JP2011109385A - Image encoder - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform high-accuracy rate control by suppressing the occurrence of a control error of a code amount even in one-pass processing. <P>SOLUTION: An image encoder is provided with a quantization step size calculation part 20 that calculates a quantization step size Δ from a target code amount BT(n) and a statistic E(n) so as to give the quantization step size Δ to a quantization part 13 and a lower-bit truncation part 16 that truncates all or a part of lower bits outputted without being compressed from an entropy coding part 15 when a code amount B(n) of code data outputted from the entropy coding part 15 to an output buffer 17 exceeds a target code amount BT(n). <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an image encoding apparatus that controls transmission so that the code amount of code data becomes a constant value and realizes transmission at a desired encoding rate.

In order to transmit a high-resolution image at a constant rate, the image is divided into an appropriate size, and the quantization step size is controlled and encoded for each divided image so that each divided image has a constant code amount. Technology is generally adopted.
FIG. 13 is a block diagram showing a general image encoding apparatus that employs a technique for controlling and encoding the quantization step size.
Hereinafter, a method for controlling the quantization step of the stripe n using a stripe n (n = 1, 2,..., N) having a plurality of lines × image width as a rate control unit will be described (for example, see Patent Document 1). reference).

(1) Calculation process of target code amount of stripe n In this calculation process, the target code amount of the stripe n to be encoded is calculated from the code amount of the already encoded stripe.
At this time, if a large amount of code has occurred in the past stripe, the target code amount is decreased in the next stripe, and conversely, if the code amount is small, the target code amount is determined in the next stripe. Increase

(2) Statistic Calculation Processing for Stripe n In this calculation processing, a statistic indicating the information amount of stripe n is calculated in order to predict the code amount generated in the stripe to be encoded.
For example, as the statistic of the stripe n, the distribution of the stripe n, the difference power between adjacent pixels, and the like are calculated.

(3) Quantization Step Size Calculation Processing for Stripe n In this calculation processing, a quantization step size that is predicted that the code amount of stripe n becomes the target code amount is determined with reference to the statistic of stripe n.
For example, if the stripe has a complicated pattern, the quantization step size is increased. Conversely, if it is a simple stripe, the quantization step size is decreased.
Thereby, an appropriate value of the quantization step size is set for each stripe, and the generated code amount in each strip is controlled.

EP0631443A1

Since the conventional image coding apparatus is configured as described above, the quantization step size predicted when the code amount of the stripe n becomes the target code amount is determined, but the accurate code amount of the stripe n is If you do not actually implement Huffman coding, you will not know. For this reason, there is a problem that a control error of a certain amount of code occurs in the one-pass process.
If multi-pass processing is performed, an appropriate quantization step size can be found, but Huffman encoding is performed many times, which increases the amount of processing.

  The present invention has been made to solve the above-described problems, and is an image coding capable of realizing high-accuracy rate control by suppressing the occurrence of a code amount control error even in one-pass processing. The object is to obtain a device.

  An image encoding apparatus according to the present invention includes a target code amount calculating unit that calculates a target code amount from a code amount of an already encoded stripe, and a statistical amount calculation that calculates a statistical amount of a stripe divided by the image dividing unit. A quantization step size calculated from the target code amount calculated by the means and the target code amount calculation unit and the statistic calculated by the statistic calculation unit, and the quantization step size is given to the quantization unit When the code amount of the code data output from the calculating means and the entropy encoding means to the output buffer means exceeds the target code amount calculated by the target code amount calculating means, the code data is output without being compressed from the entropy encoding means. And lower bit truncation means for truncating all or part of the lower bits.

  According to this invention, the target code amount calculating means for calculating the target code amount from the code amount of the already encoded stripe, the statistical amount calculating means for calculating the statistical amount of the stripe divided by the image dividing means, and the target A quantization step size calculating unit that calculates a quantization step size from the target code amount calculated by the code amount calculating unit and the statistic calculated by the statistic calculating unit, and gives the quantization step size to the quantizing unit; When the code amount of the code data output from the entropy encoding unit to the output buffer unit exceeds the target code amount calculated by the target code amount calculation unit, the lower bit of the lower bits output without being compressed from the entropy encoding unit Since it is configured to provide a lower-order bit truncation means that truncates all or part of the code amount, a code amount control error occurs even in one pass processing. To suppress an effect that can realize highly accurate rate control.

