JP4243218B2 - Quantization control device, method and program thereof, and adaptive quantization encoding device - Google Patents

Description

  The present invention relates to an image compression coding technique that reduces the amount of information by quantization, and more specifically, a quantization control device that changes the degree of quantization based on the characteristics of the image, its method, and its The present invention relates to a program and an adaptive quantization coding apparatus.

  Currently, MPEG-1, MPEG-2 and MPEG-4 standardized by ISO / IEC JTC1 / SC29 / WG11 (MPEG), ISO / IEC JTC1 / SC29 / WG1 (JPEG) are used as compression encoding methods for images. ) Standardized by JPEG and ITU-T recommended by H.264. 261 and H.H. H.263, H.264 standardized by ITU-T and ISO / IEC JTC1 / SC29 / WG11. H.264 / MPEG-4 AVC (see Non-Patent Document 1, for example).

  In these compression coding methods, an image is divided into regions of a specific size called blocks (for example, 16 × 16 pixels), and motion compensation prediction and conversion processing (discrete cosine transform [DCT: Discrete Cosine Transform]) is performed in units of the blocks. Cosine Transform] and the like, and the conversion result is quantized so as to greatly reduce high-frequency components, and is encoded by variable-length coding.

  Here, the quantization is to express the amplitude of a digital signal (conversion coefficient) subjected to conversion processing such as DCT by a certain representative value. If this quantization is performed finely, the original transform coefficient can be reproduced more faithfully, but since many representative values are required, the amount of information after quantization increases. Conversely, if quantization is performed roughly, the amount of information after quantization is reduced, but the reproducibility of the original transform coefficient is deteriorated. That is, in the compression encoding method, the coarser the quantization, the more deteriorated the image quality of the reproduced image. In the compression coding method, a parameter for setting the degree of quantization (level) is generally called a quantization parameter.

  The quantization parameter can be set to the same value throughout the entire image. However, when the encoding target is a moving image, the quantization parameter can be set to a different value for each image in the time direction. Alternatively, different quantization parameters can be set in units of areas called slices obtained by dividing one image into several band-like areas. Furthermore, different quantization parameters can be set in units of 16 × 16 pixel areas called macroblocks.

  As described above, when the quantization parameter is not set to the same value for the entire image but is set to a different value, it is necessary to set the quantization parameter based on some standard. As this setting method, any method may be used as long as it outputs encoded data (encoded stream) conforming to the encoding method standard, and various methods have been proposed.

  For example, in the encoding method employed in MPEG-2 TM-5 (Test Model 5), which is one of the compression encoding methods described above, an image is quantized with a quantization parameter in units such as slices and macroblocks. By changing the value, rate control is performed so that the bit rate of the entire encoding result becomes a predetermined bit rate.

  In this TM-5 encoding method, an amount of activity (image feature amount) indicating the degree of flatness of an image is defined in units of macroblocks. Then, TM-5 reduces the bit rate by reducing the quantization parameter value and finely quantizing the conversion coefficient of the macroblock in the flat part of the image that is visually prominently deteriorated depending on the activity value. In a complicated portion of a pattern that is increased and is relatively inconspicuous, deterioration is controlled by increasing the value of the quantization parameter and coarsely quantizing the conversion coefficient of the macroblock.

As described above, in the TM-5 encoding method, the encoding parameter value is set on the basis of the local feature amount of the image, thereby reducing the image quality deterioration within a predetermined bit rate. Generated data.
As described above, various other methods for encoding an image using the feature amount of the image have been proposed, and the control of the code is performed by defining various feature amounts (for example, Patent Document 1).
"ISO / IEC 13818-2: 2000 Information technology-Common coding of video and associated audio information: Video", December 21, 2000 JP 2001-245303 A (paragraphs 0017 to 0022, FIGS. 1 and 2)

  In the conventional compression coding method described above, the feature amount of an image is extracted in a certain region unit (for example, macroblock unit), the quantization parameter is changed based on the feature amount, and the degree of quantization in each region unit By changing, encoded data in which deterioration of image quality is suppressed is generated within a predetermined bit rate. However, in the conventional compression coding method, for example, a part where the picture pattern is complex is determined as a region where deterioration in image quality is hardly noticeable, and the quantization is coarsened. However, when there is no movement (less) in the time direction, there is a problem that if the quantization is coarsened, the deterioration of the image quality is conspicuous.

  In the conventional compression coding method, for example, when an object (subject) such as a person is reflected in an image, a macroblock including only an object and a macroblock including a boundary between the object and the background, There is a case where the image feature amount changes and the quantization parameter may be different, and there is a problem that the image quality is deteriorated at the contour portion of the subject on the boundary.

  The present invention has been made in view of the above-described problems. In compression coding of an image in which the amount of information is reduced by quantization, the quantization is controlled based on the local feature amount of the image. By performing the quantization control device, the method and the program thereof, and the adaptive quantum, which are capable of suppressing the degradation of the image quality of the decoded image in the screen and in the time direction, and further suppressing the degradation of the image quality on the boundary of the object. An object of the present invention is to provide a coding / encoding device.

The present invention was devised to achieve the above object. First, the quantization control apparatus according to claim 1 encodes an image in units of blocks of a specific size. A quantization control device that controls the degree of quantization of the block, comprising: a feature amount extraction unit, a correction block determination unit, and a quantization parameter association unit; and the correction block determination unit includes a flag The image generation unit and the expansion / contraction unit are provided .

  According to such a configuration, the quantization control device extracts the feature quantity indicating the feature of the image from the block by the feature quantity extraction unit. The feature amount of the image can be expressed by various features such as an average value of pixel values, dispersion, and color information. Thus, by extracting the feature amount for each block, it is possible to determine whether there is similarity as an image between the blocks. Note that if the degree of quantization is changed based on the feature amount of the image, the degree of quantization may be different in blocks having no similarity as images.

