JP2007228519A - Image encoding device and image encoding method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce operational throughput for calculating the feature value of an image used for quantization control when encoding a moving picture. <P>SOLUTION: When encoding an I picture in an original image memory 1, a feature value arithmetic part 16 calculates a variance value for each MB, stores it in a feature value storage memory 19, calculates quantization step size information of the MB and sends it to a quantization control part 13. When encoding the P picture, a feature value prediction part 17 acquires the variance value of MB of the I or P picture that becomes the predictive image of the MB, from the featured value storage memory 19, calculates the variance value, stores it in the featured value storage memory 19, calculates quantization step size information, and sends it to the quantization control part 13. When encoding the B picture, the feature value prediction part 17 also acquires the variance value of MB of the I or P picture that becomes the predictive image of MB of the B picture, from the feature value storage memory 19, calculates the variance value, calculates a quantization step size, and sends it to the quantization control part 13. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

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

  2. Description of the Related Art Conventionally, as a technique in the field of image encoding, an encoding method represented by the MPEG (Moving Picture Experts Group) standard is known.

  FIG. 12 is a block diagram showing an example of a conventional image encoding apparatus according to the MPEG-2 standard, where 1 is an original image memory, 2 is a subtractor, 3 is an orthogonal transform unit, 4 is a quantization unit, Encoding unit, 6 transmission buffer, 7 inverse quantization unit, 8 inverse orthogonal transformation unit, 9 adder, 10 frame memory, 11 motion detection / compensation unit, 12 intra-frame prediction unit, 13 Is a quantization control unit, and 14 is an activity calculation unit.

  In the figure, in the MPEG-2 standard encoding, intra-screen predictive coding and inter-screen predictive coding are used for each screen, and a screen encoded using intra-screen predictive coding is called an I picture. There are two types of pictures that are encoded using inter-picture predictive encoding: P and B pictures. When a P picture is inter-picture prediction encoded, the I picture or P picture is used as a reference picture. When a B picture is inter-frame predictive encoded, the I or P picture and the next P picture are the reference pictures. It is used as

  In the following, a description is given for a picture in units of frames. Therefore, intra prediction encoding is referred to as intra prediction encoding, and inter prediction encoding is referred to as inter prediction encoding.

So, for example,
..., I, B1, B2, P1, B3, B4, P2, ...
When the pictures input in this order are read from the original image memory 1,
..., I, P1, B1, B2, P2, B3, B4, ...
Are read and encoded in this order. The I picture is subjected to intraframe prediction encoding, while the P1 picture is interframe prediction encoded using the encoded I picture as a reference picture, and the P2 picture is interframe predictive encoding using the encoded P1 picture as a reference picture. Is done. The B1 and B2 pictures are inter-frame predictive encoded using the encoded I and P1 pictures as reference pictures, and the B3 and B4 pictures are inter-frame predictively encoded using the encoded P1 and P2 pictures as reference pictures. . For this reason, the I picture is encoded before the P picture and the B picture, and the P picture is encoded before the B picture using this as a reference picture. Therefore, before the B1 and B2 pictures are encoded, the I and P1 pictures are encoded, and the B1 and B2 pictures are encoded using these decoded I and P1 pictures as reference pictures. Such encoding is performed using a 16 × 16 pixel macroblock (MB) or an 8 × 8 pixel or 4 × 4 pixel block obtained by dividing the macroblock (MB) as a unit block. Such an encoding unit block is hereinafter referred to as an encoding target block.

  The input image data composed of sequential picture sequences is temporarily stored in the original image memory 1, and each picture is read out in a predetermined order as described above and supplied to the subtracter 2. In the subtractor 2, when an I picture is supplied for encoding (in this way, a picture read out and supplied from the original image memory 1 for encoding is hereinafter referred to as an encoding target picture). A reference image for intra-frame prediction encoding generated by the intra-frame prediction unit 12 is also supplied, and when P and B pictures are supplied as encoding target frames, an inter-frame prediction code from the frame memory 10 is supplied. A reference image for encoding is also supplied, a pixel value of a pixel of the reference image (hereinafter referred to as a reference pixel) is subtracted for each pixel of the encoding target frame, and encoding is made up of pixels of the difference value (difference value pixel) The target picture is supplied to the orthogonal transform unit 4.

  When the first macroblock encoded at the beginning of the I picture is supplied to the subtracter 2, a pixel having a constant pixel value “128” is supplied to the subtractor 2 as a reference pixel, and the difference value from this is supplied. A macro block composed of the pixels (difference value pixels) is formed. In other macroblocks of the I picture, the preceding macroblock encoded in the same I picture becomes a reference image, and this is supplied to the subtracter 2. Here, the macro block is a block of pixels composed of 16 × 16 pixels, and is a unit block for encoding. The first macroblock of the I picture is a macroblock located at the upper left corner on the screen of this frame.

  In the orthogonal transform unit 3, the macro block is divided into four blocks of 8 × 8 pixels, and each block is orthogonally transformed by DCT (Discrete Cosine Transform) operation. The DCT coefficient obtained by the orthogonal transform is supplied to the quantization unit 4 and quantized, and the quantized data is variable-length encoded by the encoding unit 5. The image data that has been predictively encoded and compressed in this manner is temporarily stored in the transmission buffer 6, and is read and transmitted so as to have a constant bit rate.

  Also, the quantized data from the quantization unit 4 is supplied to the inverse quantization unit 7, subjected to inverse quantization processing to be restored to the original DCT coefficient, and further supplied to the inverse orthogonal transform unit 8 for inverse DCT processing. The picture data of the picture made up of difference value pixels is restored. This picture is supplied to the adder 9, and at this time, the reference image used for the subtraction unit 3 is read from the frame memory 10 or the intra-frame prediction unit 12 for this picture, and is supplied to the adder 9, and the difference pixel It is added every time and is decoded into a picture having the original pixel value. This decoded picture is stored and saved in the frame memory 10 for use as a reference picture.

  If the picture supplied to the adder 9 is an I picture, the I picture itself is read out from the intra-frame prediction unit 12 as a reference picture and supplied to the adder 9. When the macroblock is supplied, a predetermined pixel value “128” is read from the intra-frame prediction unit 12 and supplied to the adder 9. Thereby, each pixel of this leading macroblock is decoded into a pixel having the original pixel value.

  The decoded first macroblock of the I picture is supplied to and stored in the frame memory 10 and is also supplied to the intra-frame prediction unit 12 to generate a reference image of the macroblock to be encoded next to the I picture. Is done. Therefore, when the second macroblock is encoded, the reference image of the first macroblock is supplied to the subtracting unit 2, and the subtraction process with the second macroblock is performed to perform intra prediction encoding. At the same time, the encoded second macroblock is decoded by the adder 9 with the same reference image, stored in the frame memory 10, and supplied to the intra-frame prediction unit 12 for further subsequent macroblocks (third A reference image for encoding (macroblock) is formed. In this way, the macroblocks of the I picture are sequentially encoded, and the decoded macroblocks are sequentially stored in the frame memory 10. When the intra-picture encoding of the I picture is completed, the frame memory 10 The decoded I picture is stored in the frame memory 10 and used as a reference picture for the next P picture.

