JP2015103982A - Image encoding device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image encoding device including a quantization part capable of adaptively allocating a QP index in accordance with characteristics of an image for each macro block without storing the QP index in a large storage area that is expanded in proportion to an image size.SOLUTION: The image encoding device includes a color space conversion part, a frequency conversion part, a quantization part 10, a coefficient prediction part, an adaptive scan part, a VLC encoding part and a bit stream output part. The quantization part 10 is formed from: an adaptive QP index selection portion 20 for adaptively selecting one QP index (i) on the basis of coefficient data in a neighboring block; a quantization parameter selection part 30 for selecting a quantization parameter Q(i) corresponding to the QP index (i); a quantization arithmetic processing part 40 for quantizing the coefficient data in the block by using the quantization parameter Q(i); and a block counter part 50 for counting the number of blocks.

Description

  The present invention relates to an image encoding apparatus that compresses and encodes image data.

  2. Description of the Related Art Conventionally, there has been known an image encoding device that compresses and encodes image data for the purpose of reducing a load or the like in communication. In recent years, a standard for image coding with higher compression efficiency has been proposed (see Non-Patent Document 1).

  FIG. 13 shows a schematic configuration of an image encoding device according to the standard. As shown in FIG. 13, the image encoding apparatus 100 includes a color space conversion unit 110, a frequency conversion unit 120, a quantization unit 130, a coefficient prediction unit 140, an adaptive scan unit 150, a VLC encoding unit 160, bits The stream output unit 170 is configured. Note that VLC in the name of the VLC encoding unit 160 is an abbreviation for Variable Length Coding (Variable Length Coding).

  The image data is composed of pixels (pixels) which are the minimum unit of an image, and a 256-pixel image area divided by 16 pixels (horizontal direction) × 16 pixels (vertical direction) is regarded as one unit, and this is a macroblock. A unit of 4 pixels × 4 pixels obtained by dividing this macroblock into four parts in the horizontal direction and the vertical direction is called a block. In the image processing according to the standard, the processing proceeds in units of the macroblock and the block.

  The color space conversion unit 110 performs color space conversion processing on the input image data for each pixel in units of macroblocks. The color space conversion process depends on the pixel format defined in the standard of Non-Patent Document 1. When the pixel format requires color space conversion, the original image data in the RGB format is converted into the YUV format, and the original image data in the CMYK format is converted into the YUVK format. The Further, when the pixel format does not require color space conversion, the original image data is processed so as not to be color space converted. Then, the color space conversion unit 110 inputs the processed image data to the frequency conversion unit 120.

  The frequency conversion unit 120 converts the image data that has been color space processed by the color space conversion unit 110 into a raster order in units of macroblocks, and within each macroblock, in the order of the color components after the color space conversion in each macroblock. For each block, processing (frequency conversion) for converting the data of each pixel into coefficient data for each frequency band is performed. The frequency conversion unit 120 performs frequency conversion on the 16 pixel data of 4 × 4 pixels in each block and converts the pixel data into coefficient data, and then removes the remaining data except the coefficient data on the upper left of the 16 blocks. Fifteen coefficient data are collected to form first layer data. This is HP coefficient (high pass coefficient) data which is a high frequency component. Next, the coefficient data at the upper left of each of the 16 blocks are collected to generate 16 coefficient data of 4 × 4, and the collected 4 × 4 coefficient data is further frequency-converted to obtain 4 × 4 The second hierarchical data is formed by the remaining 15 coefficient data except the one coefficient data at the upper left of the 4 × 4 coefficient data subjected to frequency conversion, and after the second frequency conversion. Among the 4 × 4 coefficient data, the third hierarchical data is formed by one coefficient data at the upper left. The second layer data corresponds to LP coefficient (low-pass coefficient) data that is a low frequency component, and the third layer data corresponds to DC coefficient (DC coefficient) data that is a direct current component. As described above, the frequency conversion unit 120 generates the first layer data (HP coefficient), the second layer data (LP coefficient), and the third layer data (DC coefficient), and each of these layer data is quantized. Input to the unit 130.

  The quantization unit 130 divides the first layer data, the second layer data, and the third layer data after the frequency conversion input from the frequency conversion unit 120 by the selected quantization parameter, respectively. The first layer data, the second layer data, and the third layer data after processing are input to the coefficient prediction unit 140.

  The quantization unit 130 includes a QP index acquisition / storage unit 111, a quantization parameter selection unit 112, a quantization operation unit 113, and a block counter unit 114, as shown in FIG. Hereinafter, a description will be given with reference to a processing flow centering on the quantization processing unit 130 shown in FIG. 15 and specific setting examples of the QP index shown in FIGS.

  In the standard of Non-Patent Document 1, quantization processing can be performed using a parameter called a QP index. The QP index designates one of 16 parameters called a QP index composed of numerical values of 0 to 15 for each macro block of 16 pixels × 16 pixels, and a quantum corresponding to the designated QP index. By using the quantization parameter, the frequency-converted LP coefficient data and HP coefficient data included in the macroblock are divided and quantized.