It is a block diagram which shows the image coding apparatus by Embodiment 1 of this invention. It is a flowchart which shows the processing content of the image coding apparatus by Embodiment 1 of this invention. It is a flowchart which shows the processing content of the low-order bit truncation part 16 of the image coding apparatus by Embodiment 1 of this invention. It is explanatory drawing which shows an example of the stripe divided | segmented by the stripe division part 11. FIG. It is explanatory drawing which shows the relationship between the statistic of a stripe in the case of QP = 3, 10, 25, 150 and the amount of generated codes. It is explanatory drawing which shows the conversion coefficient of "JPEG XR". It is explanatory drawing which represents typically the bit plane from MSB (Most Significant Bit) to LSB (Least Significant Bit) at the time of expressing the absolute value of a prediction error by a natural binary number. It is a table figure which shows the correspondence of the reduction bit number of a low-order bit, and the reduction code amount. It is explanatory drawing which shows the example in which the tile is comprised by the macroblock of 64x2. It is a block diagram which shows the image coding apparatus by Embodiment 3 of this invention. It is explanatory drawing which shows the uncompressed transmission of a low-order bit. It is a table figure which shows the correspondence of the number of reduction bits of a low-order bit, the presence or absence of reduction of a sign bit, and a reduction code amount. It is a block diagram which shows the general image coding apparatus by which the technique which controls and encodes a quantization step size is employ | adopted.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing an image coding apparatus according to Embodiment 1 of the present invention.
In FIG. 1, the encoding processing unit 1 uses, for example, “JPEG XR (ISO / IEC 29199-2)”, which is a standard encoding method for still images, to perform encoding processing of image data. It is.
The stripe dividing unit 11 of the encoding processing unit 1 performs a process of dividing the image to be encoded into stripe (n) (n = 1, 2,..., N) units. The stripe dividing unit 11 constitutes an image dividing unit.

For example, in “JPEG XR”, an image can be divided into rectangular regions called tiles, and encoding can be performed by setting encoding parameters such as a quantization step size for each tile.
Here, FIG. 4 is an explanatory diagram showing an example of a stripe divided by the stripe dividing unit 11.
In the example of FIG. 4, the image to be encoded is divided into 32 lines of stripes, and this stripe is used as a control unit for the code amount.
In the example of FIG. 4, the image to be encoded is divided in the horizontal direction, but the image to be encoded may be divided in the vertical direction.

The conversion processing unit 12 performs a conversion process on the image signal of the stripe (n) divided by the stripe dividing unit 11, and performs a process of outputting the conversion coefficient of the image signal.
The quantization unit 13 performs a process of quantizing the transform coefficient output from the transform processing unit 12 with the quantization step size Δ given from the quantization step size calculating unit 20 and outputting a quantized value of the transform coefficient. .
The transform processing unit 12 and the quantization unit 13 constitute a quantization unit.

The prediction processing unit 14 performs processing for calculating a prediction error of the quantized value output from the quantization unit 13.
The entropy encoding unit 15 performs a process of outputting the lower bits of the plurality of bits indicating the absolute value of the prediction error calculated by the prediction processing unit 14 without compression, and outputting the upper bits after performing Huffman coding. carry out. When the prediction error absolute value is non-zero, 1-bit information indicating the sign of the prediction error is also transmitted.
The prediction processing unit 14 and the entropy encoding unit 15 constitute entropy encoding means.

  The lower bit truncation unit 16 outputs the code data of the stripe (n) output from the entropy encoding unit 15 (in addition to the Huffman encoded upper bit code data, the lower bit uncompressed code output without being compressed) If the code amount B (n) of the data (including data) exceeds the target code amount BT (n) calculated by the target code amount calculation unit 18, the lower-order bits output without being compressed from the entropy encoding unit 15 Implement the process of truncating all or part of it. The lower bit truncation unit 16 constitutes lower bit truncation means.

The output buffer unit 17 temporarily accumulates the code data (n) subjected to the truncation process by the lower bit truncation unit 16 and performs a process of outputting the code data (n) at a constant speed. The output buffer unit 17 constitutes output buffer means.
The target code amount calculation unit 18 performs processing for calculating the target code amount BT (n) from the code amount of the already encoded stripe. The target code amount calculation unit 18 constitutes a target code amount calculation unit.

The statistic calculation unit 19 performs a process of calculating the statistic E (n) of the stripe (n) divided by the stripe division unit 11. The statistic calculator 19 constitutes a statistic calculator.
The quantization step size calculation unit 20 calculates a quantization parameter QP (1 to 255) from the target code amount BT (n) calculated by the target code amount calculation unit 18 and the statistic E (n) calculated by the statistic calculation unit 19. Is calculated), a quantization step size Δ corresponding to the quantization parameter QP is calculated, and a process of giving the quantization step size Δ to the quantization unit 13 is performed. The quantization step size calculation unit 20 constitutes a quantization step size calculation unit.