Therefore, the quantization control device determines a block in which the feature amount between blocks changes by a certain value or more based on the feature amount of each block extracted by the feature amount extraction unit by the correction block determination unit. It is determined that there is no block, and it is determined as a block for correcting the degree of quantization.
At this time, the quantization control device includes the feature amount of each block extracted by the feature amount extraction unit by the flag image generation unit of the correction block determination unit included in the range specified by one or more preset threshold values. A flag image associated with the pixel is generated using the determination result of each block as a pixel value. As a result, a flag image that is divided into a threshold value that is greater than or less than the threshold value by one threshold value or a flag image that is classified according to whether or not it is included in a range indicated by a plurality of threshold values is generated. This flag image represents, for example, a binary value of “1” and “0” for regions having different feature values with the feature value specified by the threshold as a boundary. In this way, by binarizing the image according to the feature amount, the image quality becomes discontinuous at the boundary between the object and the background, etc. at different locations as the values of “1” and “0” (blocks are similar) It is possible to determine that the location is not present.
Then, the quantization control device performs expansion or contraction of the flag image by converting the pixel value of each pixel of the flag image into an adjacent pixel value by the expansion / contraction unit.

Then, the quantization control device changes the feature amount of the block subjected to correction by the correction block determination unit to the feature amount approximate to the feature amount of the adjacent block of the block by the quantization parameter association unit. , you output a quantization parameter indicating the degree of quantization which is associated in advance for the feature quantity of the feature quantity and changes that have not been blocked in the changed blocks for each block. That is, the quantization parameter associating means approximates (or makes identical) the quantization parameter of the block whose feature quantity changes at the boundary (contour) portion of the object (person, etc.) to the quantization parameter of the adjacent block. Make corrections. This reduces the change in the quantization parameter in each block between the object and the object boundary.

  The quantization parameter in the quantization parameter association means is corrected by correcting the feature quantity of the block to be corrected based on the feature quantity of the adjacent block, and then determining the quantization parameter corresponding to the feature quantity. Alternatively, the quantization parameter corresponding to the feature amount of the block to be corrected may be corrected based on the quantization parameter corresponding to the feature amount of the adjacent block.

  Further, the quantization control device according to claim 2 is the quantization control device according to claim 1, wherein the difference information calculation unit is provided in a preceding stage of the feature amount extraction unit, and the feature amount extraction unit includes The feature amount of the block is extracted based on the difference information calculated by the difference information calculation means.

  According to such a configuration, the quantization control device takes the difference between the target block and the block of the past image in the time direction at the same position on the screen by the difference information calculation unit, Difference information in the time direction is calculated. Then, the feature amount extraction unit extracts the feature amount of the block based on the difference information calculated by the difference information calculation unit, so that the feature amount of the block includes a change in the image in the time direction as a feature. Information. As a result, the correction block determination unit can determine a block with a small motion that is likely to cause deterioration in image quality as a block to be corrected.

Furthermore, the adaptive quantization coding apparatus according to claim 3 , wherein the image is coded by dividing the image into blocks of a specific size and changing the degree of quantization according to the feature amount of the block. An adaptive quantization coding apparatus that includes a block dividing unit, a quantization control device according to claim 1 or 2, and a quantization coding unit.

  According to such a configuration, the adaptive quantization coding apparatus divides the image into blocks each having a specific size (for example, a macro block of 16 × 16 pixels) by the block dividing unit. Then, the adaptive quantization coding apparatus determines, for each block, a quantization parameter value indicating the degree of quantization based on the feature amount of the block image (local image) by the quantization control apparatus. Then, the adaptive quantization coding apparatus compresses the block divided by the block dividing means by the quantization coding means by changing the degree of quantization according to the value of the quantization parameter determined by the quantization control apparatus. Turn into.

The quantization control method according to claim 4 is a quantization control method for controlling the degree of quantization of the block when compressing and encoding an image in units of a block having a specific size. The procedure includes a feature amount extraction step, a correction block determination step, and a quantization parameter correction step.

  According to this procedure, the quantization control method extracts a feature quantity indicating a feature of an image from a block by a feature quantity extraction step. Thus, by extracting the feature amount for each block, it is possible to determine whether there is similarity as an image between the blocks. Note that if the degree of quantization is changed based on the feature amount of the image, the degree of quantization may be different in blocks having no similarity as images.

Therefore, the quantization control method determines whether the feature amount of each block extracted in the feature amount extraction step is included in a range specified by one or more preset threshold values in the correction block determination step. The flag image is generated by associating the determination result of each block with a pixel value, and the flag image is expanded or contracted by converting the pixel value of each pixel of the flag image into a neighboring pixel value. It is determined that the block corresponding to the pixel that is the target of expansion or contraction is a block that corrects the degree of quantization .

In the quantization control method, the feature amount of the block determined to correct the degree of quantization in the correction block determination step by the quantization parameter correction step is changed to a feature amount approximate to the feature amount of the neighboring block of the block. While changing, the quantization parameter which shows the degree of said quantization previously matched with the feature-value of the changed block and the feature-value of the block which was not changed is output for every block .

Furthermore, the quantization control program according to claim 5 , when compressing and encoding an image with a block having a specific size as a unit, in order to control the degree of quantization of the block, The calculation unit, the feature amount extraction unit, the flag image generation unit, the expansion / contraction unit, and the quantization parameter association unit are configured to function.

  According to such a configuration, the quantization control program obtains the difference between the target block and the block of the past image in the time direction at the same position on the screen by the difference information calculation unit. Difference information in the time direction is calculated. Then, the quantization control program extracts the feature quantity indicating the feature of the image from the difference information by the feature quantity extraction unit. This feature amount is information including a change in the image in the time direction as a feature.

Then, the quantization control program determines whether or not the feature amount of each block extracted by the feature amount extraction unit is included in a range specified by one or more preset threshold values by the flag image generation unit. Then, a flag image associated with the pixel is generated using the determination result as a pixel value. As a result, a flag image that is divided into a threshold value that is greater than or less than the threshold value by one threshold value or a flag image that is classified according to whether or not it is included in a range indicated by a plurality of threshold values is generated.

Further, the quantization control program performs expansion or contraction of the flag image by converting the pixel value of each pixel of the flag image into an adjacent pixel value by expansion / contraction means.
Then, the quantization control program approximates the feature amount of the block corresponding to the pixel that is the target of expansion or contraction in the flag image to the feature amount of the adjacent block of the block by the quantization parameter association unit . While changing to the feature amount, a quantization parameter indicating the degree of quantization associated with the feature amount of the changed block and the feature amount of the block that has not been changed is output for each block. As a result, it is possible to reduce blocks in which image quality is discontinuous.

