JP4293175B2 - Code amount control method - Google Patents

Description

  The present invention relates to a high-efficiency encoding method for use in high-efficiency encoding and recording / transmission of an image signal. The present invention relates to a code that performs high-speed control to suppress a generated code amount in a specific code amount control unit in an image to a certain level or less. The present invention relates to a quantity control method.

With the recent development of digital compression technology, digital VTRs that record video signals with high-efficiency encoding have become widespread. As a feature of digital VTR,
・ High image quality ・ Resistant to errors ・ Special reproduction is possible, and sufficient image quality can be realized. In order to satisfy these conditions, in the above-described digital VTR, there is a code amount control method in which the entire image is divided into specific small code amount control units, and the code amount generated in this control unit falls within the specific code amount. It has been adopted. The DV format (IEC-61834) is an example of a format that employs this method.

  In the DV format, code amount control is performed in units of segments as code amount control units. Hereinafter, the configuration of the segments will be described in detail with reference to FIG. As shown in the figure, one frame of data is divided into DCT blocks composed of 8 × 8 pixels, which are basic units of DCT calculation. Since the luminance signal and the color difference signal have different sampling frequencies, the luminance signal has four DCT blocks, and the color difference signal has one DCT block in the same position and the same area on the screen. A group of these six DCT blocks is called a macro block. In the figure, a macro block 600 is composed of a total of six DCT blocks, namely, DCT blocks 610, 611, 612, and 613 for luminance signals and DCT blocks 614 and 615 for color difference signals.

  In encoding, fixed-length encoding is performed so that one frame of data does not exceed a specific number of bits. In practice, however, code amount control is performed so that a fixed code amount is obtained in units of five macroblocks. These five macroblocks are called segments.

  When the control is performed so that the code amount is fixed on a segment basis, it is desirable that the information amount be uniform among the segments. That is, when there is a situation where the amount of information is extremely small in a certain segment and the amount of information is extremely large in a certain segment, the segment with a small amount of information is highly reproducible, that is, the compression distortion is not noticeable, but the amount of information is large. This is because the segment has a low reproducibility, that is, a phenomenon in which compression distortion is conspicuous.

  The information amount of the image signal is not uniformly distributed on the screen, but a region with a large amount of information and a region with a small amount of information are mixed locally. Therefore, it is possible to equalize the amount of information between segments by configuring segments from macroblocks that are separated on the screen as shown in FIG. 8 instead of macroblocks that are continuous on the screen.

  As shown in the figure, a segment 605 is composed of five macroblocks located at positions on the screen of macroblocks 600, 601, 602, 603, and 604. Such a process is called a segment structure by shuffling. On the other hand, as shown in FIG. 6, configuring a segment from macroblocks continuous on the screen is called a segment configuration in raster order.

  Next, a high-efficiency encoding apparatus using a conventional code amount control method will be described with reference to FIG. In FIG. 7, an image signal of one frame (or one field) input from the input unit 500 is input to the segment configuration unit 501. The segment construction unit 501 divides the input image into DCT blocks each consisting of 8 × 8 pixels, collects 6 DCT blocks at the same position on the screen to form a macro block, and further collects 5 macro blocks. Thus, a segment which is a code amount control unit is configured. Each segment is input to the orthogonal transform calculation unit 502.

  The orthogonal transform calculation unit 502 performs orthogonal transform on the input segment in units of DCT blocks and outputs orthogonal transform data. The orthogonal transform data is simultaneously transmitted to 16 encoding units, that is, the first encoding unit 503, the second encoding unit 504, the third encoding unit 505,..., And the sixteenth encoding unit 506. Is input.

  Control of the amount of generated code for a segment is performed in the quantization process. Quantization is a process of dividing each coefficient of an orthogonally transformed DCT block by an integer value called a quantization step. The larger the quantization step, the smaller the value of the DCT coefficient after quantization. In particular, in the high frequency range, the amount of generated codes is reduced by increasing the coefficient that is 0. Conversely, the smaller the quantization step, the larger the generated code amount.

  In the conventional high-efficiency encoding device, each of the 16 encoding units has a quantizer for performing quantization processing. The 16 quantizers hold different quantization steps in advance. Each quantizer is given a number for identifying a quantizer called a quantization number. The smaller the quantization number, the larger the quantization step, that is, the smaller the generated code amount. On the other hand, the larger the quantization number, the smaller the quantization step, that is, the larger the generated code amount. In this conventional high-efficiency encoding device, the encoding unit number and the quantization number are equal. For example, the quantization number of the quantizer in the first encoding unit is 1.