  In the example of FIG. 1, the stripe division unit 11, the transformation processing unit 12, the quantization unit 13, the prediction processing unit 14, the entropy coding unit 15, the lower bit truncation unit 16, and the output buffer, which are components of the image coding device Each of the unit 17, the target code amount calculation unit 18, the statistic calculation unit 19, and the quantization step size calculation unit 20 is configured by dedicated hardware (for example, a semiconductor integrated circuit in which a CPU is mounted or a one-chip microcomputer). However, when the image coding apparatus is configured by a computer, the stripe dividing unit 11, the transform processing unit 12, the quantization unit 13, the prediction processing unit 14, the entropy coding unit 15, and the lower order Professionals describing the processing contents of the bit truncation unit 16, output buffer unit 17, target code amount calculation unit 18, statistic calculation unit 19 and quantization step size calculation unit 20 Storing the ram in a memory of the computer, may execute a program that the CPU of the computer is stored in the memory.

FIG. 2 is a flowchart showing the processing contents of the image coding apparatus according to Embodiment 1 of the present invention.
FIG. 3 is a flowchart showing the processing contents of the lower bit truncation unit 16 of the image coding apparatus according to Embodiment 1 of the present invention.

Next, the operation will be described.
In the first embodiment, the encoding processing unit 1 performs encoding processing of image data using “JPEG XR (ISO / IEC 29199-2)” which is a standard encoding method for still images. However, the encoding method is not limited to “JPEG XR (ISO / IEC 29199-2)”, and other encoding methods may be used.

When image data indicating an image to be encoded is input, the stripe dividing unit 11 of the encoding processing unit 1 includes 32 lines by dividing the image to be encoded in the horizontal direction as illustrated in FIG. Dividing into N stripes (n) (n = 1, 2,..., N) (step ST1 in FIG. 2).
Here, the image to be encoded is divided in the horizontal direction, but the image to be encoded may be divided in the vertical direction.
Thereafter, in the image coding apparatus, the stripe is the control unit of the code amount.

The target code amount calculation unit 18 calculates the target code amount BT (n) from the code amount of the stripe already encoded by the encoding processing unit 1 (stripe before the encoding target stripe) (step ST2). .
In other words, the target code amount calculation unit 18 calculates the quantization step size Δ used by the encoding processing unit 1 for the encoding process of the stripe (n) which is the nth stripe. The target code amount BT (n) is calculated from the code amount.

ただし、式（１）において、ＢＳはストライプ当りの目標平均符号量、ストライプ（ｎ）で実際に発生する符号量である。
式（１）によれば、過去の５つのストライプで発生したレート制御の誤差がストライプ（ｎ）で補償されるように、目標符号量ＢＴ（ｎ）が算出される。 The target code amount calculation unit 18 calculates the target code amount BT (n) of the stripe (n) to be small when the code amount is excessive in the past stripes. When the number is small, the target code amount BT (n) of the stripe (n) is calculated to be large.
Specifically, the target code amount of the stripe (n) is calculated by the following equation (1).

However, in Equation (1), BS is the target average code amount per stripe and the code amount actually generated in the stripe (n).
According to the equation (1), the target code amount BT (n) is calculated so that the rate control error generated in the past five stripes is compensated for by the stripe (n).

When the stripe dividing unit 11 divides the image to be encoded into N stripes, the statistic calculating unit 19 calculates a statistic E (n) indicating the information amount of the stripe (n) (step ST3).
As the statistic E (n) of the stripe (n), the distribution of the stripe (n), the difference power between adjacent pixels, and the like may be calculated. Here, in the stripe (n), the conversion in “JPEG XR” is performed. The average value of the absolute value of the LP component of the coefficient (details will be described later, but corresponds to the LP coefficient in FIG. 6), and the average value is used as the statistic E (n) of the stripe (n). To do.

When the target code amount calculation unit 18 calculates the target code amount BT (n) and the statistic calculation unit 19 calculates the statistic E (n), the quantization step size calculation unit 20 calculates the target code amount BT (n ) And the statistic E (n), a quantization parameter QP (a parameter taking a value of 1 to 255) is calculated.
Specifically, the quantization parameter QP is calculated as follows.

First, the quantization step size calculation unit 20 investigates the relationship between the stripe statistic and the generated code amount while changing the quantization parameter QP of “JPEG XR” using a general image.
FIG. 5 is an explanatory diagram showing the relationship between the stripe statistic and the generated code amount when QP = 3, 10, 25, 150.
When the quantization step size calculation unit 20 investigates the relationship between the statistic of the stripe and the generated code amount, the quantization step size calculation unit 20 tabulates the relationship between the statistic and the generated code amount.