According to the first , fourth , and fifth aspects of the present invention, when the image is compressed and encoded with a block having a specific size as a unit, the quantization parameter is changed based on the feature amount of the image of each block. Thus, it is possible to finely quantize a block where deterioration is conspicuous as an image, and to roughen quantization of a block where deterioration is not conspicuous. As a result, the present invention can improve the reproducibility of a block in which deterioration is conspicuous as an image and improve the image quality as compared with the case where the same quantization is performed on the entire image. Furthermore, the present invention occurs between blocks by correcting the quantization parameter of each block so as to reduce the change in the degree of quantization even when the change in the feature amount between the blocks is large. Degradation of image quality can be suppressed.
Further, according to the present invention, it is possible to detect a block in which the image quality is discontinuous at the boundary between the object and the background by binarizing the image with the feature amount and generating a flag image. Then, by expanding or contracting the flag image, it is possible to approximate the block whose image quality is discontinuous at the boundary or the like to the feature amount of the adjacent block and reduce the discontinuous block. As a result, it is possible to suppress deterioration in image quality.

According to the second aspect of the present invention, it is possible to detect, for each block, an area in the moving image in which there is little movement that is likely to cause image quality degradation. Thus, according to the present invention, it is possible to suppress the deterioration of the image quality by making the quantization fine in association with the region where the motion is small.

According to the third aspect of the present invention, the image is divided into blocks of a specific size, and the compression coding is performed by changing the degree of quantization of the block based on the feature amount of the image of the block. Therefore, the image quality can be improved as compared with the case where the same quantization is performed. Furthermore, since the present invention can change the degree of quantization based on the feature amount of a block that changes in the time direction, even when compressing and coding a moving image, the image quality can be improved in the same way as a still image. Can be improved.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
[Configuration of Adaptive Quantization Encoding Device]
First, the configuration of an adaptive quantization coding apparatus (encoding apparatus) according to the present invention will be described with reference to FIG. FIG. 1 is a block diagram showing a schematic configuration of an adaptive quantization coding apparatus according to the present invention.
As shown in FIG. 1, the adaptive quantization coding apparatus 1 divides an image into blocks of a specific size, and adapts the feature amount of the local image that is the block to change the degree of quantization. Is encoded to generate encoded data. Here, the adaptive quantization coding apparatus 1 includes a block division unit 10, a quantization control unit 20, and a quantization coding unit 30.

  The block dividing means 10 divides an input image into blocks each having a specific size. The blocks (local images) divided here are sequentially output to the quantization control means 20 and the quantization coding means 30. The block dividing means 10 may perform image division in units of 16 × 16 pixel macroblocks according to an encoding standard such as MPEG-2, or 16 × 8, 8 × 16, 8 ×. You may perform by the block unit of the magnitude | size along various encoding standards, such as 8 and 8x4 pixel.

  The quantization control means (quantization control device) 20 sets a quantization parameter indicating the degree of quantization when compressing and encoding an image in units of a specific size as a feature (feature amount) of the block. Control based on this. The quantization control unit 20 receives the block divided by the block dividing unit 10, determines a quantization parameter based on the feature amount of the block, and outputs the quantization parameter to the quantization encoding unit 30.

The quantization encoding unit 30 generates encoded data by reducing the amount of information by performing quantization on an image input in units of blocks and performing compression encoding. The quantization encoding means 30 receives the blocks divided by the block dividing means 10 and changes the degree of quantization based on the quantization parameter input from the quantization control means 20 to compress each block. Encoding is performed.
Hereinafter, each structure of the quantization control means 20 and the quantization encoding means 30 is demonstrated in detail.

(Configuration of quantization control means)
Here, the configuration of the quantization control means 20 will be described with reference to FIG. FIG. 2 is a block diagram showing the configuration of the quantization control means (quantization control device) according to the present invention. Here, the quantization control unit 20 includes a feature amount extraction unit 21, a correction block determination unit 22, and a quantization parameter association unit 23.

The feature amount extraction means 21 extracts a unique feature of the local image from the input block (local image) as a feature amount. The feature amount extracted here is output to the correction block determination unit 22 and the quantization parameter association unit 23.
Here, the feature amount extraction unit 21 extracts the variance of the pixel values of the local image as the feature amount. In general, the greater the variation in brightness or color in the image, the greater the variance, and the smaller the variation, the smaller the variance. In addition, human vision is more likely to detect image quality degradation in flat areas with less variation. Therefore, this feature amount (dispersion) can be used as an indicator of whether or not the block is a block in which deterioration in image quality is likely to be detected.
As described above, the feature quantity extraction unit 21 includes an average value calculation unit 21a and a variance calculation unit 21b in order to extract variance as a feature quantity from a block.

The average value calculator 21a calculates the average value of each pixel of the input block. The average value calculated here is output to the variance calculation unit 21b. For example, when the block is a macro block of 16 × 16 pixels and the pixel values of 256 pixels are P _{1 to} P ₂₅₆ , the average value calculation unit 21a calculates the average value P _mean by the following equation (1).

The variance calculating unit 21b calculates the variance based on the average value calculated by the average value calculating unit 21a and the original pixel value of each pixel of the block. The variance calculated here is output to the correction block determining unit 22 and the quantization parameter associating unit 23 as the feature amount of the block. For example, when the average value of each pixel of the block is calculated by the equation (1), the variance calculating unit 21b calculates the variance V _blk by the following equation (2).

  In the feature quantity extraction unit 21, the variance of the pixel values of the local image is extracted from the block as the feature quantity of the block. However, other feature quantities may be extracted. Examples of other feature amounts include a maximum value, a minimum value, a median (median value) of pixel values indicating luminance and color difference, or a frequency component indicating color variation and its period. For the calculation of the frequency component, Fourier transform, discrete cosine transform (DCT), Hadamard transform, KL transform (Karhunen-Loeve Transform), or the like can be used.