  Each encoding unit performs code amount calculation using the quantizer. The first encoding unit 503 quantizes the input orthogonal transform data with a quantizer of quantization number 1 and further performs variable length encoding to calculate the code amount of the block. Similar processing is performed on 30 blocks in the segment, and the total code amount in the segment is calculated. The total code amount calculated by the first encoding unit 503 is input to the final quantizer determining unit 507. Here, in each quantizer, the larger the quantization number, the smaller the quantization step. That is, the larger the quantization number, the larger the generated code amount. Similarly, the other encoding units calculate the generated code amount and input the calculated total code amount to the final quantizer determination unit 507.

  The final quantizer determination unit 507 outputs the quantization number corresponding to the maximum code amount not exceeding the specific target code amount to be given to the segment to the generated code amount by 16 types of quantization thus obtained. The serial number is input to the encoding unit 508.

  The encoding unit 508 quantizes the orthogonal transform data input from the orthogonal transform calculation unit 502 with a quantizer (final quantizer) corresponding to the final quantization number input from the final quantizer determination unit 507. Further, variable length encoding is performed, and encoded data of the segment is output. The encoded data is output to the output unit 509. The output unit 509 outputs the input encoded data to a recording / transmission medium or the like.

  As described above, the conventional high-efficiency encoding apparatus realizes the process of obtaining the optimal quantization number by hardware by simultaneously performing quantization in the quantization steps corresponding to a plurality of different quantization numbers. (For example, refer to Patent Document 1).

  Along with the improvement in CPU performance, non-linear editing equipment that performs high-efficiency encoding with software on a PC has attracted attention. When performing encoding processing with software, unlike devices that perform encoding processing with hardware as described above, parallel processing is limited, so it is difficult to perform multiple quantization processing at the same time. In order to realize this, it is necessary to reduce the number of operations.

  A case where the code amount control processing in the conventional example shown in FIG. 7 is realized by software processing will be described. In order to realize the code amount control processing by software processing, quantization is performed sequentially in the quantization steps corresponding to a plurality of different quantization numbers, and the quantization number that can obtain the optimum code amount from the obtained result is obtained. The process of obtaining is performed. Hereinafter, this process will be specifically described with reference to a flowchart. FIG. 9 shows a flowchart for determining the optimum quantization number for controlling the amount of generated code when encoding the segment n to be encoded. The function calc_code (q_a) is a function that quantizes the segment in the quantization step corresponding to the quantization number q_a and further performs variable length coding. Note that the quantization number q_a given as a parameter when performing quantization and encoding processing while changing the quantization number is hereinafter referred to as a reference quantization number. The function calc_max (code_a) is the maximum code that does not exceed the target code amount among the generated code amounts code_a [i] (i = 1, 2,..., 16) calculated for each of the 16 quantization numbers. This is a function for calculating a quantization number giving a quantity.

In step 700, the reference quantization number q_a is initialized with 1. In step 701, the segment n to be encoded is quantized with the reference quantization number q_a and further variable-length encoded, and the generated code amount at that time is stored in code_a [q_a]. In step 702, if q_a is smaller than 16, the process proceeds to step 703, and if q_a is 16 or more, the process proceeds to step 704. When the process proceeds to step 703, the value of q_a is incremented by 1, and the process proceeds to step 701. When the process proceeds to step 704, the function calc_max (code_a) substitutes the maximum quantization number that does not exceed the target code amount, that is, the final quantization number, into q_n, and ends the processing of the segment n.
JP-A-6-125537

  However, the conventional code amount control method using software calculates the generated code amount by performing brute force encoding for a plurality of quantization numbers, and cannot be processed in parallel, so it takes a lot of computation time. there were.

  The present invention has been made in view of such problems, and an object thereof is to reduce the amount of calculation in code amount control by software.

In order to solve the above problem, the present invention collects a plurality of blocks at distant positions on the screen to form a segment, and selects one of the parameters related to quantization consisting of a plurality of stages for each segment. A code amount control method for controlling a code amount generated by encoding, wherein an initialization step uses a quantization parameter applied to an already encoded segment on the screen as an initial value of a quantization parameter When the sign-calculating step of calculating a generated code amount by performing a quantization on the segment of interest by using the quantization parameter, whether to adopt the generation sign-amount as a final sign-amount determination marks having a termination determination step, and a parameter changing step to change the quantization parameter so that the generated sign-amount approaches the target sign-amount It is obtained by the amount control method.