When the quantization step size calculator 20 tabulates the relationship between the statistic and the generated code quantity, the quantization parameter QP corresponding to the target code quantity BT (n) and the statistic E (n) is referred to the table. Is identified.
For example, when the relationship shown in FIG. 5 is tabulated, if the target code amount BT (n) is “bt1” and the statistic E (n) is “e1”, QP = 3, 10, 25, Among 150, since QP = 10 is the closest quantization parameter, it is specified that the quantization parameter QP corresponding to the target code amount BT (n) and the statistic E (n) is “10”.
If the target code amount BT (n) is “bt2” and the statistic E (n) is “e2”, QP = 150 is the closest quantization among QP = 3, 10, 25, 150. Since it becomes a parameter, it is specified that the quantization parameter QP corresponding to the target code amount BT (n) and the statistic E (n) is “150”.

式（２）において、Ａ％Ｂは、ＡをＢで除算したときの余りを示す演算記号である。
また、Ｃ＜＜Ｄは、ＣをＤビットだけ左にシフトする旨を示す演算記号であり、Ｅ＞＞Ｆは、ＥをＦビットだけ右にシフトする旨を示す演算記号である。
量子化ステップサイズ算出部２０は、量子化パラメータＱＰに対応する量子化ステップサイズΔを算出すると、その量子化ステップサイズΔを量子化部１３に与える。 When the quantization step size calculation unit 20 specifies the quantization parameter QP corresponding to the target code amount BT (n) and the statistic E (n), the quantization step size calculation unit 20 calculates the quantization step size Δ corresponding to the quantization parameter QP. (Step ST4).
That is, the quantization step size calculation unit 20 calculates the quantization step size Δ corresponding to the quantization parameter QP using the following equation (2).

In equation (2), A% B is an arithmetic symbol indicating the remainder when A is divided by B.
C << D is an operation symbol indicating that C is shifted to the left by D bits, and E >> F is an operation symbol indicating that E is shifted to the right by F bits.
After calculating the quantization step size Δ corresponding to the quantization parameter QP, the quantization step size calculation unit 20 gives the quantization step size Δ to the quantization unit 13.

When the stripe dividing unit 11 divides the image to be encoded into N stripes, the conversion processing unit 12 performs conversion processing on the image signal of the stripe (n), thereby generating a conversion coefficient of the image signal. The transform coefficient is output to the quantization unit 13.
In other words, the conversion processing unit 12 divides the image of the stripe (n) into 16 × 16 macroblocks, and executes two-stage LBT (Lapped Biorthogonal Transform) on each macroblock, so that FIG. A transform coefficient (DC coefficient, LP coefficient, HP coefficient) having a three-stage hierarchical structure as shown is generated.

Specifically, conversion coefficients (DC coefficient, LP coefficient, HP coefficient) of the image signal are generated as follows.
First, a macroblock (16 × 16 pixels) is divided into blocks each consisting of 4 × 4 pixels, and LBT is executed.
With this first stage LBT, 15 AC coefficients and 1 DC coefficient are generated in each block. Hereinafter, the AC coefficient is referred to as “HP coefficient”, and the number of HP coefficients per macroblock is 240 (= 15 coefficients × 16 blocks).
Next, LBT is performed again on the 16 DC coefficients collected from each block. Hereinafter, the 15 AC coefficients generated by the LBT are referred to as “LP coefficients”.
The number of LP coefficients per macroblock is 15, and one DC coefficient is generated.
Hereinafter, the coefficient of the DC component generated in the second stage is referred to as “DC coefficient”, and the number of DC coefficients per macroblock is one.

  When the quantization unit 13 receives the conversion coefficient of the image signal of the stripe (n) from the conversion processing unit 12, the quantization unit 13 quantizes the conversion coefficient by the quantization step size Δ given from the quantization step size calculation unit 20, The quantized value of the transform coefficient is output to the prediction processing unit 14.

When the prediction processing unit 14 receives the quantized value of the transform coefficient from the quantizing unit 13, the prediction processing unit 14 calculates a prediction error of the quantized value.
That is, the prediction processing unit 14 has already eliminated the correlation between the pixels to some extent by the conversion processing of the conversion processing unit 12, but by applying further prediction to the quantized value of the conversion coefficient, We are trying to improve efficiency.
When the transform coefficient is a DC coefficient, prediction is performed with reference to DC coefficients of neighboring macroblocks located around the macroblock to be encoded.
For example, the value of the DC coefficient of the macroblock located above the macroblock to be encoded is T, the value of the macroblock located on the left is L, and the value of the macroblock located on the upper left is In the case of D, the predicted value P of the DC coefficient of the macroblock to be encoded is calculated from the following equation (3).