If the feature amount of the image is used as described above, for example, when an object (subject) such as a person is reflected in the image and the block (local image) includes a boundary between the object and the background, Since the local image and the local image that is a flat region such as only an object have different feature amounts, the continuity of the image quality is impaired near the boundary depending on the degree of quantization.
Accordingly, whether or not the feature amount extracted by the feature amount extraction unit 21 is discontinuous with respect to the feature amount of the adjacent block and whether or not the degree of quantization of the corresponding block is corrected is corrected block determination described below. Determined by means 22.

Correction block determination unit 22 based on the feature amount of each block extracted by the feature extraction means 21 is for determining whether to correct the degree of quantization for each block. Here, the correction block determination unit 22 includes a flag image generation unit 22a, a flag image storage unit 22b, and an expansion / contraction unit 22c.

  The flag image generation unit (flag image generation unit) 22a is configured based on the feature amount (here, variance) extracted by the feature amount extraction unit 21 and the threshold value (threshold information) given from the outside. A flag image (binary image) in which the amount is equal to or greater than the threshold value or less than the threshold value is represented by a value of “1” or “0”.

Here, the flag image will be described with reference to FIG. FIG. 3 is a diagram illustrating an example of a flag image generated by the flag image generation unit.
As shown in FIG. 3, the flag image has a value of “1” when the feature amount of one block B (for example, a macroblock) is equal to or greater than a threshold value, and a value of “0” when the feature value is less than the threshold value It is expressed as For example, when the feature value is variance, a block B having a variance value equal to or greater than a given threshold is a complex part of an object boundary or a pattern in the image, and is assigned a value of “1”. In addition, the block B whose variance value is smaller than the given threshold value is a flat portion of the image and is given a value of “0”. In this way, by binarizing the feature amount of the image using the threshold, it is possible to detect a boundary having different image quality as a boundary between “1” and “0”.
Returning to FIG. 2, the description of the quantization control means 20 will be continued.

  The flag image storage unit 22b temporarily stores the flag image generated by the flag image generation unit 22a, and is a storage unit such as a memory. The flag image stored in the flag image storage unit 22b is read by the expansion / contraction unit 22c.

The expansion / contraction unit (expansion / contraction means) 22c converts the pixel value of each pixel of the flag image generated by the flag image generation unit 22a and stored in the flag image storage unit 22b into adjacent pixel values. Thus, the flag image is expanded or contracted. Note that a pixel whose pixel value is converted from “0” to “1” or “1” to “0” due to the expansion or contraction of the flag image indicates a location where the continuity of image quality is interrupted. . Therefore, the expansion / contraction unit 22c notifies the quantization parameter associating unit 23 of the pixel position that is the conversion target of expansion or contraction in the flag image as the block position for correcting the degree of quantization.
Note that the expansion or contraction of the flag image can be performed by, for example, a close process or an open process in the morphological process of the binary image.

  Here, with reference to FIG. 4, the expansion / contraction processing (morphological processing) of the flag image in the expansion / contraction unit 22c will be specifically described. 4A and 4B are explanatory diagrams for explaining the expansion / contraction processing. FIG. 4A is an explanatory diagram for explaining an example of expansion / contraction of the flag image by the 8-neighbor processing, and FIG. 4B is an expansion diagram of the flag image. For example, (c) shows a contraction example of a flag image.

As shown in FIG. 4A, here, in the flag image, the value of the pixel to be expanded / contracted (expansion / deflection target pixel P ₀ ) is set to eight neighborhoods close to the expansion / contraction target pixel P _0. This is determined by the values of the pixels P _{1 to} P ₈ .
For example, when the value of at least one pixel among the eight neighboring pixels P _{1 to} P ₈ is “1”, the value of the expansion / contraction target pixel P ₀ is set to “1”. As a result, the value of the expansion / contraction object pixel P ₀ is expanded to “1”.
Further, when the values of all the eight neighboring pixels P _{1 to} P ₈ are “0”, the value of the expansion / contraction target pixel P ₀ is set to “0”. As a result, the value of the expansion / contraction object pixel P ₀ is contracted to “0”.
That is, assuming that the values of the pixels P _{1 to} P _{8 in} the vicinity of ₈ are f (P ₁ ) to f (P ₈ ), the value f (P ₀ ) of the expansion / contraction pixel P ₀ is the following (3) It can be determined by the logical operation shown in the equation.

Note that the value f (P ₀ ) of the expansion / contraction object pixel P ₀ uses the values of the pixels P _{1 to} P _{8 in} the vicinity of ₈ , so that the value of the pixel P ₈ is input as the value of the flag image. Will be decided. That is, in the flag image, the value of the expansion / contraction target pixel P ₀ is determined after a delay of one line + 1 pixel indicated by a thick line in the drawing.

In this way, by determining the value of the expansion / contraction target pixel P ₀ , as shown in FIG. 4B, when the value of at least one pixel is “1” in 8 neighboring pixels, expansion / contraction is performed. Even if the value of the target pixel is “0”, it is converted to “1” (corresponding to the close process in the morphological process).

Further, as shown in FIG. 4C, when the values of the pixels in the vicinity of 8 are all “0”, even if the value of the expansion / contraction pixel is “1”, it is converted to “0”. (Corresponding to the open process in the morphological process).
Returning to FIG. 2, the description of the quantization control means 20 will be continued.

  The quantization parameter association unit 23 outputs a quantization parameter indicating the degree of quantization, which is associated in advance with the feature amount, based on the feature amount of the block extracted by the feature amount extraction unit 21. It is. Further, when the correction block determination unit 22 notifies the quantization parameter association unit 23 that the target block is a block that corrects the degree of quantization (quantization parameter), the quantization parameter association unit 23 quantizes the block. It also corrects the parameters and outputs them. Here, the quantization parameter associating unit 23 includes a feature amount storage unit 23a, a feature amount change unit 23b, a correspondence table storage unit 23c, and a corresponding quantization parameter output unit 23d.