  16 and 17 show specific setting examples of the QP index. FIG. 16 shows an example of setting a quantization parameter with a QP index of 0 to 7, and FIG. 17 shows an example of setting of a quantization parameter with a QP index of 8 to 15. As for the DC coefficient, since only one type of quantization parameter can be taken, there is no designation of the QP index. However, assuming that the QP index corresponds to 0, the first color component to the fourth color are assigned to the portion where the QP index is 0. 1 to 4 quantization parameters are specified for the color components. According to the example shown in FIGS. 16 and 17, for example, for the first color component of the LP coefficient, the QP parameter can be designated as 10 to 25 corresponding to the QP indexes 0 to 15, and LP For the second color component, the third color component, and the fourth color component, the QP parameters are designated as 30 to 45, 50 to 65, and 70 to 85, corresponding to the QP indexes 0 to 15, respectively. Can do. The same applies to the HP coefficients, and different values or the same values can be designated for the quantization parameters corresponding to the QP indexes 0 to 15. An integer in the range of 0 to 255 can be specified as the quantization parameter. The setting examples shown in FIGS. 16 and 17 are examples of designation, and other values may be designated for the quantization parameter corresponding to the QP index. As an example of setting, an example is shown in which quantization parameters are specified for all of the first color component to the fourth color component and the QP indexes 0 to 15. However, the number of color components is determined by the original image format. The number of QP indexes to be used can be specified in the range of 1 to 16 depending on the setting at the time of compression, and can be increased or decreased depending on the intended use and purpose.

  FIG. 18 shows an example in which an image is divided into 16 × 16 pixel macroblock unit areas and a QP index is designated for each area. A rectangular area delimited by a thin line is a macro block, an area inside the image delimited by a thick line is area 1, and an area outside the image delimited by a thick line is area 2. For example, assuming that the QP index designated for the region 1 is 0 and the QP index designated for the region 2 is 1, the quantization parameter corresponding to each QP index can be designated. In general, there is an important and fine image at the center of the image, and there are often less important images at the periphery. For example, the QP index specified for the area 1 in the center of the image corresponds to 0. The value of the quantization parameter to be set is a small value, and the compression is reduced to a delicate image, and the value of the quantization parameter corresponding to the QP index set to 1 in the peripheral region 2 is set to a large value for high compression. By doing so, a rough image can be obtained.

  FIG. 19 shows another example in which an image is divided into macro block units and a QP index is designated for each region. The left area is the area 1 and the right area is the area 2 in the central part of the image separated by the thick line, the upper area of the image separated by the thick line is the area 3, and the lower area of the image separated by the thick line is the area. 4, 0 to 5 are assigned to the area 1 as QP indexes, 6 to 11 are assigned to the area 2, 12 to 13 are assigned to the area 3, and 14 to 13 are assigned to the area 4. To 15 are assigned. As in FIG. 18, for example, the quantization parameter value corresponding to the QP index specified in the central region 1 and the region 2 corresponding to 0 to 11 is reduced to a low value to obtain a delicate image, and the peripheral portion. By setting the quantization parameter value corresponding to 12 to 15 in the QP index set in the regions 3 and 4 to a high value and performing high compression, a rough image can be obtained.

  The operation of the quantization process using the QP index that can be designated as described above will be described with reference to FIGS.

  The quantization parameter selection unit 112 acquires and stores a quantization parameter corresponding to each QP index from an external controller or the like before starting the compression process (step S21). At the start of the compression process, the block counter unit 114 initializes the macroblock count value m, the color component count value c, and the block count value n to 1 (step S22), and is input as the compression process proceeds. The count values of m, c, and n are updated while counting the number of coefficient data to be obtained (steps S28, S30, and S32).

  The frequency conversion unit 120 performs frequency conversion on the image data of the first macroblock (count value m is 1) to generate first layer data (HP coefficient), second layer data (LP coefficient), and third layer data ( Coefficient data consisting of DC coefficients is generated and input to the quantization unit 130 (step 23). Subsequent processing will be described for the first layer data (HP coefficient). For the second layer data (LP coefficient), the number of blocks in the macroblock is set to 1, and the same processing is performed. However, for the third layer data (DC coefficient), the QP index is not used for the quantization process. QP acquisition is not necessary.

  The QP index acquisition / storage unit 111 outputs a data request signal for the QP index corresponding to the first macroblock (count value m is 1), acquires the QP index from the data storage unit, and stores it (step S24). ). Note that the value of the QP index is specified for each macroblock, and as the image size increases, the number of macroblocks included in the image data also increases, and the number of QP indexes corresponding to each macroblock also increases. Generally, a QP index corresponding to each macroblock is held in an external or internal data storage area.

  The quantization parameter selection unit 112 refers to the QP index i corresponding to the first macroblock (count value m is 1) stored in the QP index acquisition / storage unit 111 and performs quantization corresponding to the QP index i. The parameter Q (i) is selected from the quantization parameters acquired before starting the compression process, and the selected quantization parameter Q (i) is input to the quantization operation unit 113 (step S25).

  The quantization operation unit 113 receives the first block (count value n is 1) of the first color component (count value c is 1) of the first macroblock (count value m is 1) input from the frequency converter 120. ) Is divided by the QP parameter Q (i) acquired from the quantization parameter selection unit 112 to generate quantization data, and the generated quantization data is input to the coefficient prediction unit 140 (Step S26).

  If it is not the last block (counter value n is 15) (step S27), the block counter unit 114 counts up the count value n by 1 (step S28). Similarly, the quantization operation unit 113 determines the first block. Quantization processing of the next block is performed using the same quantization parameter Q (i) (step S26). Then, the count value n is sequentially counted up, and the quantization process for each block is performed. After the quantization processing of the block, if the processed block is the final block, the block counter unit 114 checks whether the color component is the final block (step S29).