  When inter-picture prediction encoding is performed on the next P picture after the I picture, the same processing as described above is performed using the I picture stored in the frame memory 10 as a reference image. In this case, the P picture to be encoded The I picture stored in the frame memory 10 is supplied to the motion detection / compensation unit 11, and a motion vector MV is detected for each macroblock. Then, based on the motion vector MV, an I-picture macroblock for a P-picture macroblock supplied to the subtractor 2 is detected, and this is supplied to the subtractor 2 to form a macroblock composed of difference value pixels. The macroblock is encoded and supplied to the adder 9 to decode the encoded macroblock, which is supplied to the frame memory 10 and stored therein.

  In the case of inter-picture predictive coding of a P picture that is coded after the coded P picture, the motion detection / motion compensation unit 11 processes the decoded P picture stored in the frame memory 10 for reference. Other than this, it is the same as the P picture encoding process described above. When a B picture is encoded, a reference obtained by processing a decoded picture of an already encoded I or P picture stored in the frame memory 10 with a motion vector of the motion detection / motion compensation unit 11 is obtained. An image is used. However, since the B picture is not used as a reference image, the B picture is not stored in the frame memory 10.

  When the P and B pictures are encoded, the motion vector MV detected by the motion detection / compensation unit 11 is supplied to the encoding unit 5 for each macroblock. In the encoding unit 5, the motion vector MV is added to the image data from the quantization unit 4 of the macroblock in which the motion vector MV is used as described above by the subtracter 2 of the encoding target picture. Variable length coding.

  As described above, the data is encoded into image data including an array of I, P, and B pictures, and is transmitted from the transmission buffer 6 at a constant bit rate. 13 monitors the image data, controls the quantization unit 4 according to the monitoring result, and transmits the quantization step size QS (quantization level interval) in the quantization unit 4 so as to be transmitted at a constant bit rate. ). Hereinafter, this point will be described.

  The quantization control unit 13 controls the data allocation amount (allocation code amount) for each picture in order to transmit the encoded image data at a predetermined data amount (bit rate). For example, in the MPEG encoding system, there are three types of pictures (I, P, and B pictures described above) having different properties (encoding systems), and therefore the allocated code amount is determined according to the type of picture. That is, when I and P pictures are used as reference pictures for P and B pictures, the allocated code amount for I and P pictures is made larger than that for B pictures. It is also possible to divide a picture into blocks after determining the allocated code amount in units of pictures and determine the allocated code amount in units of blocks.

In the MPEG-2 Test Model 5 (hereinafter referred to as TM5) quantization control method used in the MPEG-1 standard,
Step 1: Assign a code amount to each picture Step 2: Use a virtual buffer to quantize each block in the picture
Determining the meter QP Step 3: Considering the visual characteristics in the picture, the quantization parameter of each block
A three-step process is performed in which the quantization step size is determined by weighting the data QP (see, for example, Non-Patent Document 1 and Non-Patent Document 2).

  In step 3, human visual characteristics that distortion is conspicuous in the flat part of the image and inconspicuous in the complicated part of the image pattern are used. Use. Then, the quantization step size QS is changed based on the activity Act. The activity calculation unit 14 obtains an activity Act for each picture, and the quantization control unit 13 controls the quantization step size QS in the quantization unit 4 based on the obtained child activity Act.

As specific processing in the activity calculation unit 14, first, the picture is divided into macro blocks each having a size of 16 × 16 pixels. Next, as shown in FIG. 13, the macroblock is divided into four blocks of 8 × 8 pixels, and the variance value of the luminance signal is calculated for each block. The variance value Var [n] of the nth block is expressed by the following Equation 1,

Obtained by the calculation formula shown in. Here, the first block (n = 1) is the upper left 8 × 8 pixel block 1 among the four blocks in the macroblock shown in FIG. 13, and the second block (n = 2). Is the upper right 8 × 8 pixel block 2, the third block (n = 3) is the lower left 8 × 8 pixel block 3, and the fourth block (n = 4) is the lower right This is a block 4 of 8 × 8 pixels. In Equation 1, P _n (x, y) is the pixel value of the pixel at the coordinate position (x, y) in the nth block.

Next, the activity Act of this macroblock is obtained from the variance value Var [n] of the four blocks shown in the above equation 1. This activity Act has the following number 2, namely:

Is calculated from Here, min () is a function for selecting the variance value Var [n] that becomes the minimum value in the parentheses, and the variance value Var [n] that becomes the minimum value is selected in one macro block. This is because when there is a flat part even in only a part, the quantization step size QS is reduced to make the quantization finer.

Next, the weighting coefficient N _act of the quantization step size QS is _expressed by the following equation 3, that is,

Calculate with Here, Avg_act is an average value of activities of all macroblocks in the picture encoded immediately before.

Subsequently, the quantization parameter QP calculated and set in the above step 2 is multiplied by the weighting coefficient N _act obtained in Equation 3 as shown in _Equation 4, and the quantization in the quantization unit 4 for the encoding target block is performed. The step size QS is calculated.

  The quantization step size QS obtained in this way is supplied to the quantization control unit 13, and the quantization control unit 13 determines the quantization level in the quantization unit 4 based on the quantization step size QS. .

  The quantization control unit 13 is also supplied with bit amount information from the transmission buffer 6, adjusts the quantization step size QS based on the information, and transmits from the transmission buffer 6 at a constant bit rate. The quantization level in the quantization unit 4 is adjusted.

  By the above method, in a complex image region with a pattern that is not easily noticeable visually, the quantization step size in the quantizer 4 is increased to perform rough quantization, and a flat image that is easily noticeable visually deteriorates. In the area, the quantization step size in the quantization unit 4 is reduced to perform fine quantization, and the generated code amount is controlled according to the image quality.

  However, in the above-described quantization control method, the variance value becomes large at the edge portion where deterioration is conspicuous and is roughly quantized. For this reason, the deterioration of the edge portion becomes conspicuous, and at a low bit rate, the quantization step (quantization level) inevitably becomes rough, and the block distortion increases, resulting in a visually uncomfortable feeling.