The feature amount storage unit 23a temporarily stores the feature amount (here, distributed) extracted by the feature amount extraction unit 21, and is a storage unit such as a memory. In the feature amount storage unit 23a, at least block feature amounts corresponding to pixels that are to be expanded / contracted by the expansion / contraction unit 22c of the correction block determination unit 22 and pixels that are close to (eight adjacent to) the pixel. Is memorized. The feature amount stored in the feature amount storage unit 23a is read by the feature amount change unit 23b.

The feature amount changing unit 23b is a feature of a block corresponding to a pixel position that is a conversion target of expansion or contraction notified from the expansion / contraction unit 22c of the correction block determination unit 22, that is, a block that corrects the degree of quantization. The amount is changed based on the feature amount of the adjacent block adjacent to the block.

For example, the feature amount changing unit 23b sets the feature amount of the target block as an average value or an intermediate value of the feature amounts of adjacent blocks stored in the feature amount storage unit 23a. Alternatively, the feature amount of the block is obtained by linearly extrapolating the feature amount of a plurality of blocks adjacent to the left side (or upper side) of the target block. Thereby, the feature amount of the block becomes a feature amount approximate to the feature amount of the adjacent block.
Here, the feature amount of the changed block is output to the corresponding quantization parameter output unit 23d. The feature amount of the block that has not been changed is output to the corresponding quantization parameter output unit 23d as it is.

  The correspondence table storage unit 23c stores a correspondence table in which block feature amounts and quantization parameters are associated in advance, and is a storage unit such as a hard disk. For example, when the feature amount is variance, as shown in FIG. 5, the correspondence table Tv in which the quantization parameter Qp and the variance var are associated is stored in the correspondence table storage unit 23c. As a result, the quantization parameter is uniquely specified by the value of the feature amount (variance).

The corresponding quantization parameter output unit 23d reads out and outputs the quantization parameter corresponding to the feature amount from the correspondence table storage unit 23c. In this example, the corresponding quantization parameter output unit 23d acquires the quantization parameter associated with the feature amount from the correspondence table stored in the correspondence table storage unit 23c. Either one of the two quantization parameters classified according to the threshold value of the feature amount may be output.
Alternatively, the corresponding quantization parameter output unit 23d may define the feature amount and the quantization parameter by a relational expression, and calculate the quantization parameter by the relational expression. In this case, the correspondence table storage unit 23c can be omitted from the configuration.

  Here, an example in the case where the feature quantity and the quantization parameter are expressed by, for example, a linear relational expression is shown in the following expression (4). Qp represents the quantization parameter, and maxQp and minQp represent the maximum value and the minimum value of the quantization parameter, respectively. Ch is a feature amount, and maxCh and minCh are the maximum value and the minimum value of the feature amount, respectively. Norm represents a coefficient for normalization.

In the case where variance is used for the feature value Ch in the equation (4), for example, the quantization parameter Qp is calculated assuming that maxQp = 51, minQp = 0, maxCh = 255, minCh = 0, norm = 20.
The quantization parameter associating unit 23 sets the feature amount of the block for correcting the degree of quantization determined by the correction block determining unit 22 in the feature amount changing unit 23b, which is the previous stage of the corresponding quantization parameter output unit 23d. On the other hand, although the quantization parameter is corrected by approximating the feature amount of the adjacent block stored in the feature amount storage unit 23a, the correction of the quantization parameter is not limited to this.

  For example, the quantization parameter associating unit 23 may be configured by a correspondence table storage unit 23c, a corresponding quantization parameter output unit 23d, and a quantization parameter changing unit (not shown). In this case, the corresponding quantization parameter output unit 23d reads the quantization parameter corresponding to the feature amount extracted by the feature amount extraction unit 21 from the correspondence table storage unit 23c, and the quantization parameter change unit (not shown) performs the correction. The quantization parameter of the block to be corrected determined by the block determination unit 22 may be corrected with an average value, an intermediate value, or the like of the quantization parameter of the adjacent block.

  As described above, the quantization control means 20 can control the value of the quantization parameter for each block according to the feature amount of the local image that is a block obtained by dividing the image into blocks of a specific size. . Also, for example, when a block includes a boundary (for example, a person and a background), the feature amount of the block is corrected based on the feature amount of the block adjacent to the block, so that The difference in the degree of quantization can be reduced, and the continuity of image quality can be maintained in the screen.

  The quantization control means (quantization control device) 20 can be realized by causing a general computer to execute a program and operating an arithmetic device or a storage device in the computer. This program (quantization control program) can be distributed via a communication line, or can be distributed by writing in a recording medium such as a CD-ROM.

(Configuration of quantization coding means)
Next, the configuration of the quantization encoding means 30 will be described with reference to FIG. FIG. 7 is a block diagram showing the configuration of the quantization coding means. Here, the quantization encoding unit 30 includes a prediction unit 31, a conversion unit 32, a quantization unit 33, an inverse quantization unit 34, an inverse conversion unit 35, and an encoding unit 36.

The prediction unit 31 predicts a block to be encoded from the differential local decoded image generated by the inverse conversion unit 35 described later, and the block (local image) divided by the block dividing unit 10 (FIG. 1). By taking the difference, a difference image that is a prediction error is generated. The difference image generated here is output to the conversion means 32.
The prediction unit 31 includes a local decoded image storage unit (not shown), stores the differential local decoded image generated by the inverse conversion unit 35, and performs inter-frame prediction based on motion compensation prediction or prediction from adjacent blocks. By performing intra-frame prediction to be performed, a block to be encoded is predicted, and a difference image is generated by taking a difference from the input block that is an original image.

The conversion unit 32 generates a conversion coefficient by converting the difference image generated by the prediction unit 31 into information in the frequency domain. The transform coefficient generated here is output to the quantization means 33.
Note that the frequency conversion performed by the conversion means 32 is, for example, discrete cosine transform (DCT) performed in units of blocks of 8 × 8 pixels used in MPEG-2 or the like, or H.264. It is possible to use a 4 × 4 pixel block unit integer or decimal precision conversion method used in H.264 / MPEG-4 AVC. H. The frequency conversion used in H.264 / MPEG-4 AVC is a floating point calculation of discrete cosine transform converted to an integer (or a decimal number).