  If it is not the final color component, the block counter 114 initializes the block count value n to 1, increments the color component count value c by 1 (step S30), and then selects the quantization parameter. The unit 112 selects the quantization parameter Q (i) corresponding to the QP index i of the first macroblock (count value m is 1) in the current color component (color component corresponding to the count value c) (step S25). ) The quantization operation unit 113 performs the quantization process on the coefficient data for the number of blocks using the selected quantization parameter (step S26). Similarly, each time the quantization process is advanced for each block, and the quantization process for the final block is completed, the block counter unit 114 checks the count value c of the color component, and the count value c is the final value c. If it is not the color component, the count value c is counted up, and if it is the final color component, it is confirmed whether the count value of the macroblock is the final one (step S31).

  If it is not the final macro block, the block counter unit 114 initializes the count value n of the block and the count value c of the color component to 1, and increments the count value m of the macro block by 1 (step S32). ), The frequency conversion unit 120 converts the frequency of the image data of the macroblock into coefficient data in units of blocks (step S23), and the QP index acquisition / storage unit 111 receives a QP index from the data storage unit. (Step S24), the quantization parameter selection unit 112 selects a quantization parameter Q (i) corresponding to the acquired QP index i (step S25), and the quantization calculation unit 113 The coefficient data generated by the conversion unit 120 is converted by the quantization parameter selection unit 112. Is divided by the selected quantization parameter Q (i), it generates quantized data, and outputs to the coefficient prediction unit 140. When the quantization process in units of macroblocks proceeds and the processed macroblock becomes the final macroblock, the quantization unit 130 ends the series of quantization processes. Whether or not it is the last macroblock is determined based on whether or not the counter value m of the macroblock matches (horizontal image size / 16) × (vertical image size / 16).

  As described above, the quantization unit 130 updates the macroblock, color component, and block count value in the block counter unit 114 while the QP index acquisition / storage unit 111 updates each macroblock. The quantization parameter selection unit 112 selects a quantization parameter from the acquired quantization index, and the quantization operation unit 113 selects coefficient data in the current macroblock. The quantization process of dividing by the quantization parameter to generate quantized data is repeated for the number of macroblocks. Then, a QP index that is a basis for selecting a quantization parameter is acquired for each macroblock from the data storage unit.

  The coefficient prediction unit 140 compares coefficient values between adjacent blocks for the first layer data, second layer data, and third layer data for each block input from the quantization unit 130. Thus, a prediction mode is calculated, and a process of subtracting between blocks having high correlation in the direction of the derived prediction mode is performed based on a predetermined arithmetic expression. Then, the coefficient prediction unit 140 inputs the processed first layer data, second layer data, and third layer data to the adaptive scan unit 150 for each block.

  The adaptive scan unit 150 performs a process of replacing the coefficient data for each block with respect to the first layer data and the second layer data input from the coefficient prediction unit 140. In this replacement process, non-zero coefficient data is collected on the side where the pixel position in the block after frequency conversion is small (low frequency side), and the zero value is continuous on the side where the pixel position in the block is large (high frequency side). Since coefficient data is collected, the compression rate of encoding can be increased. Then, the adaptive scan unit 150 inputs the first layer data, the second layer data, and the third layer data on which the replacement process is not performed to the VLC encoding unit 160.

  The VLC encoder 160 performs entropy encoding on the first layer data, the second layer data, and the third layer data processed by the adaptive scan unit 150 to generate a code stream. The code stream is input to the bit stream output unit 170.

  The bit stream 170 sequentially concatenates the code stream encoded by the VLC encoding unit 160 to the image setting data as header information, and collects one file as a bit stream for one image data. Create and output the formed file to the outside.

JPEG XR Recommendation ITU-T T.832: Information technology -JPEG XR image coding system-Image coding specification. [Online] .Recommendation T.832 (01/12). Approval 2012-01-13. Status: In force. 2012-9-14, ITU (International Telecommunication Union). [Retrieved on 2013-10-29]. Retrieved from the Internet: <URL: http://www.itu.int/rec/T-REC-T.832 -201201-I / en>.

  By the way, in the conventional image encoding device shown in FIG. 14, since the QP index for each macroblock is stored in advance, it is necessary to include a data storage unit having a storage area that increases in proportion to the image size. Every time the frequency conversion of the block is completed, it is necessary to obtain a QP index from the data storage unit. For this reason, a large storage area is required as the image size increases, and there is a problem that access to the data storage unit is necessary for data transfer of the QP index every time the macroblock is processed.

  In addition, the quantization in the image compression using the QP index requires that a QP index value be determined in advance for each macro block. For example, as described above with reference to FIGS. A QP index is specified such that a QP index is assigned to each area divided by a plurality of macroblocks in units of. For this reason, the QP index can be assigned only for each segment of the image area, and the QP index cannot be adaptively assigned during the encoding process according to the characteristics of the image for each macroblock. there were.

  The present invention has been made in view of the above circumstances, and it is possible to adaptively assign a QP index in accordance with the characteristics of an image for each macroblock without storing the QP index in a large storage area. An object of the present invention is to provide an image encoding device including a quantization unit.