Therefore, in order to solve the problem, a technique that enables high-compression and high-quality encoding by performing preprocessing on original image data before encoding has been proposed (for example, , See Patent Document 1 and Patent Document 2).
MPEG-2, Test Model 5 (TM5), Doc.ISO / IEC JTC1 / SC29 / WG11 / N0400, Test Model Editing Committee, Apr. 1993. Joint Video Team (JVT) of ISO / IEC MPEG & ITU-T VCEG: "Text of ISO / IEC 14496 10 Advanced Video Coding 3rd Edition", Jul, 2004. JP2002-16923 JP2002-10268

  By the way, in the above preprocessing in Patent Document 1 and Patent Document 2, as the difference between the encoded original image data and the decoded image data becomes larger, filter processing for limiting the frequency band and spatial processing of the original image data are performed. DFT (Discrete Fourier Transform) and DCT orthogonal transform processing are required to acquire accurate frequency information. In addition, a mid-high pass filter process for emphasizing the edge portion is also required. Therefore, when these pre-processing is performed on all of the I, P, and B pictures, the calculation processing time is significantly increased. In particular, the P picture and the B picture are compared with the I picture by the motion search process. There was a problem that the amount of processing became very large.

  In addition, when encoding a large image such as an HD (High Definition) image, it is difficult to perform the encoding process in real time.

  An object of the present invention is to provide an image coding apparatus and an image coding method that can solve such a problem and that can substantially reduce the amount of calculation spent for calculating a feature amount while substantially maintaining image quality. It is in.

  In order to achieve the above object, according to the present invention, a feature amount of an I or P picture is calculated in an image feature amount calculation process including an activity, and a feature amount of a P or B picture is calculated based on the feature amount. This is a prediction, and realizes a significant reduction in the amount of calculation spent on the feature amount calculation while substantially maintaining the image quality.

  Specifically, the present invention is an image encoding device that encodes moving image information including a series of pictures, a feature amount calculating unit that calculates a feature amount of a picture to be encoded, and a feature amount calculation unit. Characteristic amount storage means for storing the variance value at the time of the calculation, and feature quantity prediction means for calculating the feature value of the picture to be encoded based on the variance value stored in the feature value storage means. It is a feature.

  In the image coding apparatus according to the present invention, the picture whose feature value is calculated by the feature value calculating unit is an I picture that is predictive-encoded in the screen, and the feature value is calculated by the feature value predicting unit. The pictures are P pictures and B pictures that are inter-picture predictively encoded, and the variance value of the P picture is stored in the feature amount storage means and the already encoded I picture or P picture Based on the variance value, calculated by the feature amount prediction means, the calculated variance value is stored in the feature amount storage means, and the variance value of the B picture is already stored in the feature amount storage means The feature amount is calculated by the feature amount prediction unit based on the encoded feature amount of the I picture or the P picture, and storage in the feature amount storage unit is prohibited.

  Also, in the image encoding device according to the present invention, the feature amount calculation means divides the I picture into blocks, and generates the variance value for each block based on the pixel value of each pixel of the block. It is characterized by.

  The image coding apparatus according to the present invention includes means for dividing the I picture into blocks, and for each block, using a difference value from a reference image in the I picture as a pixel, wherein the feature amount calculating means includes: The variance value is generated for each block based on the pixel value of the difference value.

  Also, in the image coding apparatus according to the present invention, the feature quantity prediction means divides the P and B pictures into blocks, and the feature quantity storage means that is a reference block for the block for each block. The variance value is generated based on the variance value of the block or its neighboring blocks.

  In the image coding apparatus according to the present invention, the feature amount calculated by the feature amount calculation unit is a quantization step size when the I picture is encoded, and is calculated by the feature amount prediction unit. The feature amount is a quantization step size when the P and B pictures are encoded.

  In the image encoding device according to the present invention, the feature prediction means predicts a motion vector for each block to be encoded of the P picture based on a motion vector of a block already encoded around the block. The block of the block to be encoded by predicting the reference block of the I or P picture based on the motion vector, obtaining a variance value of the predicted reference block or its neighboring blocks from the feature amount storage means It is characterized by obtaining a value.

  Further, in the image encoding device according to the present invention, the feature prediction means includes, for each block to be encoded of the B picture, a forward prediction motion vector for the I or P picture forward in the display order of the B block and the B block. Predicting a backward predicted motion vector for a backward P picture in display order, predicting the forward reference block in the forward I or P picture based on the forward predicted motion vector, and Based on this, the reference block in the back in the back P picture is predicted, and a variance value of the reference block in the front or its surrounding block and a variance value of the reference block in the back or its surrounding block are used as the feature amount It is characterized in that the variance value of the block to be encoded is obtained from the storage means.

  The image coding apparatus according to the present invention further includes a feature amount prediction process determination unit that determines whether or not a feature amount is predicted by the feature amount prediction unit, and the feature amount prediction process determination unit determines whether the feature amount prediction process is performed. When it is determined that the feature amount is not predicted, the feature amount calculation unit calculates the feature amount, and the variance value at the time of the calculation is stored in the feature amount storage unit.

  The image encoding apparatus according to the present invention further includes a scene change detection unit for detecting a scene change of a moving image, and the feature amount prediction unit determines a feature amount for the scene-changed picture in the feature amount prediction processing determination unit. It is characterized in that it is determined not to predict.

  The image encoding apparatus according to the present invention is characterized in that information on a quantization step size obtained from the variance value is stored in the feature quantity storage means instead of the variance value.

  The image encoding apparatus according to the present invention is characterized in that the feature value storage means stores a pixel value after pre-processing a picture to be encoded.

  The image encoding device according to the present invention is characterized in that the preprocessing is processing for enhancing an edge portion of an image.

  The present invention also relates to an image encoding method for encoding moving picture information composed of a series of I, P, and B pictures, wherein the feature amount of the I picture is calculated when the I picture is encoded. When the P and B pictures are encoded, the feature amount is calculated based on the stored feature amount, and the calculated feature amount of the P picture is stored and held. It is characterized by this.

  In the image coding method according to the present invention, the stored feature amount is a variance value calculated for each divided block of the I, P, and B pictures. Information for determining a quantization step size at the time of encoding P and B pictures is calculated.

  According to the present invention, in the moving picture coding system such as the MPEG standard system, the processing for calculating the feature amount of the image used for the quantization control can be greatly reduced, and the image quality is substantially maintained. However, the encoding process can be performed at high speed.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 is a block diagram showing the configuration of an embodiment of an image encoding apparatus and an image encoding method according to the present invention, in which 15 is a picture type determining unit, 16 is a feature amount calculating unit, and 17 is a feature amount predicting unit. , 18 is a motion vector memory, and 19 is a feature quantity storage memory. The same reference numerals are assigned to the portions corresponding to those in FIG.

  In the figure, the first embodiment is different from the configuration of the conventional image coding apparatus shown in FIG. 12 in that a picture type determination unit 15, a feature amount calculation unit 16, a feature amount prediction is used instead of the activity calculation unit 14. Unit 17, motion vector memory 18, and feature quantity storage memory 19. A feature quantity such as an activity is obtained from an I picture, and a feature quantity of P and B pictures is obtained from this I picture. The quantization control unit 13 controls the quantization step size of the quantization unit 4 in accordance with these feature amounts.