The quantization unit 33 quantizes the transform coefficient generated by the transform unit 32 based on the quantization parameter output from the quantization control unit 20 (FIG. 1), and generates a quantized transform coefficient. The quantized transform coefficient generated here is output to the inverse quantization means 34 and the encoding means 36. The quantization parameter is also output to the inverse quantization unit 34 and the encoding unit 36.
In this quantization means 33, the transform coefficient corresponding to the block to be encoded is roughly quantized or finely quantized with a quantization parameter that differs depending on the feature amount of the block.

  The inverse quantization unit 34 generates an inverse quantization transform coefficient by performing inverse quantization on the quantized transform coefficient generated by the quantization unit 33. The inverse quantization transform coefficient generated here is output to the inverse transform means 35. The inverse quantization means 34 inversely quantizes the quantized transform coefficient based on the same quantization parameter as that used in the quantization means 33.

  The inverse transform unit 35 performs inverse transform of the transform performed by the transform unit 32 on the inverse quantization transform coefficient generated by the inverse quantization unit 34. An image (difference local decoded image) generated by being inversely transformed by the inverse transformation means 35 is output to the prediction means 31.

  The encoding unit 36 generates encoded data obtained by encoding an image by entropy encoding and multiplexing the quantized transform coefficient generated by the quantizing unit 33 and the quantization parameter. . When the image is a moving image, the encoding unit 36 indicates the motion information (motion vector) when the motion compensation prediction is performed in the prediction unit 31, which past frame is used for the prediction, etc. Information is also multiplexed into the encoded data.

  Note that the quantization encoding unit 30 described here is not limited to this configuration as long as it realizes an encoding method for compressing and encoding an image by quantization. That is, the quantization encoding means 30 includes MPEG-1, MPEG-2, MPEG-4, JPEG, H.264, and the like. 261, H.H. 263, H.M. An encoding device using various encoding methods such as H.264 / MPEG-4 AVC can be used.

  By configuring the adaptive quantization coding apparatus 1 as described above, the image is quantized by changing the degree of quantization to a specific block (local image) unit according to the characteristics of the local region, Encoded data can be generated.

  The adaptive quantization coding apparatus 1 can be realized by causing a general computer to execute a program and operating an arithmetic device or a storage device in the computer. This program (adaptive quantization coding program) can be distributed via a communication line, or can be distributed by writing in a recording medium such as a CD-ROM.

[Operation of Adaptive Quantization Encoder]
Next, the operation of the adaptive quantization coding apparatus according to the present invention will be described with reference to FIG. 8 (refer to FIGS. 1 and 2 as appropriate for the configuration). FIG. It is a flowchart which shows operation | movement. Here, description will be made mainly on the operation of the quantization control means (quantization control device) 20.

(Block division step)
First, the adaptive quantization coding apparatus 1 divides the image into blocks (local images) having a specific size by the block dividing means 10 (step S1). The blocks (local images) divided here are sequentially output to the quantization control means 20 and the quantization coding means 30.

(Feature extraction step)
Then, the adaptive quantization coding apparatus 1 extracts the feature amount for each block by the feature amount extraction unit 21 of the quantization control unit 20 (step S2). The feature amount extracted here is output to the correction block determination unit 22 and the quantization parameter association unit 23.

  Here, the feature quantity extraction means 21 calculates the average value of each pixel of the block by the average value calculation unit 21a, and the average value calculated by the average value calculation unit 21a by the variance calculation unit 21b and the block The variance is calculated based on the original pixel value of each pixel. As a result, the variance that is the feature amount is extracted from the local image that is the block.

(Correction block judgment step)
Then, the adaptive quantization coding apparatus 1 is based on the feature amount (for example, variance) extracted in step S2 by the flag image generation unit 22a of the correction block determination unit 22 and the threshold value given from the outside. A flag image (binary image) representing whether the feature amount of the block is equal to or greater than the threshold value or less than the threshold value by a value of “1” or “0” is generated (step S3). This flag image is stored in the flag image storage unit 22b.

Subsequently, the adaptive quantization coding apparatus 1 converts the pixel value of each pixel of the flag image generated in step S3 into a neighboring pixel value by the expansion / contraction unit 22c of the correction block determination unit 22. Then, the flag image is expanded or contracted (step S4).
It is determined that the block subjected to expansion or contraction in step S4 is a block for correcting the degree of quantization.

(Quantization parameter correction step)
Furthermore, the adaptive quantization coding apparatus 1 is determined to be a block that corrects the degree of quantization by being expanded or contracted in step S4 by the feature amount changing unit 23b of the quantization parameter association unit 23. Change the block feature.

  That is, the feature amount changing unit 23b determines whether or not the block (the processing target block) corresponds to the expanded or contracted pixel (step S5). When the block corresponds to a pixel that has been expanded or contracted (Yes in step S5), the feature amount changing unit 23b approaches the feature amount of the block to the block stored in the feature amount storage unit 23a. Is changed based on the feature amount of the adjacent block to be processed (step S6), and is output to the corresponding quantization parameter output unit 23d.

On the other hand, when the block does not correspond to the expanded or contracted pixel (No in step S5), the feature amount of the block is output as it is to the corresponding quantization parameter output unit 23d, and the process proceeds to step S7.
Subsequently, the adaptive quantization coding apparatus 1 uses the corresponding quantization parameter output unit 23d to read and determine the quantization parameter corresponding to the feature amount of the block from the correspondence table storage unit 23c (step S7).

(Encoding step)
Then, the adaptive quantization coding apparatus 1 performs coding by compressing and coding the block divided in step S2 based on the quantization parameter determined in step S7 by the quantization coding means 30. Data is generated (step S8).

  Then, when there is a continuing block (Yes in step S9), adaptive quantization coding apparatus 1 returns to step S2 and continues the operation. Further, if there is a next image (frame) (Yes in step S10), the process returns to step S1 to continue the operation. Then, when there is no subsequent image (frame) (No in step S10), the operation ends.

  As mentioned above, although embodiment of this invention was described, this invention is not limited to this. Hereinafter, another embodiment of the quantization control means 20 (FIG. 2) will be described with reference to the drawings.