In order to solve the above problems, an image encoding device of the invention provides:
The image data is frequency-converted for a macroblock consisting of m blocks in each horizontal and vertical direction in units of a block consisting of n pixels in each horizontal and vertical direction, and the upper left pixel position of each block in the macroblock The data corresponding to the pixel position other than the above is used as the coefficient data of the high frequency component, the data corresponding to the upper left pixel position of each block in the macro block is collected, further frequency-converted, and the upper left pixel position after the frequency conversion. The frequency conversion unit which makes the corresponding data the coefficient data of the direct current component, and the data corresponding to the pixel position other than the upper left the coefficient data of the low frequency component,
A quantization unit that quantizes the coefficient data of each component frequency-converted by the frequency conversion unit by dividing the coefficient data by a quantization parameter held in an internal or external storage area to obtain quantized coefficient data;
A coefficient scanning unit that performs a process of changing the order of pixels in the block of quantized coefficient data of the low frequency component and the high frequency component quantized by the quantizing unit, and sets the coefficient data after replacement;
An image coding unit comprising: an encoding unit that encodes the quantized coefficient data of the DC component after quantization by the quantization unit and the post-replacement coefficient data after the replacement process by the coefficient scanning unit to generate encoded data In the device
The quantization unit is
The difference between the DC component coefficient data and the low frequency component coefficient data of the block adjacent to the left and right in the upper block in the macro block and the upper and lower adjacent blocks in the left block, and the low frequency in the adjacent block Each QP that takes the difference between the coefficient data of the components, converts it to an absolute value, takes the average, and designates one quantization parameter by p numerical values from 0 to p−1. A reference value corresponding to an index, which is a reference value held in an external or internal storage area, is compared in order from the smallest reference value, and the difference average value is initially smaller than the reference value. An adaptive QP index selection unit that sets the QP index corresponding to the selected reference value as the selected QP index (i);
A quantization parameter selection unit that selects a quantization parameter Q (i) corresponding to the selected QP index (i);
A quantization operation processing unit that quantizes each coefficient data of a low frequency component and a high frequency component in each block in the macroblock using the selected quantization parameter Q (i).
However, n, m and p are integers of 2 or more, and i is an integer of 0 to p-1.

  According to the image coding apparatus according to the present invention having the above-described configuration, first, image data for one image is input to the frequency conversion unit, and the image data is divided into n in each of the horizontal and vertical directions. Using a block of pixels as a unit, frequency conversion is performed on a macroblock consisting of m blocks in each of the horizontal and vertical directions to obtain coefficient data of a high frequency component, a low frequency component, and a direct current component, and input to the quantization unit Is done.

  Then, in the quantization unit, the coefficient data of each component subjected to frequency conversion is quantized by dividing by the quantization parameter held in the internal or external storage area.

  The quantized coefficient data is input to the coefficient prediction unit. The coefficient prediction unit derives a prediction mode and performs a process of subtracting between highly correlated blocks, and then inputs the coefficient data to the coefficient scan unit. In the scanning unit, processing for changing the arrangement order of the pixels in the block is performed so that non-zero coefficient data is concentrated on the low frequency side in the block, and the coefficient data in the block unit after the replacement is encoded in the encoding unit. And encoded data is generated.

  Then, in the quantization unit, in the adaptive QP index selection unit, the DC component coefficient data and the low-frequency component coefficient of the block adjacent to the left and right in each upper block in the macro block and vertically adjacent in each left block The difference between the data and the difference between the coefficient data of the low frequency components in the adjacent blocks is taken, converted into an absolute value, and the difference average value obtained by taking the average and the reference value corresponding to each QP index, The QP index corresponding to the reference value whose difference average value is smaller than the reference value is set as the selected QP index (i) first, and the quantization parameter selection unit selects the QP index (i). The quantization parameter Q (i) corresponding to the QP index (i) thus selected is selected, and the quantization operation processing unit Te, by using the selected quantization parameter Q (i), the coefficient data of the block unit is divided and quantized.

  As described above, according to the configuration of the present invention, the adaptive QP index selection unit is provided in the quantization unit, and the difference between the coefficient data of the DC component and the low-frequency component in each adjacent block in the upper and left blocks of the macroblock. Since an average value is taken and compared in order from the smallest set reference value corresponding to the QP index, one QP index is selected from a predetermined number of QP indexes. In addition, it is not necessary to provide a data storage unit that stores a QP index that increases in proportion to the image size, and processing for obtaining a QP index from the data storage unit is also unnecessary. In addition, since the difference average value is obtained from the coefficient data of the DC component and the low frequency component in the adjacent block and is compared with the reference value corresponding to the QP index, one QP index is selected. A QP index that matches the characteristics of the image area being processed can be determined, and each coefficient data in a block unit is divided by the quantization parameter corresponding to the QP index, which is suitable for the image characteristics of each area of the image. Quantization processing is possible, and image encoding processing can be optimized.

In addition, the adaptive QP index selection unit of the image encoding device of the present invention,
The coefficient data of the low frequency component is subtracted from the coefficient data of the DC component between the adjacent blocks in the macro block, or the coefficient data of the low frequency component of the adjacent block vertically or horizontally from the coefficient data of the low frequency component. And a difference processing unit that generates the difference data by subtracting
An absolute value calculation unit that generates absolute value data by taking the absolute value of each difference data generated by the difference processing unit;
An average value calculation unit that averages each absolute value data generated by the absolute value calculation unit and generates difference average value data for coefficient data of the DC component and the low frequency component between the adjacent blocks;
The difference average value data generated by the average value calculation unit and each reference value determined for each QP index are compared in order from the smallest reference value, and the difference average value is initially smaller than the reference value. A reference value comparison unit that uses the QP index corresponding to the reference value as the selected QP index (i).

  According to the adaptive QP index selection unit in the image coding apparatus according to the present invention having the above configuration, first, in the difference processing unit, the DC component and the low frequency component between the adjacent blocks of the upper and lower blocks of the macroblock The absolute value of the difference data is generated by the absolute value calculation unit, and the average value of the difference data is obtained by the average value calculation unit. Then, the reference value comparison unit compares the difference average value with the reference value determined for each QP index, and selects one QP index from a predetermined number of QP indexes.