  When an encoding target picture (here, referred to as a frame) read out from the original image memory 1 in the above order is encoded by the subtractor 2 or the orthogonal transform unit 3, the picture type determination unit 15 performs the encoding process. It is determined whether the target picture is an I, P, or B picture, and the determination result information CR is supplied to the feature amount calculation unit 16 and the feature amount prediction unit 17. When the determination result information CR indicates that the picture is an I picture, the feature amount calculation unit 16 starts operating, and the image data of the I picture is supplied from the original image memory 1 to obtain the feature amount and the like. The calculation process for this is started. When the determination result information CR indicates that the picture is a P or B picture, the feature quantity prediction unit 17 starts operation, and the feature quantity of these P or B pictures is started using the feature quantity of the I or P picture. The arithmetic processing for obtaining the above is started.

  In the following, the I, P, and B pictures to be encoded that are processed by the feature amount calculating unit 16 and the feature amount predicting unit 17 are referred to as the calculation target I, P, and B pictures, respectively. Further, when the feature amount prediction unit 17 performs a calculation process for a calculation processing target P or B picture, the calculation processing target P or B picture is supplied to the feature amount prediction unit 17 as described above. Instead, the feature amount of the I picture obtained by the feature amount calculation unit 16 is used.

  In the feature amount calculation unit 16, when the picture type determination unit 15 determines that the picture is an I picture and the calculation target I picture is supplied from the original image memory 1, the feature amount including the activity Act becomes a macroblock (MB ) And the variance value Var for each block used to calculate the activity Act is stored in the feature amount storage memory 19. Then, when the same picture as the calculation processing target I picture is read from the original image memory 1 and encoded as the encoding target I picture as described above, the activity calculated by the feature amount calculation unit 16 is calculated. The quantization step size QS is obtained from Act and is sent to the quantization control unit 13. As a result, the quantization step size in the quantization unit 4 and thus the quantization level interval are set for each macroblock for the I picture to be encoded.

  When the picture type determination unit 15 determines that the picture read and encoded from the original image memory 1 is a P picture, the feature amount calculation unit 16 starts its operation. The variance value Var is calculated for each block (operation processing target block). The variance value Var of the operation processing target block is calculated using the variance value stored in the feature amount storage memory 19. For this reason, the necessary variance value is obtained from the feature amount storage memory 19. It is read by the quantity prediction unit 17. The motion vector MV detected by the motion detection / compensation unit 11 is used for reading the dispersion value. That is, in the feature quantity predicting unit 17, when a block to be subjected to the computation process of the computation target P picture is determined, a block that becomes a reference image for the computation target block (reference block: a variance value corresponding thereto is stored in the feature quantity (Stored in the memory 19) is searched using this motion vector MV, and the variance value of this reference block or its surrounding blocks is fetched from the feature amount storage memory 19, and based on this, the variance of the block to be processed is calculated. Value.

  The feature amount prediction unit 17 further predicts, for each macroblock, a feature amount including an activity from the variance value Var for each block of the P picture to be calculated in this way, and uses the variance value Var as the feature value. It is stored in the quantity storage memory 19. When the encoding target P picture is encoded as described above, the feature amount prediction unit 17 obtains the quantization step size QS from the activity Act calculated from the calculation processing target P picture. Is sent to the quantization control unit 13. As a result, the quantization step size in the quantization unit 4 and thus the quantization level interval are set for each macroblock for the P picture to be encoded. The motion vector MV captured from the motion detection / compensation unit 11 is held in the motion vector memory 18 until such processing for the calculation processing target P picture is completed.

  When a preceding P picture in which a P picture has already been encoded is encoded as a reference image, the variance value Var for each block of the preceding P picture is also stored in the feature amount storage memory 19. When obtaining the variance value for each block of the P picture to be processed with respect to the P picture to be coded, the reference block in the preceding P picture for this block is obtained using the motion vector MV in the same manner as described above. Are distributed from the feature amount storage memory 19.

  When a B picture is read out from the original image memory 1 and is encoded, the reference picture of the encoding target B picture is an I picture that has already been encoded before and after this B picture, and is encoded next. Or two already encoded P pictures before and after this B picture. Therefore, when the B picture in the original image memory 1 is encoded, the feature quantity storage memory 19 already stores the feature quantity (variance value Var) of the I picture or P picture that is the reference picture.

  Therefore, when the picture type determination unit 15 determines that the picture read and encoded from the original image memory 1 is a B picture, the feature amount prediction unit 17 starts the operation, and the feature quantity prediction unit 17 For each block (arithmetic processing target block), a motion vector MV detected by the motion detection / compensation unit 11 is used to search for a block (reference block) of an arithmetically processed I picture or P picture that becomes a reference image, The variance value Var of this reference block or its peripheral blocks is fetched from the feature amount storage memory 19. Then, based on the fetched variance value Var, the variance value Var of this operation processing target block is obtained. Further, the feature amount predicting unit 17 predicts a feature amount including the activity Act of the B picture to be processed for each macroblock based on the variance value Var obtained in this way, and this predicted activity The quantization step size QS is obtained from Act and is sent to the quantization control unit 13. Thereby, with respect to the encoding target B picture, the quantization step size in the quantization unit 4 and thus the quantization level interval are set for each macroblock. The motion vector MV captured from the motion detection / compensation unit 11 is held in the motion vector memory 18 until such processing for the calculation processing target B picture is completed.

  In this way, the feature quantities of the I, P, and B pictures are calculated, and the respective quantization step sizes QS are obtained using the feature quantities. At this time, the feature amounts of the I and P pictures as reference images of the P picture and B picture are stored in the feature amount storage memory 19 until they are not used.

  FIG. 2 is a flowchart showing the flow of processing performed by the picture type determination unit 15, the feature amount calculation unit 16, the feature amount prediction unit 17, and the feature amount storage memory 19 described above. Explained.

  When a picture read out from the original image memory 1 and encoded is determined as an I picture by the picture type determination unit 15 (“YES” in step 101 in FIG. 2), the feature amount calculation unit 16 operates. This picture is supplied as an arithmetic processing target I picture, and its feature amount is calculated for each macroblock (step 102 in FIG. 2).

  Here, this feature amount will be described using the activity Act as an example.

  Here, for calculating the activity Act, Formula 1 for calculating the previous variance value Var and Formula 2 for calculating the activity Act are used, but Formula 1 is calculated for each 8 × 8 pixel block for calculating the Activity Act. Instead of using the variance value Var obtained in step 1, the average value is obtained from the pixel values in the 8 × 8 pixel block, and the sum of the absolute values of the differences between the average value and each pixel value in the block is required. You may use the value calculated | required with the other formula about a block. Further, the block size can be applied to any size as long as it is composed of a plurality of pixels such as a 16 × 16 pixel unit or a 4 × 4 pixel unit instead of an 8 × 8 pixel unit.