When the prediction processing unit 14 calculates the prediction error of the quantized value, the entropy encoding unit 15 calculates lower order bits among a plurality of bits indicating the absolute value of the prediction error in order to simplify entropy encoding. Are output to the lower bit truncation unit 16 without compression, and the upper bits are Huffman encoded and output to the lower bit truncation unit 16 (step ST5).
Specifically, the prediction error of the quantized value is entropy-coded as follows.

For LP and HP coefficients, 15 coefficients are scanned in a predetermined order in units of blocks (15 coefficients), and Huffman coding is applied to non-zero coefficient values and the number of consecutive zero coefficients (zero run). To do. The Huffman code table is prepared by preparing a plurality of candidates and learning the code table to be used from the encoded coefficients.
Here, FIG. 7 is an explanatory diagram schematically showing bit planes from MSB (Most Significant Bit) to LSB (Least Significant Bit) when the absolute value of the prediction error is expressed in a natural binary number.
As shown in FIG. 7, the upper bits are compressed by Huffman coding, but since the lower bits are signals that are close to random, efficient compression cannot be expected even if Huffman coding is performed, so the signal value is not compressed. Transmit as it is.

However, when the absolute value of the prediction error is a non-zero value, 1-bit information (sign bit) indicating the sign of the prediction error is added to the lower bits and transmitted.
The number of bits of the lower bits is determined so that the ratio of the coefficients at which all the values of the upper bits to be Huffman encoded are “0” is approximately 3/4 of the whole.
Further, the HP component can be reduced by the number of bits designated from the lower bits to be transmitted without compression. The number of bits to be reduced can be specified for each tile, and the number of bits specified by all the HP coefficients in the tile is reduced from the lowest.

The lower bit truncation unit 16 obtains the code data of the stripe (n) from the entropy encoding unit 15 (in addition to the Huffman encoded upper bit code data, the lower bit uncompressed code data output without being compressed). The code amount B (n) of the code data is compared with the target code amount BT (n) calculated by the target code amount calculation unit 18.
If the code amount B (n) of the code data does not exceed the target code amount BT (n), the lower bit truncation unit 16 does not perform the lower bit truncation process, but the code amount B (n of the code data ) Exceeds the target code amount BT (n), a process of truncating all or a part of the lower bits output without being compressed from the entropy encoding unit 15 is performed (step ST6).

The quantization unit 13 performs the process of quantizing the transform coefficient using the quantization step size Δ corresponding to the target code amount BT (n). If the Huffman coding is not actually performed, Since an accurate code amount cannot be grasped, a control error of a certain amount of code occurs in one-pass processing.
Therefore, when the generated code amount of each stripe exceeds the target code amount, the code amount is controlled more accurately by truncating all or a part of the lower bits transmitted without compression in the HP coefficient. be able to.
Hereinafter, the processing content of the lower-order bit truncation unit 16 will be specifically described.

FIG. 8 is a table showing the correspondence between the number of reduced bits of the lower bits and the reduced code amount.
In FIG. 8, the reduction bit number k indicates the number of lower bits to be rounded down (the number of bits from the least significant bit), and the reduction code amount b (k) is when the lower bits are rounded down by the reduction bit number k. The code amount to be reduced is shown in FIG.
First, when the lower bit truncation unit 16 receives the code data of the stripe (n) from the entropy encoding unit 15, it prepares a counter representing the reduction bit number k, and the counter value k (reduction bit number k). Is initialized to “0” (step ST11 in FIG. 3).

Next, the lower-order bit truncation unit 16 performs a process of truncating the least significant bit among the lower-order bits that have not yet been discarded.
At this time, with reference to the table of FIG. 8, the reduced code amount b (k) corresponding to the count value k (reduced bit number k) of the counter is subtracted from the code amount B (n) of the code data. The code amount (= code amount B (n) −reduced code amount b (k)) is compared with the target code amount BT (n) (step ST12).

The lower-order bit truncation unit 16 determines that the code amount after subtraction (= code amount B (n) −reduced code amount b (k)) is equal to or smaller than the target code amount BT (n) or the count value k of the counter If (reduction bit number k) has reached the maximum value kmax, the lower-order bit truncation process is terminated.
On the other hand, if the code amount after subtraction (= code amount B (n) −reduced code amount b (k)) exceeds the target code amount BT (n), the lower bits to be cut off are increased, The count value k (reduction bit number k) is incremented by 1 (step ST13), and the process returns to step ST12.
As a result, the code data output to the output buffer unit 17 is truncated in order from the least significant bit until the code amount of the code data falls below the target code amount BT (n).