  As described above, according to the configuration of the present invention, the difference processing unit takes the difference data of the coefficient data of the direct current component and the low frequency component between the adjacent blocks of the upper block and the lower block of the macroblock, and calculates the absolute value The difference data is converted into an absolute value at, and the difference value is averaged by converting the difference data into an absolute value at the average value calculation unit, and the difference average value and the reference for each QP index are calculated at the reference value comparison unit. Since one QP index is selected by comparing the values with each other, it becomes possible to select a QP index corresponding to the state of the image in the image area being processed, and the quantization corresponding to the selected QP index. By quantizing with parameters, it is possible to perform quantization that matches the characteristics of the image area being processed.

  As described above, according to the present invention, the difference average value is obtained from the coefficient data of the DC component and the low frequency component between the blocks adjacent to the upper side and the left side of the macroblock in the image area being processed. Since one QP index is selected by comparing the value with the reference value corresponding to the QP index, there is no need to provide a data storage unit for storing the QP index. Therefore, the QP index is obtained from the data storage unit. The processing to obtain is not necessary, and the QP index that matches the characteristics of the image area being processed is determined, and the coefficient data for each block is divided by the corresponding QP parameter. Quantization processing suitable for image characteristics becomes possible, and image encoding processing can be optimized.

It is the block diagram which showed schematic structure of the image coding apparatus which concerns on one Embodiment of this invention. It is the block diagram which showed the structure of the quantization part which concerns on this embodiment. It is the block diagram which showed the structure of the adaptive QP index selection part which concerns on this embodiment. It is the block diagram which showed the structure of the reference value comparison part which concerns on this embodiment. It is a truth table showing the function of the priority encoder according to the present embodiment. It is the flowchart which showed the process sequence centering on the quantization part which concerns on this embodiment. It is explanatory drawing for demonstrating an example of the difference average value processing method in the adaptive QP index selection part which concerns on this embodiment. It is explanatory drawing for demonstrating an example of the difference average value process in the adaptive QP index selection part which concerns on this embodiment. It is explanatory drawing for demonstrating an example of the difference average value process in the adaptive QP index selection part which concerns on this embodiment. It is explanatory drawing for demonstrating an example of the difference average value process in the adaptive QP index selection part which concerns on this embodiment. It is explanatory drawing for demonstrating an example of the reference value comparison in the quantization part which concerns on this embodiment, and quantization parameter selection. It is explanatory drawing for demonstrating an example of the reference value comparison in the quantization part which concerns on this embodiment, and quantization parameter selection. It is the block diagram which showed schematic structure of the conventional image coding apparatus. It is the block diagram which showed the structure of the conventional quantization part. It is the flowchart which showed the processing procedure centering on the conventional quantization part. It is explanatory drawing for demonstrating the relationship between a QP index and a quantization parameter. It is explanatory drawing for demonstrating the relationship between a QP index and a quantization parameter. It is explanatory drawing for demonstrating an example of the system which divides | segments the conventional image and allocates a QP index. It is explanatory drawing for demonstrating an example of the system which divides | segments the conventional image and allocates a QP index.

  Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. 1 is a block diagram illustrating a configuration of an image encoding device 1 according to the present embodiment, and FIGS. 2 to 5 are block diagrams illustrating a configuration and functions of a quantization unit according to the present embodiment. It is a truth table. FIG. 6 is a flowchart showing a processing procedure centering on the quantization unit according to the present embodiment.

  As shown in FIG. 1, the image encoding device 1 of the present embodiment includes a quantization unit 10 having a configuration different from that of the quantization unit 130 in the conventional image encoding device 100 of FIG. 13. However, only in this respect, the configuration of the conventional image encoding apparatus 100 is different. Therefore, the other components are denoted by the same reference numerals as those in the image encoding device 100, and detailed description thereof is omitted.

  As shown in FIG. 2, the quantization unit 10 includes an adaptive QP index selection unit 20 that adaptively selects a QP index, a quantization parameter selection unit 30 that selects a quantization parameter corresponding to the selected QP index, and a selection. The quantization operation unit 40 performs quantization using the quantization parameter and the block counter unit 50 counts a counter value such as a block.

  In the quantization unit 10, similarly to the quantization unit 130 described above, the first layer data and the second layer data output from the frequency conversion unit 120 are handled in units of macroblocks. Process in block units. In the following, the processing of the first layer data will be described, but the same applies to the processing of the second layer data. The operation flow centering on the quantization unit 10 of the present invention shown in FIG. 6 will also be described.

  First, before the image encoding process, the adaptive QP index selection unit 20 acquires and stores a reference value corresponding to each QP index from an external controller or the like (step S1), and the quantization parameter selection unit 30 Similarly, a quantization parameter corresponding to each QP index is acquired from an external controller or the like and stored (step S2).

  Next, at the start of the image encoding process, the block counter unit 50 initializes the macroblock, the color component, and the count values m, c, and n of the block to 1 (step S3). Thereafter, the block counter unit 50 updates the count values m, c, and n while counting the number of coefficient data input from the frequency conversion unit 120 (steps S13, S15, and S17).

  After the start of the image encoding process, the frequency converting unit 120 converts the frequency of the first macroblock (step S4), and the adaptive QP index selecting unit 20 acquires DC coefficient data and LP coefficient data of adjacent blocks ( Step S5). In order to explain the subsequent processing, the block configuration of the adaptive QP index selection unit 20 is shown in FIG.

  The adaptive QP index selection unit 20 includes a difference processing unit 21 that calculates a difference between DC coefficient data and LP coefficient data of adjacent blocks, an absolute value calculation unit 22 that converts the difference data that has been subjected to difference processing into an absolute value, and an absolute value An average value calculation unit 23 for obtaining an average of difference data, a reference value storage unit 24 for storing a reference value for each QP index, and a reference value comparison for selecting the QP index by comparing the calculated difference average value with the reference value The unit 25 is configured.