  Next, the variance value Var of each block obtained when calculating the activity Act in step 102 in FIG. 2 is stored in the feature amount storage memory 19 (step 103 in FIG. 2). As a storage method in the feature quantity storage memory 19, for example, an MB address that determines the storage address of the macroblock is set for each I and P picture, and the computation target I is sequentially calculated by the feature quantity computation unit 16. The variance value Var of each block in the macro block of the picture is stored in the corresponding MB address of the feature amount storage memory 19.

  When the encoding target I picture corresponding to the arithmetic processing target I picture is read from the original image memory 1 and encoded, the feature amount calculation unit 16 uses the calculated activity act based on the above equations 3 and 4. A quantization step size QS is calculated. Then, the quantization step size QS is supplied to the quantization control unit 13, whereby the quantization step size is controlled by the quantization unit 4 to perform the I picture encoding process (step 106 in FIG. 2).

  The above processing is repeated until all the macroblocks in the calculation target I picture are completed (step 107 in FIG. 2).

  When the picture read out from the original image memory 1 and encoded is determined as a P picture by the picture type determination unit 15 (“NO” in step 101 of FIG. 2), the feature amount prediction unit 17 The operation is started, and the activity Act is predicted using the variance value Var of the I picture stored in the feature amount storage memory 19 (step 104 in FIG. 2).

  FIG. 3 is a diagram showing a specific example of a part of the activity Act prediction process for the P picture to be processed.

  In the figure, the encoding target block in the encoding target P picture corresponding to the operation target P picture is X, the adjacent block on the left side is A, the adjacent block on the upper side is B, and the adjacent block on the right side of the adjacent block is C. These adjacent blocks A, B, and C are already encoded blocks, and motion vectors MVA, MVB, and MVC for I or P pictures as reference screens have already been obtained.

  In the feature prediction unit 17, if the block that is the target of the arithmetic processing target P picture is the block X, the block X that is the target of the arithmetic processing is not yet encoded, so the motion vector MV is obtained. Although not coded, motion vectors MVA, MVB, and MVC are obtained and stored in the motion vector memory 18 in the adjacent blocks A, B, and C that have already been encoded. Therefore, the feature prediction unit 17 acquires the motion vectors MVA, MVB, and MVC of these adjacent blocks A, B, and C from the motion vector memory 18, and the motion vector having an intermediate absolute value among them is the operation processing target block. Let XV be a motion vector (hereinafter referred to as a predicted motion vector) PMV. Then, the feature quantity predicting unit 17 specifies an I picture or P picture block (reference block) which is a reference image at the time of encoding based on the predicted motion vector PMV, and its variance value Var or its peripheral blocks Is obtained from the feature quantity storage memory 19, and based on this, the variance value Var of the operation processing target block X is obtained. For this purpose, the motion vector MV for each block of the calculation target P picture detected by the motion detection / motion compensation unit 11 is stored and held in the motion vector memory 18.

  Note that the motion vector MVA, MVB, and MVC of the adjacent blocks A, B, and C may be the minimum motion vector or vector average vector as the predicted motion vector PMV.

  In addition, there are blocks in which there are no adjacent blocks corresponding to the peripheral blocks A, B, and C, such as the upper side block on the screen of the encoding target P picture. For such a block, when a motion vector MV detected by the motion detection / compensation unit 11 is supplied to the feature prediction unit 17 when such a block becomes an encoding target block (which is of course also a motion) (Preserved in the vector memory 18) directly as the predicted motion vector of the operation target block for this encoding target block, and using this, the corresponding reference block is identified from the feature amount storage memory 19 as described above, The variance value Var is read and set as the variance value Var of the operation processing target block.

  In this way, with respect to the P picture to be processed, for a part of the blocks, the motion detection / motion compensation unit 11 waits for the motion vector MV to be detected, and the detected motion vector MV is processed. Although the prediction motion vector of the target block is used, for most blocks, the prediction motion vector MV is obtained based on the motion vector of an adjacent block that has already been arithmetically processed for the P picture to be processed. The calculation processing amount for acquiring the vector is greatly reduced, and a quick process can be performed.

  FIG. 4 is a diagram showing a specific example of a method for obtaining the variance value Var of the operation processing target block X in the operation processing target P picture. In FIG. 4, 20 is an operation processing target macroblock of the operation processing target P picture, and 21 is This is a part of an I or P picture that has been subjected to arithmetic processing as a reference image for this arithmetic processing target P block.

  In the figure, an arithmetic processing target macroblock 20 composed of 16 × 16 pixels is composed of four blocks A, B, C, and D composed of 8 × 8 pixels. Further, the I or P picture 21 that has been subjected to the arithmetic processing is also divided into blocks each having 8 × 8 pixels. Here, the picture 21 is described as an I picture, but the same applies to a P picture.

  Now, the upper left arithmetic processing target block A in the arithmetic processing target macroblock 20 will be described. An area of 8 × 8 pixels on the I picture 21 determined by the predicted motion vector PMV obtained from the arithmetic processing target block A as described above is defined as an area X ′, and on the I picture 21 closest to the area X ′. If the block A ′ is a block A ′, this block A ′ becomes a reference block of the operation processing target block A. Then, the variance value Var of the reference block A ′ is set as the variance value Var of the operation processing target block A. However, the present invention is not limited to this, and the variance value of the operation processing target block A is obtained by weighting and averaging the variance values Var of any of the blocks E, F,..., L around the block A ′. It may be Var.

  Similarly, the variance value Var is also obtained for the other blocks B, C, and D of the operation processing target macroblock 20, and the activity Act of the operation processing target macroblock 20 is calculated from these distribution values Var by the above-described Expression 2. To do. At this time, since the operation processing target picture is a P picture (step 105 in FIG. 2), the variance value Var of these blocks A, B, C, and D is stored in the feature amount as in the case of the operation processing target I picture. The data is stored in the memory 19 (step 106 in FIG. 2).

  The above processing is repeated until all the macroblocks in the calculation target P picture are completed (step 107 in FIG. 2).

  In FIG. 2, when a picture read and encoded from the original image memory 1 is determined as a B picture by the picture type determination unit 15 (“NO” in step 101 in FIG. 2), the feature amount The activity Act is predicted using the variance values of the I and P pictures stored in the storage memory 19 (step 104 in FIG. 2).

  FIG. 5 is a diagram showing a specific example of a process for obtaining a predicted motion vector in a calculation process for predicting an activity Act in a B picture.