  When a segment is formed by collecting blocks located at distant locations on the screen, there are areas with a large amount of information including edges and textures, and areas with a small amount of information such as flat areas. The amount of information at is equalized. Therefore, the final quantization number of an arbitrary segment is expected to take a value close to the final quantization number of the surrounding segments. In the present invention, with respect to the encoding target segment, the final quantization number of the encoded segment is used as the initial value of the reference quantization number, and a search is performed from the periphery to obtain the maximum code amount not exceeding the target code amount. When the final quantization number, which is the quantization number, is found, there is an advantage that the calculation amount necessary for calculating the final quantization number can be greatly reduced by terminating the processing in the segment.

  Embodiments of the present invention will be described with reference to the drawings. The present invention is a code amount control method in an encoding method in which blocks located at distant positions on a screen are assembled to form a segment and an encoding process is performed in units of segments, such as the DV format.

  FIG. 1 shows a hardware configuration diagram for executing the present code amount control method. This code amount control method is realized by the CPU 102 executing a program stored in the program storage unit 104 in advance. One frame (or one field) image signal input from the video input means 100 is input to and stored in the memory 101. Based on the request command input from the input unit 103 such as a mouse or a keyboard, the CPU 102 executes the program stored in the program storage unit 104. The CPU 102 performs encoding processing including code amount control on the video signal stored in the memory according to the program, generates a processed image, and temporarily stores the processed image in the memory 101. The video output unit 105 transmits the processed image stored in the memory 101 to the video display unit 106. The video display means 106 displays the input processed image on a display device such as a monitor.

  FIG. 2 is a flowchart showing an algorithm of the code amount control method of the present invention. In the figure, the final quantization number of segment n is denoted q_ [n]. According to this notation, the final quantization number of the (n−1) -th segment is q_ [n−1]. The function calc_code (q_a) is a function that quantizes the segment in a quantization step corresponding to the reference quantization number q_a and further performs variable length coding, and outputs a generated code amount code_t. The reference quantization number q_a takes a value from 1 to 16. Further, the larger the quantization number, the smaller the corresponding quantization step, and as a result, the generated code amount increases. That is, if quantization and encoding are performed with the function calc_code () while increasing the reference quantization number, the generated code amount also increases. code_L is the target code amount per segment. “over_flag” is a flag indicating that the generated code amount exceeds the target code amount in the process of calculating the final quantization number of the encoded segment n. under_flag is a flag indicating that the generated code amount does not exceed the target code amount in the process of calculating the final quantization number of the encoded segment. q_c is a quantization number that is a candidate for the final quantization number.

  The embodiment of the present invention will be specifically described below.

  In step 150, the final quantization number q_ [n-1] of the already encoded segment (n-1) is substituted as the initial value of the reference quantization number q_a of the segment n to be encoded.

  Two over_flags and under_flags indicating the state of the generated code amount of the segment n are defined. If the segment has exceeded the target code amount even once, the value of over_flag is set to 1. Conversely, if the target code amount has never been exceeded, over_flag is set to 0. Also, if the segment has never reached the target code amount even once, the value of under_flag is set to 1. Conversely, if the code amount never falls below the target code amount, under_flag is set to 0.

  Both over_flag and under_flag are initialized to 0. q_c is initialized with 1.

  In step 151, the segment n is quantized in the quantization step corresponding to the reference quantization number q_a, further variable-length coded, and the generated code amount code_t at that time is output. In step 152, the generated code amount code_t and the target code amount code_L are compared. If code_t does not exceed code_L, the process proceeds to step 153, and if code_t exceeds code_L, the process proceeds to step 157.

  When the process proceeds to step 153, q_t becomes a candidate for the final quantizer, and is substituted for q_c.

  In step 154, the value of over_flag is checked, and over_flag == 1, that is, in the process of calculating the final quantizer of the encoded segment, when the generated code amount exceeds the target code amount, or when q_a is 16, Transition to step 155. If it is not over_flag == 1 and q_a == 16, the process proceeds to step 156.

  When the process proceeds to step 156, under_flag = 1 is set, the value of q_a is increased by 1, and the process proceeds to step 151.

