as the Internet. It can also be used by being recorded on a recording medium. Such a recording medium can be widely applied to recording media such as an optical disk and a magneto-optical disk.

  In the above embodiment, the unit of the input image has been described as a picture. However, the present invention can also be applied to the case where encoding is performed in units of a plurality of divided slices. Further, the present invention can also be applied to the case where an interlaced image is encoded as input data and is encoded as a field. In the above embodiment, the case where processing is performed in units of macroblocks (16 × 16 pixels) has been described. However, when processing is performed using blocks of an arbitrary size such as 32 × 32 pixels or 8 × 8 pixels. Is also applicable. The present invention can be widely applied to various image data and encoding methods including the H.264 encoding method.

It is a block diagram which shows the structure of the conventional image coding apparatus. It is a block block diagram which shows 1st, 2nd, 3rd embodiment of the image coding apparatus by this invention. It is a flowchart which shows the flow of the process for the setting of the quantization step in 1st Embodiment. It is the conceptual diagram which showed the relationship between the prediction error and generated code amount in an encoding process. In this invention, it is the figure which showed the positional relationship of the encoding picture and the picture used for target code amount calculation. In this invention, it is the figure which showed the method to determine which calculates a quantization parameter from target code amount. It is the conceptual diagram of the buffer which showed transition of the buffer occupation amount. It is a flowchart which shows the flow of the process for the setting of the quantization step in 2nd Embodiment. It is the figure which showed the determination method of the target code amount of I picture in the underflow boundary in 2nd Embodiment, and the generation | occurrence | production code amount to accept | permit. It is the figure which showed the determination method of the target code amount of P and B pictures in the underflow boundary in 2nd Embodiment, and the generation | occurrence | production code amount to accept | permit. It is the figure which showed the determination method of the target code amount of each picture and the allowable generated code amount in the overflow boundary in 2nd Embodiment. It is a flowchart which shows the flow of the process for the setting of the quantization step in 3rd Embodiment. It is the figure which showed the characteristic of the activity and quantization parameter in 3rd Embodiment. It is the figure which showed the characteristic of the activity and quantization parameter in 3rd Embodiment. It is an example of the correspondence table which determines the relationship between the prediction error and the generated code amount in the present invention.

Explanation of symbols

101 ... Original image memory,
102, 109 ... adder,
103… Direct conversion part,
104 ... Quantization part,
105: Encoding unit,
106 ... buffer,
107: Inverse quantization unit,
108 ... Inverse transformation unit,
110… Frame memory,
111: Motion detection / compensation unit,
112 ... Quantization control unit,
113… Activity calculation part,
114 ... Intra prediction part,
201 ... Scene change detection unit,
202 ... Picture type determination unit,
203 ... Prediction error information update unit,
204: Target code amount determination unit.

Claims (17)

An image encoding device that encodes moving image information including a series of pictures,
Original image storage means for storing an original image of a picture input from the outside;
Based on the original image, intra prediction means for performing intra prediction and inter prediction means for performing inter prediction,
Feature amount calculating means for calculating a feature amount of the picture to be encoded,
Prediction error information updating means for updating prediction error information based on the difference between the prediction image generated by the intra prediction means or the inter prediction means and the original image;
Target code amount determining means for determining a target code amount for each picture to be encoded according to the updated prediction error information and the generated code amount of the encoded picture;
A quantization control unit configured to determine a quantization width of each of the plurality of blocks to be encoded based on the feature amount and the target code amount in the picture to be encoded divided into a plurality of blocks; An image encoding device. The image encoding device according to claim 1,
A buffer for storing the generated code amount;
The image coding apparatus, wherein the target code amount is determined based on a storage capacity state of the buffer and prediction error information. The image encoding device according to claim 1,
The image coding apparatus according to claim 1, wherein the prediction error information is a value calculated based on a past picture coded before the picture to be coded. The image encoding device according to claim 1,
Means for multiplying a generated code amount of the past picture by an average value of quantization widths in the picture;
When the picture to be coded is intra prediction, an image coding apparatus is characterized in that a target code amount is determined based on the multiplication value. The image encoding device according to claim 1,
Means for formulating a relationship between a prediction error of the past picture and a generated code amount of the picture to be encoded;
When the picture to be encoded is inter-screen prediction, a target code amount is determined based on the relational expression. In any one of the image coding apparatuses of Claim 1 thru | or 5,
A scene change detecting means for detecting a difference value between an attribute of the picture to be encoded and an attribute of a picture encoded immediately before the picture to be encoded, and detecting the presence or absence of a scene change;
When it is determined that the difference value is smaller than a predetermined threshold, a target code amount is determined based on a relational expression between prediction error information of the past picture and a generated code amount of the picture to be encoded. Image encoding device. The image encoding device according to claim 6, wherein
When the scene change is detected, an image coding apparatus, wherein a relational expression between prediction error information of the past picture and a generated code amount of the picture to be coded is updated. The image encoding device according to claim 6, wherein
An image encoding apparatus comprising: a picture type determination unit that determines a type of a picture to be encoded. The image encoding device according to claim 2, wherein
An image encoding apparatus, wherein the target code amount is determined by a relational expression prepared in advance based on a relationship between a prediction error of a picture encoded immediately before the picture to be encoded and a generated code amount. The image encoding device according to claim 1,
Provide an allowable range of generated code amount of the picture to be encoded,
An image encoding apparatus that determines an upper limit value and a lower limit value of the quantization width based on the allowable range. The image encoding device according to claim 10.
Means for detecting a generated code amount situation for each picture;
The image encoding device, wherein the allowable range is determined based on the generated code amount situation. The image encoding device according to claim 1,
An image encoding apparatus, wherein the maximum value, the minimum value, and the average value of the feature amount are calculated in units of pictures. The image encoding device according to claim 12, wherein
An image code characterized by determining the quantization width of the block to be encoded by associating the maximum value and minimum value of the quantization width with the maximum value and minimum value of the feature amount, respectively. Device. The image encoding device according to claim 12, wherein
Determining the quantization width of the block to be encoded by associating the maximum value, minimum value, and reference value of the quantization width with the maximum value, minimum value, and average value of the feature amount, respectively; An image encoding device characterized by the above. A video encoding program having a step of encoding video information consisting of a series of pictures,
Original image storage means for storing an original image of a picture input from the outside;
Based on the original image, an intra prediction step for performing intra-picture prediction and an inter prediction step for performing screen prediction;
Calculating a feature amount for calculating a feature amount of the picture to be encoded, and
Updating prediction error information based on the difference between the prediction image generated by the intra prediction step or the inter prediction step and the original image;
Determining a target code amount for determining a target code amount for each picture to be encoded, according to the updated prediction error information and a generated code amount of the encoded picture;
A moving image for causing a computer to execute a step of controlling a quantization width of each of the plurality of blocks to be encoded based on the feature amount and the target code amount in the picture to be encoded divided into a plurality of blocks Encoding program. In the moving image encoding program according to claim 15,
A moving picture encoding program comprising a picture type determination step for determining a type of picture to be encoded. In the moving image encoding program according to claim 16,
Detecting a difference value between an attribute of the picture to be encoded and an attribute of a picture encoded immediately before the picture to be encoded, and detecting the presence or absence of a scene change,
And determining a target code amount based on a relational expression between the prediction error information of the past picture and the generated code amount of the picture to be encoded when it is determined that the difference value is larger than a predetermined threshold. A moving image encoding program as a feature.

Images