A specific example of the difference average value calculation in this embodiment is shown in FIG. A macro block composed of 16 × 16 pixels is divided into 16 blocks (one block is 4 × 4 pixels), and the upper left pixel of each block is set to DC (DC coefficient data) or LP (LP coefficient data). Pixels other than are HP (HP coefficient data). The DC / LP coefficients of the upper block of the macro block are DC, LP ₁ , LP ₂ , LP _{3 in} order from the left, and the DC / LP coefficients of the left block of the macro block are DC, LP ₄ , LP ₅ , when the LP _6, each of the difference result of the adjacent blocks _{a = DC-LP 1, b} = LP 1 -LP 2, c = LP 2 -LP 3, d = DC-LP 4, e = LP 4 - LP ₅ , f = LP ₅ −LP _6, and absolute values were taken and averaged for each of the difference results a, b, c, d, e, and f (| a | + | b | + | c | + | d) | + | E | + | f |) / 6 is the difference average value.

An example in which a difference average value is obtained for specific numerical values of coefficient data obtained by frequency-converting a part of a natural image is shown in FIGS. FIG. 8 shows specific numerical values of coefficient data obtained by frequency conversion. DC is −60, and LP _{1 to} LP ₆ are 262, −116, 45, −935, −44, and 223, respectively. At this time, when the difference values DC-LP ₁ , LP ₁ -LP ₂ , LP ₂ -LP ₃ , DC-LP ₄ , LP ₄ -LP ₅ , and LP ₅ -LP ₆ of the coefficient data are obtained in FIG. -322, 378, -161, 875, -891, and -267, respectively. In FIG. 10, taking the absolute value of this difference value gives 322, 378, 161, 875, 891, 267, and 482 when the average value (difference average value) of these is obtained. The difference average value is calculated in such a procedure. The difference average value is used for reference value comparison described later.

  A process of the adaptive QP index selection unit 20 will be described. The difference processing unit 21 obtains a difference between DC coefficient data and LP coefficient data of adjacent blocks input from the frequency conversion unit 120, and inputs the difference data to the absolute value calculation unit 22 (step S6).

  The absolute value calculation unit 22 converts the difference data subtracted by the difference processing unit 21 into an absolute value, and inputs the difference absolute value data to the average value calculation unit 23 (step S7).

  The average value calculation unit 23 calculates a difference average value that is an average value of the difference absolute value data converted into an absolute value by the absolute value calculation unit 22 and inputs the difference average value data to the reference value comparison unit 25. (Step S8).

  The reference value comparison unit 25 calculates the difference average value data averaged by the average value calculation unit 23 in order from the smaller QP index stored in the reference value storage unit 24 in advance, in accordance with the reference corresponding to the QP index. The QP index corresponding to the reference value that is first determined to be smaller than the reference value is selected by comparing with the value (step S9). The reference value is assigned with a small reference value in order from the smallest QP index value.

FIG. 4 shows a block configuration of the reference value comparison unit 25. The reference value comparison unit 25 compares the reference values T _{0 to} T ₁₄ with the difference average value P and outputs comparison results C _{0 to} C ₁₄ , and priorities are added to the comparison results. The priority encoder 27 encodes the value of the QP index.

The comparators 26 a to 26 o respectively compare the reference values T _{0 to} T ₁₄ input from the reference value storage unit 24 with the difference average value P averaged by the average value calculation unit 23. Namely, the comparator 26a compares the differential average value P and the reference value T _0, and 1 if is smaller the difference average value P, and outputs the comparison result C ₀ 0 otherwise.

The comparator 26b compares the difference average value P and the reference value T1 in the same manner as the comparator 26a, and outputs the comparison result C _1. Similarly, until the comparator 26C～26o, respectively to compare the difference average value P and the reference value _T 2 _{through T 14,} and outputs the comparison result _C 2 _{-C 14.} The reference value T ₀ through T ₁₄ is the number of subscripts reference value T ₀ as a minimum value in order of ascending ones, it assumed to be set so that the value becomes larger.

If the reference value _T 0 _{through T 14} is in the above configuration, for example, if it is less than the reference value _{T 2} by a difference average value P is the reference value above _{T 1,} the reference value _{T 0} and _{T 1} is the difference mean value Since the relationship is such that the reference values T _{2 to} T ₁₄ are smaller than P and the difference average value P is not less than (not smaller), the comparison results C ₀ and C ₁ are 0, and the comparison results C _{2 to} C ₁₄ are 1. become. As described above, the comparison results C _{0 to} C ₁₄ are compared with the comparison result C _{0 to} C _n−1 , where the difference comparison value P is smaller than a certain reference value and the comparison result serving as the boundary is C _n. Takes a value of 0, and the comparison results C _{n to} C ₁₄ take a value of 1. When the difference average value P is smaller than the reference value T ₀ , the comparison results C _{0 to} C ₁₄ are all 1, and when the difference average value P is greater than or equal to the reference value T _14, all the comparison results C _{0 to} C ₁₄ are all. 0.

FIG. 5 shows a truth table showing the functions of the priority encoder 27. The comparison results C _{0 to} C ₁₄ on the left side are inputs, and the right QP index i is an output. The x in the truth table indicates that it is arbitrary (it may be 0 or 1).