(Configuration of quantization control means: second embodiment)
First, a second embodiment of the quantization control means (quantization control device) will be described with reference to FIG. FIG. 9 is a block diagram showing the configuration of the quantization control means.
As illustrated in FIG. 9, the quantization control unit 20 </ b> B sets a quantization parameter indicating a degree of quantization when compressing and encoding an image in units of a block having a specific size. Control based on the quantity. Here, the quantization control unit 20B includes a feature amount extraction unit 21, a correction block determination unit 22, a quantization parameter association unit 23, and a difference information calculation unit 24. The feature quantity extraction unit 21, the correction block determination unit 22, and the quantization parameter association unit 23 have the same configuration as the quantization control unit 20 described in FIG. Is omitted.

  The difference information calculation unit 24 calculates difference information in the time direction by taking the difference between the input block (local image) and the past block at the same position on the screen. Here, the difference information calculation means 24 includes a difference calculation unit 24a and a frame storage unit 24b.

  The difference calculation unit (difference calculation unit) 24a is configured to calculate a difference (pixel) between the input block (local image) and a block at the same position on the screen in the past frame stored in the frame storage unit 24b. Value difference). Since this difference takes a positive or negative value, the absolute value of the difference is calculated as difference information here. This difference information may be calculated as the square of the difference between pixel values. Here, the calculated difference information is output to the feature amount extraction unit 21. The input current block to be encoded is sequentially stored in the frame storage unit 24b.

  The frame storage unit 24b stores an input block for each frame, and is a storage device such as a memory. The frame storage unit 24b sequentially stores the blocks input to the difference calculation unit 24a and stores them as an image for at least one frame. In this way, by storing an image for one frame, the difference calculation unit 24a can calculate a difference in the time direction in the blocks at the same position on the screen.

  As described above, the quantization control unit 20B adds the difference information calculation unit 24 to the quantization control unit 20 (FIG. 2), thereby correcting the degree of quantization according to the feature amount between the screens (in the time direction). be able to. Accordingly, it is possible to suppress a rapid change in the degree of quantization not only in the screen but also in the corresponding blocks between the screens, and it is possible to suppress deterioration in image quality due to the quantization of the moving image.

  In the quantization control unit 20B, the feature quantity extracted by the feature quantity extraction unit 21 is output to the corresponding quantization parameter output unit 23d of the quantization parameter association unit 23 as it is, thereby quantizing only between the screens. It is good also as a structure which suppresses the change of the degree of. In this case, the correction block determination unit 22 and the feature amount storage unit 23a and the feature amount change unit 23b in the quantization parameter association unit 23 can be omitted from the configuration from the quantization control unit 20B.

  The quantization control means (quantization control device) 20B can be realized by causing a general computer to execute a program and operating an arithmetic device or a storage device in the computer. This program (quantization control program) can be distributed via a communication line, or can be distributed by writing in a recording medium such as a CD-ROM.

(Configuration of quantization control means: third embodiment)
Next, a third embodiment of the quantization control means (quantization control device) will be described with reference to FIG. FIG. 10 is a block diagram showing the configuration of the quantization control means.
As illustrated in FIG. 10, the quantization control unit 20 </ b> C includes a quantization parameter indicating a degree of quantization when an image is compressed and encoded in units of a block having a specific size (feature amount) that the block has. Control is based on the color information. Here, the quantization control unit 20C includes a feature amount extraction unit 21C, a correction block determination unit 22C, and a quantization parameter association unit 23C.

  The feature amount extraction unit 21C extracts color information from the input block (local image) as a unique feature amount of the local image. Here, the feature amount extraction unit 21C includes a color information extraction unit 21a.

  The color information extraction unit 21a extracts luminance Y and two color differences U and V as color information of the input block. Here, although description is given using luminance and color difference as color information, three primary colors such as RGB may be extracted. Here, the extracted color information (luminance Y, color difference U and V) is output to the correction block determination unit 22C.

  The correction block determination unit 22C quantizes the block based on the feature amount (color information) of the block extracted by the feature amount extraction unit 21C and the feature amount (color information) of a neighboring block adjacent to the block. It is determined whether or not the degree of correction is to be corrected. Here, the correction block determination unit 22C includes a flag image generation unit 22Ca, a flag image storage unit 22b, and an expansion / contraction unit 22c. Here, since the configuration other than the flag image generation unit 22Ca is the same as the configuration of the quantization control unit 20 described with reference to FIG.

  Based on the feature amount (color information) extracted by the feature amount extraction unit 21C and the color range information that is a threshold value given from the outside, the flag image generation unit 22Ca sets the feature amount of the block within the threshold range. A flag image (binary image) indicating whether it belongs or not by a value of “1” or “0” is generated. This color range information is information indicating the upper and lower limits of the luminance Y, the upper and lower limits of the color difference U, and the color range having the upper and lower limits of the color difference V as threshold values.

Then, the flag image generation unit 22Ca is “1” when the color information extracted by the color information extraction unit 21a is within the range specified by the color range information, and is “0” when it is not. A flag image (binary image) represented by a value is generated.
The flag image is subjected to expansion or contraction processing by the expansion / contraction unit 22c, and it is determined which block is a block for correcting the degree of quantization.

  The quantization parameter associating unit 23C, based on the feature amount (color information) of the block extracted by the feature amount extracting unit 21C, is a quantization parameter indicating the degree of quantization that is associated in advance with the feature amount. Is output. Further, when the correction block determination unit 22C notifies the quantization parameter association unit 23C that the target block is a block that corrects the degree of quantization (quantization parameter), the quantization of the block is quantized. It also corrects the parameters and outputs them. Here, the quantization parameter associating unit 23C includes a feature amount storage unit 23a, a feature amount change unit 23b, a correspondence table storage unit 23Cc, and a correspondence quantization parameter output unit 23d. Here, since the configuration other than the correspondence table storage unit 23Cc is the same as the configuration of the quantization control unit 20 described with reference to FIG.

  The correspondence table storage unit 23Cc stores a correspondence table in which block feature amounts (color information) and quantization parameters are associated in advance, and is a storage unit such as a hard disk. For example, as shown in FIG. 6, the quantization parameter Qp is associated with a range of luminance Y, color difference U, and V. Thereby, the quantization parameter is uniquely specified by the value of the color information that is the feature amount.