When image data is encoded using “JPEG XR (ISO / IEC 29199-2)”, which is a standard encoding method for still images, the same reduction is achieved from all HP coefficients in the tile. The lower bits, which are uncompressed data, are reduced by the number of bits k.
Therefore, when a certain tile is composed of 64 × 2 macroblocks as shown in FIG. 9, for example, if the number of bits to be reduced is k = 3, 64 × 2 macroblocks Thus, 3 bits of uncompressed data is reduced from all HP coefficients.

When the output buffer unit 17 receives the code data (n) subjected to the truncation process by the lower bit truncation unit 16, the output buffer unit 17 temporarily accumulates the code data (n) and stores the code data (n) at a constant speed. Output.
Steps ST1 to ST6 are repeated until the last stripe encoding process is completed (step ST7).

As is apparent from the above, according to the first embodiment, the target code amount calculation unit 18 that calculates the target code amount BT (n) from the code amount of the already encoded stripe and the stripe division unit 11 perform division. A statistic calculation unit 19 that calculates a statistic E (n) of the stripe (n) that is calculated, a target code amount BT (n) that is calculated by the target code amount calculation unit 18, and a statistic calculation unit 19 A quantization step size Δ is calculated from the statistic E (n), the quantization step size Δ is supplied to the quantization unit 13, and the entropy encoding unit 15 outputs the output to the output buffer unit 17. When the code amount B (n) of the encoded data exceeds the target code amount BT (n) calculated by the target code amount calculation unit 18, the lower order output without being compressed from the entropy encoding unit 15 Since the low-order bit truncation unit 16 that truncates all or part of the bits is provided, the generation of a code amount control error is suppressed even in one pass processing, and high-accuracy rate control is realized. There is an effect that can.
That is, by reducing not only the quantization step size Δ but also the uncompressed portion of the code data, it is possible to control the code amount with higher accuracy than when the code amount is controlled only by the quantization step size Δ. become.

Embodiment 2. FIG.
In the first embodiment, it has been shown that uncompressed data corresponding to the same number k of reduction bits is reduced from all HP coefficients in all macroblocks. However, the number of reduction bits can be changed in units of macroblocks. (In this case, it is out of the standard of “JPEG XR”).
Therefore, the uncompressed data corresponding to the reduction bit number k may be reduced from the HP coefficient of a specific macroblock instead of all the macroblocks.

For example, when a tile is composed of 64 × 2 macroblocks as shown in FIG. 9, the least significant 2 bits are reduced from all HP coefficients in all macroblocks, and then the macro Consider a case in which the third bit from the least significant bit of each macroblock is reduced in the raster scan order from the block (0, 0).
At this time, it is assumed that the code amount B (n) of the code data falls below the target code amount BT (n) at the stage where the third bit from the least significant bit of the macroblocks (0, 0) to (1, 1) is reduced. .
In this case, truncation of the lower bits is stopped at this stage, and the index (1, 1) of the macroblock that becomes the transmission boundary is recorded in the header of the code stream at the number of reduced blocks of 2 bits and the third bit from the lowest. To.
In this manner, rate control can be performed with higher accuracy by reducing lower-order bits in units of macroblocks.

Embodiment 3 FIG.
10 is a block diagram showing an image coding apparatus according to Embodiment 3 of the present invention. In the figure, the same reference numerals as those in FIG.
As with the lower bit truncation unit 16 in FIG. 1, the lower bit truncation unit 21 uses the code amount B (n) of the code data of the stripe (n) output from the entropy encoding unit 15 as the target code amount calculation unit. When the target code amount BT (n) calculated by 18 is exceeded, a process of truncating all or part of the lower bits output from the entropy encoding unit 15 without being compressed is performed. In addition, a process of outputting the reduced bit number m indicating the number of bits of the lower-order bits rounded down to the quantization step size calculation unit 22 is performed. The lower bit truncation unit 21 constitutes lower bit truncation means.

  The quantization step size calculation unit 22 calculates a quantization parameter QP (1 to 255) from the target code amount BT (n) calculated by the target code amount calculation unit 18 and the statistic E (n) calculated by the statistic calculation unit 19. And a subtracting bit number m output from the lower bit truncation unit 21 from the quantization parameter QP, and a quantization step corresponding to the quantization parameter QP-m after the subtraction A process of calculating the size Δ and giving the quantization step size Δ to the quantization unit 13 is performed. The quantization step size calculation unit 22 constitutes a quantization step size calculation unit.