  In the drawing, as shown in the figure, the B2 picture is set as the operation processing picture in the display order arrangement of the pictures P1, B1, B2, and P2. In this case, the P1 and P2 pictures are P pictures as reference pictures of the calculation target picture, and have already been calculated and the variance value Var of the block is stored in the feature amount storage memory 19 (of course, The P1 and P2 pictures as encoding targets have already been encoded, and these become reference pictures for the arithmetic processing target picture B2. Also, the motion vector MVC01 from the P1 picture block to the P2 picture reference block has already been obtained by the motion detection / motion compensation unit 11 and taken into the motion vector memory 18. B1 and B2 pictures are B pictures. Even if the P1 picture is an I picture, the following calculation processing is the same.

  Here, assuming that the operation processing target block in the operation processing target picture B2 is a block X, first, the motion vector in which the position of the motion vector PMVL0 from the reference picture P1 in the operation processing target picture B2 is the operation processing target block X. PMVL0 and the motion vector PMVL1 from the operation processing target block X to the reference picture P2 are predicted and obtained. As a method thereof, these are approximated using the motion vector MVCol from the reference picture P1 to the reference picture P2. Use the method.

  As this method, first, a reference picture that is temporally closest to the operation processing target block B2 is determined. In this case, it is the reference picture P2. Next, a block having the same positional relationship as the spatial positional relationship of the operation processing target block X in the operation processing target picture B2 on the reference picture P2 is obtained. This block is called block B. Next, the motion vector MVC01 in which the block closest to the position in the reference picture P2 by the motion vector MVC01 from the reference picture P1 is the block B is obtained. Such a motion vector MVCol can be obtained from the motion vector memory 18 (FIG. 1).

When such a motion vector MVCol is obtained, the following equation 5 that divides the time in accordance with the time interval T1 between the pictures P1 and B2 and the time interval T2 between the reference pictures P1 and P2,

Thus, the predicted motion vectors PMVL0 and PMVL1 of the operation processing target block X are obtained. Here, the predicted motion vector PMVL0 is mainly a motion vector calculated by forward prediction, and the predicted motion vector PMVL1 is a motion vector calculated by backward prediction.

  Further, as another method for obtaining the predicted motion vector of the operation processing target block of the operation processing target picture B, as shown in FIG. 6, the operation processed block A around the operation processing target block X in the operation processing target picture B is shown. , B, C motion vectors MVA, MVB, MVC can also be used to determine the predicted motion vectors PMVL0, PMVL1 of the operation processing target block X. In this method, an arithmetically processed picture that is closest to the arithmetic processing target picture B2 in display time order is used as a reference picture, and an intermediate value of motion vectors of arithmetically processed blocks A, B, and C is set to the arithmetic processing target block X. This is a predicted motion vector.

  In this way, when the predicted motion vectors PMVL0 and PMVL0 for the reference pictures P1 and P2 of the operation processing target block X are obtained, the feature quantity predicting unit 17 (FIG. 1) performs the reference picture P1 similarly to the method described in FIG. The block closest to the position of the predicted motion vector PMVL0 from the operation processing target block X in FIG. 5 is used as a reference block, and the variance value Var (P1) of this reference block is read from the feature quantity storage memory 19. Similarly, the block closest to the position of the predicted motion vector PMVL1 from the operation processing target block X in the reference picture P2 is used as a reference block, and the variance value Var (P2) of this reference block is read from the feature quantity storage memory 19. Of course, as described with reference to FIG. 4, the variance values Var (P1) and Var (P2) may be obtained using the variance values Var of the neighboring blocks. Then, based on these variance values Var (P1) and Var (P2), the variance value Var of the operation processing target block X is obtained. As the method, for example, these variance values Var (P1) and Var (P2) are obtained. ) Is averaged by weighting according to time intervals T1, (T2-T1) from the operation target picture B2 to the pictures P1, P2 as reference images.

  In the same manner, the variance value Var is obtained for all the operation processing target blocks in the same macroblock, and the activity Act in this operation processing target macroblock is obtained using the previous equation 2, and the previous equation 3 is used as the basis. Then, the weighting coefficient Nact of the quantization step size in the operation processing target macroblock is calculated, and the quantization step size QS is obtained based on the above equation 4. Here, if the activity Act obtained based on the predicted motion vector PMVL0 is Act_PMVL0, and the activity Act obtained based on the predicted motion vector PMVL0 is Act_PMVL1, the weighting coefficient Nact of the quantization step size is calculated by the above Equation 3. In this case, the average value of Act_PMVL0, Act_PMVL1, Act_PMVL0 and Act_PMVL1, or the value obtained by weighting and adding Act_PMVL0 and Act_PMVL1 can be used.

  The above is the processing in step 104 in FIG. 2 for the B picture to be processed, and this B picture to be processed does not become a reference picture (“NO” in step 105 in FIG. 2). Without storing Var in the feature quantity storage memory 19, the obtained quantization step size QS is sent to the quantization control unit 13 (FIG. 1), and the quantization step size is controlled by the quantization unit 4.

  As described above, in the first embodiment, for the P and B pictures, the variance value Var is predicted using the variance value Var of the reference picture, and the quantization step size of the quantization unit 4 is based on this. Therefore, the processing amount for calculating the feature amount of the image used for the quantization control can be greatly reduced, and the encoding process can be speeded up without substantially affecting the image quality. it can.

  FIG. 7 is a block diagram showing a second embodiment of the image coding apparatus and image coding method according to the present invention, 22 is a difference information storage memory, and the same reference numerals are given to portions corresponding to FIG. A duplicate description will be omitted.

  In this figure, this second embodiment is obtained when predictive encoding of an I picture is made by using, for example, the encoding method according to the H.264 / AVC standard described in Non-Patent Document 2 above. By calculating the feature amount using the difference information, the calculation processing amount for calculating the feature amount of the I picture can also be reduced. For this purpose, the difference information storage memory 22 is added to the configuration of the first embodiment shown in FIG. 1, and the obtained difference information is stored so that it can be used for the calculation for obtaining the feature amount of the I picture.

  Hereinafter, the operation of the feature amount calculation unit 16 for calculating the feature amount of the I picture will be described.

  In the H.264 / AVC standard, for encoding an I picture, a process of predicting and generating a reference image using a pixel value of a block adjacent to an encoding target block called an intra (intraframe) prediction block is performed. In the H.264 / AVC standard, a plurality of types are defined as the prediction (intra prediction) mode, and any one of them can be arbitrarily selected.

  FIG. 8 is a diagram showing an intra prediction mode defined by the H.264 / AVC standard.

  In the intra prediction mode in the H.264 / AVC standard, a specific pixel of a peripheral block of an encoding target block is set as a prediction pixel of the encoding target block, and a direction determined from the prediction pixel (this is referred to as a prediction direction) The prediction of the encoding target block is performed by obtaining the difference value between the pixel value of the pixel in the encoding target block and the prediction pixel. A plurality of prediction directions are defined, and one of them can be arbitrarily selected and used.