In order from the top of the truth table of FIG. 5, when C ₀ becomes 1, the QP index i becomes 0, and then when C ₀ becomes 0 and C ₁ becomes 1, the QP index i becomes When C ₀ and C ₁ are 0 and C ₂ is 1, the QP index i is 2. Similarly, when C _{0 to} C _n−1 are 0 and C _n is 1, the QP index i outputs a value of n (n is an integer of 1 to 14) as the corresponding encoding result. Further, when C _{0 to} C _{14 at} the bottom are all 0, the QP index value is 15. Thereby, it is possible to adaptively select the QP index i corresponding to the case where the difference average value P first becomes smaller than the reference value.

  The quantization parameter selection unit 30 changes for each color component corresponding to the QP index i selected by the adaptive QP index selection unit 20 every time the value c of the color component count c counted by the block counter unit 50 changes. A quantization parameter Q (i) is selected (step S10).

  The quantization operation unit 40 quantizes the coefficient data input from the frequency conversion unit 120 for each block using the quantization parameter Q (i) selected by the quantization parameter selection unit 30 (step S11). ).

  In order to specifically describe the above quantization processing, the QP index and the quantization parameter corresponding to 482 which is the difference average value calculated from the coefficient data obtained from a part of the natural image in FIG. An embodiment to select automatically is shown. 11 and 12 show specific reference values, QP indexes corresponding to the reference values, and LP coefficients and HP coefficients corresponding to the QP indexes. In this embodiment, the reference value is specified in the range of 30 to 4000 and the quantization parameter is specified in the range of 25 to 90. However, the specified range is not limited to this, and can be changed as needed depending on image characteristics, image compression conditions, and the like. can do. Also, in this example, priority is given to image quality so that the smaller the difference value (difference average value) of coefficient data in each block is, that is, the smaller quantization parameter is specified on the side where the reference is larger, and the fine components of the image remain. Although the quantization parameter is specified, the specification of the quantization parameter is not limited to this.

  When the calculated difference average value 482 is applied to FIGS. 11 and 12, the difference average value 482 is compared from the smaller reference value, and a reference value smaller than the difference average value is obtained first. Therefore, the corresponding QP index is QP index 8 in the reference value 500. For the QP index 8, 80 values are selected for the LP coefficient and the HP coefficient as the quantization parameter.

  In this way, the QP index is adaptively selected from the difference average value of the coefficient data (in this example, the DC coefficient and the LP coefficient) in the adjacent block, and the corresponding quantization parameter is selected and processed. The coefficient data is quantized using the quantization parameter selected for each macroblock which is an image region.

  Each time the quantization processing of one block is completed, the block counter unit 50 checks whether the processed block is the last block, and updates the block counter value (n = n + 1) if it is not the last block (step n = n + 1). If it is the last block (step S12), it is confirmed whether it is the last color component (step S14).

  The block counter unit 50 checks the counter value of the color component, and if it is not the final color component, it initializes the block counter value (n = 1) and updates the color component counter value (c = c + 1) ( If it is the last color component (step S14), it is confirmed whether it is the last macroblock (step S16).

  The block counter unit 50 confirms the counter value of the macro block. If it is not the final macro block, the block counter unit 50 initializes the block counter value and the color component counter value (n = 1, c = 1) and also sets the macro block counter value. Is updated (m = m + 1) (step S17), and if it is the last macroblock (step S16), the quantization processing for one image is completed. Note that, as described above, whether or not it is the last macroblock depends on whether or not the counter value m of the macroblock matches (horizontal image size / 16) × (vertical image size / 16). Judge.

  According to the image encoding apparatus 1 of the present example having the above configuration, first, image data for one image is input to the color space conversion unit 110, and color space conversion is performed according to the pixel format of the image data, or Image data that has not undergone color space conversion and has undergone processing in the color space conversion unit 110 is input to the frequency conversion unit 120.

  The frequency conversion unit 120 performs processing for sequentially converting the data of each pixel into a coefficient for each frequency band in the raster direction in units of macroblocks (16 × 16 pixels), and the first layer data (HP coefficient) after processing, The second layer data (LP coefficient) and the third layer data (DC coefficient) are input to the quantization unit 10 in block units (4 × 4 pixels).

  The quantization unit 10 quantizes the first layer data, the second layer data, and the third layer data that have been coefficientized by the frequency conversion unit 120 with a quantum corresponding to a QP index selected based on the coefficient data in an adjacent block. The first hierarchical data, the second hierarchical data, and the third hierarchical data after processing are input to the coefficient prediction unit 140 after being divided by the quantization parameter and quantized.

  Similarly, the coefficient prediction unit 140 derives a prediction mode for the first tier data, the second tier data, and the third tier data for each block, and performs a process of subtracting between highly correlated blocks. The first layer data, the second layer data, and the third layer data are input to the adaptive scanning unit 150 for each block.

  The adaptive scanning unit 150 performs processing for replacing the coefficient data for each block with respect to the first layer data and the second layer data. In this replacement process, after frequency conversion, non-zero coefficients are collected on the side where the pixel position in the block is small (low frequency side), and the zero value continues on the side where the pixel position in the block is large (high frequency side). This is to increase the compression rate of encoding by collecting data. The first layer data, the second layer data, and the third layer data on which the replacement process is not performed are input to the VLC encoding unit 160.

  The VLC encoding unit 160 performs entropy encoding processing on the first layer data, the second layer data, and the third layer data processed by the adaptive scan unit 150, generates a code stream, The code stream is input to the bit stream output unit 170.

  The bit stream output unit 170 sequentially concatenates the code streams encoded by the VLC encoding unit 160 to the image setting data as header information, and forms one file as a bit stream for one image data. And output the formed file to the outside.