  When transitioning from step 152 to step 157, the value of under_flag is checked, and under_flag == 1, that is, in the process of calculating the final quantization number of the encoded segment, if the generated code amount does not exceed the target code amount, Alternatively, when q_a == 1, the process proceeds to step 155. As a result, if under_flag == 1 is not satisfied and q_a == 1 is not satisfied, the process proceeds to step 158. If under_flag == 1 and q_a == 1, q_c is never updated, so the initial value 1 is the final quantization number.

  When the process proceeds to step 158, the value of over_flag is set to 1, the value of q_a is decreased by 1, and the process proceeds to step 151.

  In step 155, q_c is substituted into q_n as a final quantization number which is a quantization number for obtaining the maximum code amount not exceeding the target code amount, and the processing of the segment n is completed.

  The process of determining the final quantization number of the encoding target segment n will be described with a specific example.

  FIG. 3 shows an example of the final quantization number determination process for the encoding target segment n. In the figure, the horizontal axis indicates the quantization number, and the vertical axis indicates the generated code amount when the encoding target segment n is encoded with each quantization number. 200 represents a generated code amount at quantization number 9, 201 represents a generated code amount at quantization number 10, 202 represents a generated code amount at quantization number 11, 203 represents a generated code amount at quantization number 12, The bits are 1000, 2000, 3000, and 4000 bits, respectively. Reference numeral 204 denotes a target code amount, which is 2500 bits.

  First, as the initial value of the reference quantization number, 10 that is the quantization number applied in the encoding of the segment (n−1) is set, and the first code amount calculation is performed. As a result of this calculation, a code amount of 2000 bits is calculated. At this time, since the generated code amount at the quantization number 10 is smaller than the target code amount, the quantization number is incremented by 1 so that the generated code amount is increased in the next code amount calculation, and the code amount calculation is performed again. That is, the quantization number is set to under_flag = 1, and the value of q_a is increased by 1 to 11. Further, q_c = 10.

  Subsequently, a code amount calculation of the second time is performed with the reference quantization number as 11, and a code amount of 3000 bits is calculated. At this time, the generated code amount by the quantization number 11 is larger than the target code amount, and under_flag = 1. That is, although it was less than the target code amount last time, it exceeded the target code amount in this calculation, so the final quantization number for the segment is set to 10. In this case, the number of code amount calculations is two, and a large amount of calculation can be realized as compared with the conventional example that requires 16 calculation times.

  FIG. 4 shows another example of the process of determining the final quantization number for the encoding target segment n. In FIG. 3, as in FIG. 3, the horizontal axis indicates the quantization number, and the vertical axis indicates the generated code amount when the encoding target segment n is encoded with each quantization number. 300 indicates a generated code amount at quantization number 9, 301 indicates a generated code amount at quantization number 10, 302 indicates a generated code amount at quantization number 11, 303 indicates a generated code amount at quantization number 12, The bits are 1000, 2000, 3000, and 4000 bits, respectively. Reference numeral 304 denotes a target code amount, which is 2500 bits.

  First, 12 which is the quantization number applied in the encoding of the segment (n−1) is set as the initial value of the reference quantization number, and the first code amount calculation is performed. As a result of this calculation, a 4000-bit code amount is calculated. At this time, since the generated code amount at the quantization number 12 is larger than the target code amount, the quantization number is decreased by 1 so that the generated code amount is reduced in the next code amount calculation, and the code amount calculation is performed again. That is, over_flag = 1, and the value of q_a is decreased by 1 to 11.

  Subsequently, a code amount calculation of the second time is performed with the reference quantization number as 11, and a code amount of 3000 bits is calculated. At this time, since the generated code amount by the quantization number 11 is larger than the target code amount and under_flag = 0, the value of q_a is similarly decreased by 1 to 10.

  Subsequently, the third code amount calculation is performed with the reference quantization number as 10, and a 2000 bit code amount is calculated. At this time, the generated code amount by the quantization number 10 is smaller than the target code amount, and over_flag = 1. That is, the previous code amount exceeded the target code amount, but in the current calculation, the target code amount was less than the target code amount. In this case, the number of code amount calculations is three, and a large amount of calculation can be realized as compared with the conventional example requiring 16 calculation times.

  The code amount control method of the present invention is premised on being applied to an encoding process in a recording format in which blocks located at distant positions on the screen are collected to form a segment and the code amount control is performed in units of segments. It is.