  FIG. 8A shows a prediction mode when the encoding target block X is a macroblock having 16 × 16 pixels, that is, an intra 16 × 16 prediction mode. In the intra 16 × 16 prediction mode, each pixel in the lowermost pixel row 23a of the macroblock (hereinafter referred to as upper MB) as the upper adjacent block of the macroblock as the encoding target block X and the adjacent block on the left side As a prediction pixel, either one or both of the pixels in the rightmost pixel row 23b of the macroblock (hereinafter referred to as the left MB). An arrow 24a indicates one example of the prediction direction, and an arrow 24b indicates another example of the prediction direction.

  If the prediction direction is selected in the downward direction as indicated by the arrow 24a, 16 pixels in the upper MB pixel column 23a are prediction pixels of the encoding target block X. For each prediction pixel, the difference between the pixel value and the pixel value of the prediction pixel for all the pixels in the encoding target block X arranged in the downward direction from the prediction pixel (that is, the encoding target pixel). A value is calculated. Therefore, the encoding target block X is composed of pixels having such a difference value. Such a prediction pixel is created in the intra-frame prediction memory 12, and difference processing between the encoding target pixel in the encoding target block X and the corresponding prediction pixel is performed by the subtracter 2.

  When the prediction direction is selected in the right direction as indicated by the arrow 24b, 16 pixels in the left MB pixel column 23b are prediction pixels of the encoding target block X. Then, for each prediction pixel, for all the encoding target pixels in the encoding target block X arranged in the right direction from the prediction pixel, a difference value between the pixel value and the pixel value of the prediction pixel is calculated. . Therefore, the encoding target block X in this case is also composed of pixels having such a difference value.

  In the intra 4 × 4 prediction mode or the intra 8 × 8 prediction mode in the H.264 / AVC standard, nine prediction directions are determined as shown in FIG. That is, the prediction direction 0 (Vertical) in the downward vertical direction, the prediction direction 1 (Horizontal) in the rightward horizontal direction, and the prediction direction 2 (the average value of the pixel values of the specific pixels in the adjacent blocks is the pixel value of the prediction pixel ( DC), angle π / 4 downward prediction direction 3 (Diagonal Down Left), angle π / 4 downward prediction direction 4 (Diagonal Down Right), angle 3π / 8 downward prediction direction 5 (Vertical Right), There are nine types of prediction direction 6 (Horizontal Down) with an angle π / 8 downward to the right, prediction direction 7 (Vertical Left) with an angle 3π / 8 downward to the left, and prediction direction 8 (Horizontal Up) with an angle π / 8 in the upper right direction.

  In the intra 16 × 16 prediction mode, four prediction directions of 0 (Vertical), 1 (Horizontal), 2 (DC), and 3 (Diagonal Down Left) are defined, and one of them is arbitrarily selected. You can choose. Therefore, in the intra 16 × 16 prediction method, there are four types of intra prediction modes. In addition, the prediction direction 0 (DC) uses the average value of the pixel values of the previous pixel of the encoding target block X as the pixel value of the prediction pixel, and all the pixels of the encoding target block X have the predicted pixel. The difference value is taken between.

  FIG. 8B shows a prediction mode when the encoding target block X is a block composed of 4 × 4 pixels, that is, an intra 4 × 4 prediction mode. Therefore, one macro block is composed of four encoding target blocks. In the intra 4 × 4 prediction mode, each pixel in the lowermost pixel column of the upper block of the encoding target block X, each pixel in the rightmost pixel column of the left block, the pixel in the lower right corner of the upper left block, and the upper right pixel One of the pixels in the lowermost pixel column of the side block is a predicted pixel, and the predicted pixel is determined according to the prediction direction. The prediction directions to be applied are all nine prediction directions 0 to 8 shown in FIG. 8D, and one of them can be arbitrarily selected and used. Therefore, in the intra 4 × 4 prediction mode, there are nine types of intra prediction modes.

  FIG. 8C shows a prediction mode when the encoding target block X is a block composed of 8 × 8 pixels, that is, an intra 8 × 8 prediction mode. Therefore, one macro block is composed of four encoding target blocks. The intra 8 × 8 prediction mode is the same as the intra 4 × 4 prediction mode, and any one of nine prediction directions 0 to 8 shown in FIG. 8D is arbitrarily selected and used. can do. Accordingly, there are nine types of intra prediction modes even in the intra 8 × 8 prediction mode.

  The predicted pixel value may be a value obtained by applying a predetermined weight to the pixel value of the above-mentioned pixel in the adjacent block, or an averaged value (in this case, the pixel value of all the predicted pixels). May be equal).

  The feature value calculation unit 16 in FIG. 7 obtains the variance value Var, activity Act, and the like for the calculation target I picture using the processing result in the intra prediction mode. Hereinafter, the process for this is demonstrated using FIG.

  When the first encoding target block (16 × 16 pixel block, 4 × 4 pixel block, or 8 × 8 pixel block) of the I picture is supplied from the original image memory 1 to the subtracter 2, as described above, A reference pixel having a value “128” is supplied from the prediction unit 12 to the subtractor 2 (step 201 in FIG. 9), and a difference value from the reference pixel having this value “128” is determined for each pixel of the first encoding target block. Is required. The encoding target block composed of pixels of the difference value is supplied to the orthogonal transformation unit 3 and is subjected to intraframe prediction coding. At the same time, the encoding block composed of pixels of the difference value from the subtracter 2 is encoded. The information on the target block is stored as difference information in the difference information storage memory 22 (step 202 in FIG. 9: the block of difference information stored in the difference information storage memory 22 is hereinafter referred to as an operation processing target block. ). Further, as described with reference to FIG. 8, in a coding target block in which a pixel of an adjacent block is used as a prediction pixel and a difference from this prediction pixel can be obtained, the prediction pixel from the intra-frame prediction unit 12 is subtracted. 2 (step 201 in FIG. 2), an encoding target block whose difference value is a pixel value, and therefore an arithmetic processing target block, is obtained. It is stored in the information storage memory 22 (step 202 in FIG. 2).

  In addition, when calculating | requiring the calculation process target block of such difference information, any mode demonstrated in FIG. 8 can be selected arbitrarily and used as an intra (intra-frame) prediction mode.

As described above, the difference information of each calculation processing target block is stored in the difference information storage block 22 (FIG. 7), but every time the difference information of one calculation processing target block is stored in the difference information storage memory 22. The difference information is read by the feature amount calculation unit 16 (FIG. 7), and the following equation 6,

In this calculation process, the variance value Var (n) of this calculation process target block is obtained. Here, P _n ′ (x, y) is a difference pixel value at the position (x, y) of the calculation processing target block of difference information, and Equation 6 shows an 8 × 8 pixel block as an example. Accordingly, the same applies to a 16 × 16 pixel block and a 4 × 4 pixel block.