  As described above in detail, according to the image encoding device 1 of the present embodiment, the quantization unit 10 obtains the difference average value of the DC coefficient and the LP coefficient in adjacent blocks in the image area being processed. Since the difference average value is compared with a preset reference, a QP index is selected, and the coefficient data is divided and quantized by the quantization parameter corresponding to the selected QP index. Compared to 130, it is not necessary to store the QP index corresponding to each image area in a large-capacity storage area that increases in proportion to the image size, and can perform quantization according to the image characteristics of the image area being processed. The encoding process can be optimized. In addition, since the reference for comparison with the difference average value can be set from the outside, even if the characteristics of the image data change, it is possible to flexibly cope with the change by changing the reference.

  As mentioned above, although one Embodiment of this invention was described, the specific aspect which this invention can take is not limited to this at all. For example, in the above example, the DC coefficient data and the LP coefficient data are used as the coefficient data of the adjacent blocks for which the difference average value is obtained in the quantization unit 10, but the present invention is not limited to this, and the HP coefficient data is used. Also good. Further, although the DC coefficient data and LP coefficient data of the upper and left blocks in the macroblock are used as the coefficient data, LP coefficient data of other blocks in the macroblock may be used. DC coefficient data and all LP coefficient data may be used.

  In the present embodiment, one QP index common to the LP coefficient data and the HP coefficient data is selected as the QP index corresponding to the area being processed. However, the present invention is not limited to this. Different QP indexes may be selected for each data and HP coefficient data.

  In this embodiment, an example in which image quality is given priority by specifying a smaller quantization parameter as the coefficient change is larger, that is, as the reference value is larger (as the QP index is larger) has been described. However, the present invention is not limited to this, and conversely, the larger the coefficient change, that is, the larger the reference value, the larger the quantization parameter is specified, and the quantum corresponding to each QP index is increased so as to increase the compression rate. You may specify a parameter.

DESCRIPTION OF SYMBOLS 1 Image coding apparatus 10 Quantization part 20 Adaptive QP index selection part 21 Difference processing part 22 Absolute value calculation part 23 Average value calculation part 24 Reference value storage part 25 Reference value comparison part 26a-26o Comparator 27 Priority encoder 30 Quantization Parameter selection unit 40 Quantization operation unit 50 Block counter unit 100 Image encoding device 110 Color space conversion unit 111 QP index acquisition / storage unit 112 Quantization parameter selection unit 113 Quantization operation unit 114 Block counter unit 120 Frequency conversion unit 130 Quantum 140 Coefficient prediction unit 150 Adaptive scan unit 160 VLC encoding unit 170 Bit stream output unit

Claims (2)

The image data is frequency-converted for a macroblock consisting of m blocks in each horizontal and vertical direction in units of a block consisting of n pixels in each horizontal and vertical direction, and the upper left pixel position of each block in the macroblock The data corresponding to the pixel position other than the above is used as the coefficient data of the high frequency component, the data corresponding to the upper left pixel position of each block in the macro block is collected, further frequency-converted, and the upper left pixel position after the frequency conversion. The frequency conversion unit which makes the corresponding data the coefficient data of the direct current component, and the data corresponding to the pixel position other than the upper left the coefficient data of the low frequency component,
A quantization unit that quantizes the coefficient data of each component frequency-converted by the frequency conversion unit by dividing the coefficient data by a quantization parameter held in an internal or external storage area to obtain quantized coefficient data;
A coefficient scanning unit that performs a process of changing the order of pixels in the block of quantized coefficient data of the low frequency component and the high frequency component quantized by the quantizing unit, and sets the coefficient data after replacement;
An image coding unit comprising: an encoding unit that encodes the quantized coefficient data of the DC component after quantization by the quantization unit and the post-replacement coefficient data after the replacement process by the coefficient scanning unit to generate encoded data In the device
The quantization unit is
The difference between the DC component coefficient data and the low frequency component coefficient data of the block adjacent to the left and right in the upper block in the macro block and the upper and lower adjacent blocks in the left block, and the low frequency in the adjacent block Each QP that takes the difference between the coefficient data of the components, converts it to an absolute value, takes the average, and designates one quantization parameter by p numerical values from 0 to p−1. A reference value corresponding to an index, which is a reference value held in an external or internal storage area, is compared in order from the smallest reference value, and the difference average value is initially smaller than the reference value. An adaptive QP index selection unit that sets the QP index corresponding to the selected reference value as the selected QP index (i);
A quantization parameter selection unit that selects a quantization parameter Q (i) corresponding to the selected QP index (i);
A quantization operation processing unit configured to quantize each coefficient data of a low frequency component and a high frequency component in each block in the macro block using the selected quantization parameter Q (i). An image encoding device.
However, n, m and p are integers of 2 or more, and i is an integer of 0 to p-1. The adaptive QP index selection unit
The coefficient data of the low frequency component is subtracted from the coefficient data of the DC component between the adjacent blocks in the macro block, or the coefficient data of the low frequency component of the adjacent block vertically or horizontally from the coefficient data of the low frequency component. And a difference processing unit that generates the difference data by subtracting
An absolute value calculation unit that generates absolute value data by taking the absolute value of each difference data generated by the difference processing unit;
An average value calculation unit that averages each absolute value data generated by the absolute value calculation unit and generates difference average value data for coefficient data of the DC component and the low frequency component between the adjacent blocks;
The difference average value data generated by the average value calculation unit and each reference value determined for each QP index are compared in order from the smallest reference value, and the difference average value is initially smaller than the reference value. The image coding apparatus according to claim 1, further comprising: a reference value comparison unit that uses a QP index corresponding to the reference value as a selected QP index (i).

Images