  In the example of FIG. 10, the stripe division unit 11, the transformation processing unit 12, the quantization unit 13, the prediction processing unit 14, the entropy coding unit 15, the lower bit truncation unit 21, and the output buffer, which are components of the image coding device Each of the unit 17, the target code amount calculation unit 18, the statistic calculation unit 19, and the quantization step size calculation unit 22 is configured by dedicated hardware (for example, a semiconductor integrated circuit in which a CPU is mounted or a one-chip microcomputer). However, when the image coding apparatus is configured by a computer, the stripe dividing unit 11, the transform processing unit 12, the quantization unit 13, the prediction processing unit 14, the entropy coding unit 15, and the lower order A profile describing the processing contents of the bit truncation unit 21, the output buffer unit 17, the target code amount calculation unit 18, the statistic calculation unit 19, and the quantization step size calculation unit 22. Stores grams in the memory of a computer, may be executed the program by the CPU of the computer is stored in the memory.

Next, the operation will be described.
Other than the lower bit truncation unit 21 and the quantization step size calculation unit 22 are the same as in the first embodiment, and in this second embodiment, the lower bit truncation unit 21 and the quantization step size calculation unit 22 Processing contents will be described.

When receiving the code data of the stripe (n) from the entropy encoding unit 15, the low-order bit truncation unit 21 determines that the code amount B (n) of the code data is the same as the low-order bit truncation unit 16 of FIG. When the target code amount BT (n) calculated by the target code amount calculation unit 18 is exceeded, a process of truncating all or part of the lower bits output from the entropy encoding unit 15 without being compressed is performed.
However, unlike the lower bit truncation unit 16 in FIG. 1, the lower bit truncation unit 21 outputs a reduced bit number m indicating the number of bits of the lower bits discarded to the quantization step size calculation unit 22.

  Similar to the quantization step size calculation unit 20 of FIG. 1, the quantization step size calculation unit 22 investigates the relationship between the stripe statistic and the generated code amount, and tabulates the relationship between the statistic and the generated code amount. .

When the quantization step size calculation unit 22 tabulates the relationship between the statistical amount and the generated code amount, the quantization parameter QP corresponding to the target code amount BT (n) and the statistical amount E (n) is referred to the table. Is identified.
For example, when the relationship shown in FIG. 5 is tabulated, if the target code amount BT (n) is “bt1” and the statistic E (n) is “e1”, QP = 3, 10, 25, Among 150, since QP = 10 is the closest quantization parameter, it is specified that the quantization parameter QP corresponding to the target code amount BT (n) and the statistic E (n) is “10”.
If the target code amount BT (n) is “bt2” and the statistic E (n) is “e2”, QP = 150 is the closest quantization among QP = 3, 10, 25, 150. Since it becomes a parameter, it is specified that the quantization parameter QP corresponding to the target code amount BT (n) and the statistic E (n) is “150”.

When the quantization step size calculation unit 22 specifies the quantization parameter QP corresponding to the target code amount BT (n) and the statistic E (n), the quantization step size calculation unit 22 outputs the quantization parameter QP from the lower bit truncation unit 21. The reduction bit number m is subtracted.
Then, the quantization step size calculation unit 22 calculates the quantization step size Δ corresponding to the quantization parameter QP-m after the subtraction using the above-described equation (2).
When the quantization step size calculation unit 22 calculates the quantization step size Δ corresponding to the quantization parameter QP-m after the subtraction, the quantization step size calculation unit 22 gives the quantization step size Δ to the quantization unit 13.

  As is apparent from the above, according to the third embodiment, the quantization step size calculation unit 22 takes into account the number of bits of the lower bits that have been truncated by the lower bit truncation unit 21 in the immediately preceding stripe. Since the conversion step size Δ is calculated, the rate can be controlled with higher accuracy.

Embodiment 4 FIG.
In the first embodiment, the entropy encoding unit 15 adds the sign bit indicating the sign of the prediction error to the lower bits (see FIG. 7), but all the upper bits are “0”. In this case, if all the lower bits except the least significant bit of the absolute value of the prediction error are “0” and the least significant bit is “1”, the least significant bit is changed to “0”, You may make it reduce the sign bit added to.
That is, in the fourth embodiment, a method for reducing the code amount by avoiding transmission of the sign bit will be described.

In the non-compressed transmission of the lower bits in “JPEG XR”, as shown in FIG. 11A, when all the upper bits to be subjected to Huffman coding are “0” among the absolute values of the prediction errors, uncompressed A sign bit representing the sign of the prediction error is added to the lower bits to be transmitted, and the sign bit is used as a code.
However, when not only the upper bits but all the lower bits are “0”, there is no need to add a sign bit.

Therefore, in the fourth embodiment, as shown in FIG. 11B, all bits except the least significant bit of the absolute value of the prediction error are “0” and the least significant bit is “1”. In this case, as shown in FIG. 11C, the least significant bit is forcibly changed to “0” to avoid the transmission of the sign bit, thereby reducing the code amount.
As described above, by changing the least significant bit to “0”, an encoding error occurs. However, the amount of code reduction can be reduced compared to reducing all the least significant bits.