  When a segment is formed by collecting blocks located at distant locations on the screen, there are areas with a large amount of information including edges and textures, and areas with a small amount of information such as flat areas. The amount of information at is equalized. The code amount control method according to the present invention uses the quantization number of the encoded segment as the initial value of the reference quantization number for the encoding target segment, searches from the surroundings, and finds the final quantization number. At this point, the processing in the segment is aborted, so that it is possible to greatly reduce the amount of calculation necessary for calculating the final quantization number.

  The segment configuration will be described with reference to FIG. Reference numerals 400, 401, 402, 403, and 404 denote macroblocks each including eight DCT blocks. As shown in the figure, five macroblocks located at distant positions on the screen are collected to constitute segment 1 (450). Similarly, 410, 411, 412, 413, and 414 represent macroblocks, and these are collected to constitute segment 2 (451).

  In segment 1, areas 400 and 404 are flat areas with a small amount of information, and areas 401, 402, and 403 are areas with a large amount of information including edges. In segment 2, areas 410 and 414 are flat areas with a small amount of information, and areas 411, 412, and 413 are areas with a large amount of information including edges. The amount of information of segment 1 and segment 2 is equalized, and the final quantization numbers of these two segments are expected to be the same or close. Therefore, when encoding the segment 2, the final quantization number of the segment 1 is set as the initial value of the reference quantization number, and the search is performed from the periphery thereof, thereby calculating the amount of computation necessary for calculating the final quantization number of the segment 2 Can be greatly reduced.

  In this embodiment, when the value of q_a is updated, it is increased (decreased) by one. However, a method of controlling the magnitude of increase / decrease according to the difference between the generated code amount and the target code amount is also conceivable. For example, when the difference between the generated code amount and the target code amount is larger than a certain threshold, the number of operations until the final quantization number is calculated can be further reduced by increasing (decreasing) by 2 or increasing (decrease) by 3. Is possible.

  According to the code amount control method of the present invention, since the amount of calculation can be reduced in the code amount control by software, it can also be applied to a non-linear editing device or the like that performs high-efficiency encoding by software.

The figure which showed the hardware constitutions of the code amount control method of this invention The flowchart which showed the code amount control method of this invention Diagram showing the operation when the predicted quantization number matches the actual quantization number Diagram showing the operation when the predicted quantization number is different from the actual quantization number Diagram showing segment structure by shuffling Diagram showing segment structure in raster order Block diagram of a conventional code amount control device Diagram showing macro block and segment structure A flowchart showing a conventional code amount control method

Explanation of symbols

100 video input means 101 memory 102 CPU
103 Input means 104 Program storage means 105 Video output means 106 Video display means 400, 401, 402, 403, 404, 410, 411, 412, 413, 414, 600, 601, 602, 603, 604 Macroblock 450 Segment 1
451 Segment 2
500 Input unit 501 Segment configuration unit 502 Orthogonal transform operation unit 503 First encoding unit 504 Second encoding unit 505 Third encoding unit 506 Sixteenth encoding unit 507 Final quantizer determining unit 508 encoding Conversion unit 509 Output unit 610, 611, 612, 613 Luminance DCT block belonging to the same macroblock 614, 615 Color difference DCT block belonging to the same macroblock

Claims (5)

A code that controls the amount of code that is generated by collecting multiple blocks located at distant locations on the screen to form a segment , selecting one of the parameters related to quantization in multiple stages , and performing encoding A quantity control method,
An initialization step in which a parameter relating to quantization applied to an already encoded segment on the screen is an initial value of the quantization parameter ;
A code amount calculation step of calculating a generated code amount by performing quantization on a target segment using the quantization parameter ;
An end determination step of determining whether to employ the generated code amount as a final code amount;
A parameter changing step said generated code amount to change the quantization parameter so as to approach the target code amount,
A code amount control method comprising: The code amount control method according to claim 1, wherein the parameter changing step changes the quantization parameter step by step so that the generated code amount approaches a target code amount. The parameter changing step changes the quantization parameter by one step so that the code amount decreases when the generated code amount exceeds the target code amount, and the code amount when the generated code amount is equal to or less than the target code amount. The code amount control method according to claim 2, wherein the quantization parameter is changed by one step so as to increase. The code amount control method according to claim 1, wherein the segment is configured by shuffling. 2. The termination determination step employs the generated code amount as a final code amount when the generated code amount for the selected quantization parameter is a maximum code amount not exceeding the target code amount. Code amount control method.

Images