  Thus, the quantization control means 20C can change the quantization parameter according to the range of color information included in the local image. Thus, for example, by setting the color range to the skin color range, it is possible to suppress a rapid change in the degree of quantization in the human skin color portion in the image, for example, the face contour portion in the face quantization. And the deterioration of the image quality of the person can be suppressed.

It is a block diagram which shows the structure of the outline of the adaptive quantization coding apparatus which concerns on this invention. It is a block diagram which shows the structure of the quantization control means (quantization control apparatus) which concerns on this invention. It is a figure which shows an example of the flag image which a flag image generation part produces | generates. It is explanatory drawing for demonstrating expansion / contraction processing, Comprising: (a) is explanatory drawing for demonstrating the example of expansion / contraction of a flag image by 8-neighbor processing, (b) is an example of expansion of a flag image, (c ) Shows an example of contraction of the flag image. It is a figure which shows an example of the correspondence table which matched the quantization parameter and dispersion | distribution. It is a figure which shows an example of the correspondence table which matched the quantization parameter and color information. It is a block diagram which shows the structure of a quantization encoding means. It is a flowchart which shows operation | movement of the adaptive quantization coding apparatus which concerns on this invention. It is a block diagram which shows the structure of the quantization control means (quantization control apparatus) which concerns on 2nd embodiment of this invention. It is a block diagram which shows the structure of the quantization control means (quantization control apparatus) which concerns on 3rd embodiment of this invention.

Explanation of symbols

1 Adaptive quantization coding device (coding device)
10 block division means 20, 20B, 20C quantization control means (quantization control device)
21 feature amount extraction means 22 correction block determination means 22a flag image generation unit (flag image generation means)
22b Flag image storage unit 23c Expansion / contraction part (expansion / contraction means)
23 Quantization parameter association unit 24 Difference information calculation unit 30 Quantization encoding unit

Claims (5)

In an encoding apparatus that compresses and encodes an image in units of a block having a specific size, a quantization control apparatus that controls a degree of quantization of the block,
A feature amount extracting means for extracting a feature amount indicating the feature of the image from the block;
Correction block determination means for determining whether or not to correct the degree of quantization of each block based on the feature amount of each block extracted by the feature amount extraction means;
The feature quantity of the block subjected to correction by the correction block determination means is changed to a feature quantity approximate to the feature quantity of the adjacent block of the block, and the feature quantity of the changed block and the block that has not been changed. A quantization parameter associating means for outputting, for each block , a quantization parameter indicating the degree of quantization that is previously associated with the feature amount of
The correction block determination means
It is determined whether the feature amount of each block extracted by the feature amount extraction unit is included in a range specified by one or more preset threshold values, and the determination result of each block is used as a pixel value as a pixel value. Flag image generation means for generating a flag image associated therewith;
An expansion / contraction means for expanding or contracting the flag image by converting the pixel value of each pixel of the flag image generated by the flag image generation means into an adjacent pixel value;
A quantization control apparatus, characterized in that a block corresponding to a pixel subjected to expansion or contraction by the expansion / contraction means is determined as a block for correcting the degree of quantization. The difference information calculating means for calculating the difference information in the time direction by taking the difference between the block and the block of the past image in the time direction at the same position on the screen as the block, In preparation for the previous stage,
The quantization control apparatus according to claim 1, wherein the feature amount extraction unit extracts a feature amount of the block based on the difference information calculated by the difference information calculation unit. An adaptive quantization encoding device that encodes the image by dividing the image into blocks of a specific size and changing the degree of quantization according to the feature amount of the block,
Block dividing means for dividing the image into the blocks;
The quantization control device according to claim 1 or 2, wherein a quantization parameter indicating the degree of quantization corresponding to the block is output based on a feature amount of the block divided by the block dividing unit;
Based on the quantization parameter output from this quantization control device, the quantization coding means for compressing and coding the blocks divided by the block dividing means, changing the degree of quantization,
An adaptive quantization coding apparatus comprising: A quantization control method for controlling the degree of quantization of a block when an image is compressed and encoded in units of a specific size block,
A feature amount extraction step of extracting a feature amount indicating the feature of the image from the block by the feature amount extraction means ;
The flag image generation means determines whether or not the feature amount of each block extracted in the feature amount extraction step is included in a range specified by one or more preset threshold values, and the determination result of each block Is generated as a pixel value, and a flag image associated with the pixel is generated, and the expansion / contraction means converts the pixel value of each pixel of the flag image into a neighboring pixel value to expand or contract the flag image. A correction block determination step for determining that the block corresponding to the pixel subjected to expansion or contraction is a block for correcting the degree of quantization ;
The feature value of the block determined to correct the degree of quantization in the correction block determination step by the quantization parameter association unit is changed to a feature value approximate to the feature value of the neighboring block of the block, and the change A quantization parameter correction step for outputting, for each block, a quantization parameter indicating the degree of quantization previously associated with the feature amount of the block that has been changed and the feature amount of the block that has not been changed ;
The quantization control method characterized by including. In order to control the degree of quantization of a block when compressing and encoding an image in units of a specific size block, a computer is provided.
Difference information calculation means for calculating difference information in the time direction by taking a difference between the block and the block of the past image in the time direction at the same position on the screen as the block,
Feature amount extraction means for extracting the feature amount of the block from the difference information calculated by the difference information calculation means;
It is determined whether or not the feature amount of each block extracted by the feature amount extraction unit is included in a range specified by one or more preset threshold values, and the determination result of each block is used as a pixel value as a pixel value. Flag image generation means for generating a flag image associated with the flag image;
Expansion / contraction means for expanding or contracting the flag image by converting the pixel value of each pixel of the flag image generated by the flag image generation means into an adjacent pixel value;
The feature quantity of the block corresponding to the pixel that is subject to expansion or contraction by the expansion / contraction means is changed to a feature quantity approximate to the feature quantity of the neighboring block of the block, and the feature quantity of the changed block And a quantization parameter associating means for outputting, for each block, a quantization parameter indicating the degree of quantization associated in advance with the feature amount of the block that has not been changed ,
Quantization control program characterized by functioning as

Images