Specifically, the lower bit truncation unit 16 (or lower bit truncation unit 21) prepares a lower bit code amount table as shown in FIG.
The lower bit code amount table in FIG. 8 shows the correspondence between the number of bits to be reduced and the reduced code amount, but the lower bit code amount table in FIG. 12 shows the reduced code amount when the sign bit is reduced. The amount of code to be reduced when the sign bit is not reduced is listed.
That is, the combination of the number of bits to be reduced and the presence or absence of sign bit reduction is represented by parameter k. For example, in parameter k = 3, the least significant 2 bits are reduced and the sign bit generated in the least significant bit is reduced. In this case, the amount of reduced code is also listed.
The lower bit truncation unit 16 (or lower bit truncation unit 21) applies the lower bit code amount table of FIG. 12 to reduce the bit amount when performing the lower bit truncation processing of FIG. A parameter k that satisfies the desired code amount B−b (k) is calculated, and an appropriate number of reduction bits and whether or not sign bits are reduced are selected.

  As apparent from the above, according to the fourth embodiment, if all the lower bits except the least significant bit of the absolute value of the prediction error are “0” and the least significant bit is “1”, Since the sign bit added to the low-order bits is reduced by changing the low-order bits to “0”, the code amount can be controlled with higher accuracy than truncating the entire bit plane of the uncompressed data. There is an effect that can be done.

  In the fourth embodiment, the example in which the least significant sign bit is reduced from all the HP coefficients in the stripe is shown. However, it is not always necessary to reduce all the sign bits in the stripe, and an arbitrary conversion coefficient is used. Therefore, it is possible to realize rate control with higher accuracy.

  DESCRIPTION OF SYMBOLS 1 encoding process part, 11 stripe division part (image division means), 12 conversion process part (quantization means), 13 quantization part (quantization means), 14 prediction process part (entropy encoding means), 15 entropy code Conversion unit (entropy encoding unit), 16, 21 lower bit truncation unit (lower bit truncation unit), 17 output buffer unit (output buffer unit), 18 target code amount calculation unit (target code amount calculation unit), 19 Statistic calculator (statistics calculator), 20, 22 quantization step size calculator (quantization step size calculator).

Claims (5)

  An image dividing unit that divides the image to be encoded into stripe units, and a conversion process on the image signal of the stripe divided by the image dividing unit, and the conversion coefficient of the image signal is set with a given quantization step size. Quantizing means for quantizing and outputting a quantized value of the transform coefficient; calculating a prediction error of the quantized value output from the quantizing means; and a plurality of bits indicating an absolute value of the predictive error The entropy encoding means for outputting the lower bits without compression and the Huffman encoding for the upper bits and the code data output from the entropy encoding means are temporarily stored, and the above-mentioned code is output at a constant rate. Output buffer means for outputting code data, target code amount calculation means for calculating a target code amount from the code amount of an already encoded stripe, and the above A statistic calculating means for calculating the statistic of the stripes divided by the image dividing means; a target code amount calculated by the target code amount calculating means; and a quantization step size from the statistic calculated by the statistic calculating means. And a quantization step size calculation unit that gives the quantization step size to the quantization unit, and a code amount of code data output from the entropy encoding unit to the output buffer unit is calculated as the target code amount calculation. An image encoding apparatus comprising: a lower bit truncation unit that truncates all or part of lower bits output without being compressed from the entropy encoding unit when the target code amount calculated by the unit is exceeded.   When the code amount of the code data output from the entropy encoding unit to the output buffer unit exceeds the target code amount calculated by the target code amount calculation unit, the low-order bit truncation unit 2. The image coding apparatus according to claim 1, wherein truncation is performed in order from the least significant bit until the code amount falls below the target code amount.   3. The image encoding device according to claim 2, wherein the lower-order bit truncation means truncates the lower-order bits related to a specific macro block among the plurality of macro blocks constituting the stripe.   3. The image code according to claim 2, wherein the quantization step size calculating means calculates the quantization step size in consideration of the number of bits of the lower bits cut off in the immediately preceding stripe by the lower bit truncation means. Device.   When the sign bit indicating the sign of the prediction error is added to the lower bit by the entropy coding unit, the lower bit truncation means is a plurality of bits indicating the absolute value of the prediction error, except for the least significant bit. If the lower bit is “0” and the least significant bit is “1”, the least significant bit is changed to “0” to reduce the sign bit added to the lower bit The image encoding device according to claim 2.

Images