  In this way, the variance value Var (n) is obtained for the sequential computation processing target blocks, and the quantum value is calculated for each computation processing target macroblock by the above formulas 2 to 4 based on the obtained variance value Var (n). The quantization step size QS is obtained and supplied to the quantization control unit 13 (FIG. 7) (step 203 in FIG. 9).

  The above processing is repeated until all macroblocks in the I picture are completed (step 204 in FIG. 9).

  The calculation of the feature amounts of the P and B pictures is the same as that in the first embodiment.

  As described above, in the second embodiment, since the feature amount is calculated using the difference pixel value between the pixel of the block to be subjected to the arithmetic processing and the reference pixel, the amount of information to be processed can be reduced.

  In addition, in the I picture, the variance value of the macro block to be processed may be calculated every other macro block and stored in the feature amount storage memory 19. Specifically, when the variance value of the surrounding blocks of the calculation target macroblock is obtained, which variance value of the surrounding blocks is determined according to the prediction direction of the intra-frame prediction mode. For example, when the intra-frame prediction mode is the prediction direction 1 (Horizontal) (FIG. 8), as shown in FIG. 8, the prediction direction is a mode predicting from the left to the right. The activity is calculated using the variance value of adjacent blocks. The activity is calculated using, for example, the above formulas 1 and 2. Here, although the variance value of the macroblock for calculating the activity is set to every other macroblock, the present invention is not limited to this and may be performed for each of a plurality of macroblocks. In addition, as a method of calculating a variance value for each of a plurality of macroblocks, any size can be applied as long as it is composed of a plurality of pixels such as an 8 × 8 pixel unit or a 4 × 4 pixel unit. Also, the present invention may be applied to P pictures and B pictures.

  FIG. 10 is a block diagram showing a third embodiment of the image coding apparatus and image coding method according to the present invention, in which 25 is a feature amount prediction processing determination unit, 26 is a user input unit, and 27 is scene change detection. The same reference numerals are given to the portions corresponding to the above-mentioned drawings, and redundant description is omitted.

  In the third embodiment, it is possible to select whether or not to perform the feature amount calculation method in the first and second embodiments for I, P, and B pictures. In general, when the prediction accuracy of feature quantities in I, P, and B pictures is increased, the visual image quality is improved, but the amount of computation associated therewith is enormous. On the other hand, if the prediction accuracy of the feature amount is lowered, the amount of calculation processing associated therewith is reduced, but the deterioration of the image quality is easily noticeable visually.

  In this third embodiment, taking this point into consideration, if an activity is taken as an example of the feature amount, an activity prediction process (as in the previous first and second embodiments, the variance value has already been encoded). It is possible to decide whether or not to perform processing obtained by predicting from the variance value of the block. Therefore, if the activity prediction process of the first embodiment in FIG. 1 is applied, the scene change detection unit 1101, the feature amount prediction process determination unit 1102, and the user input unit 1103 are added to the configuration shown in FIG. Is added.

  In FIG. 10, the scene change detection unit 27 calculates a luminance difference value and a luminance histogram between pictures that are before and after in the display timing order, and whether or not there is a scene change (scene switching) on the displayed screen. Is determined. For example, if a difference value between pictures or a histogram of luminance is larger than a certain threshold value, it is determined that a scene change has occurred. When a scene change occurs, the motion search accuracy decreases, so an instruction signal is sent to the feature amount prediction process determination unit 25 so that the activity prediction process for the P picture and B picture is not performed. Also, based on the input image, frame rate, etc., it is determined whether to perform activity prediction processing of I, P, B pictures adaptively, and the determination signal is sent to the feature amount prediction processing determination unit 25. Also good.

  FIG. 11 is a flowchart showing prediction selection processing for I, P, and B pictures in the encoding processing according to the third embodiment. Hereinafter, this process will be described.

  In the feature amount prediction processing determination unit 25 (FIG. 10), a flag (hereinafter referred to as a feature amount prediction processing flag) for determining whether or not to perform activity prediction processing for I, P, and B pictures is prepared. When the prediction processing flag is “0” (“YES” in step 301 in FIG. 11), the activity prediction processing for I, P, and B pictures is not performed, and itself is performed as in the prior art shown in FIG. The activity is calculated (step 302 in FIG. 11). As an activity calculation method, for example, Equations 1 and 2 are used. On the other hand, when the feature amount prediction processing flag is “1” (“NO” in step 301 of FIG. 11), the activity prediction processing is performed by one of the methods of the first and second embodiments (FIG. 11). Step 303).

  In this way, when the variance value is obtained by the activity prediction process, it is next determined whether or not the arithmetic processing target picture becomes a reference picture of another P or B picture (step 304 in FIG. 11). If this is the case (“YES” in step 304 in FIG. 11), the calculated variance value is stored in the feature amount storage memory 19 (step 305 in FIG. 11). A quantization step size QS is calculated based on Equations 2 to 4, and the corresponding encoding target block is encoded (step 306 in FIG. 11). When this calculation processing target picture is not a reference picture as in the calculation processing target B block, the calculated variance value is not stored in the feature amount storage memory 19 but is quantized based on the above formulas 2 to 4. A step size QS is calculated and an encoding process is performed (step 306 in FIG. 11).

  The above processing is repeated until all the macroblocks in the calculation target picture are finished (step 307 in FIG. 11).

  It should be noted that the relationship between the feature amount prediction processing flags “1” and “0” is not limited to this and may be anything as long as binary discrimination such as ON and OFF can be performed. Further, reverse association may be performed.

  The switching unit of the feature amount prediction processing flag includes, for example, a macro block, a slice, a frame, and a sequence unit. Further, the user may arbitrarily determine whether or not to perform a picture prediction process. In this case, as a configuration of the apparatus, a user input unit 26 is added as shown in FIG. The user input unit receives an instruction on whether or not to perform an activity prediction process on a calculation target picture input by a user, and sets a signal indicating whether or not to set a feature amount prediction process flag (perform a prediction process) or not This is sent to the quantity prediction process determination unit 25.

  According to the configuration of the third embodiment, for example, when encoding a large image such as an HD (high definition) image, the processing time is reduced by performing an activity prediction process for I, P, and B pictures. It can be shortened. On the other hand, when encoding a small image such as a QCIF (Quarter Common Intermediate Format) image with a small amount of processing calculation, all or any of the I, P, and B pictures are encoded. Thus, highly accurate processing can be performed by directly calculating the activity from the original image (its own image). Similarly, even in a high-performance encoding processing apparatus, that is, an encoding apparatus that can process many blocks in a short time, the activity is directly calculated from the original image for all of the I, P, and B pictures, Highly accurate processing can be performed.

  As described above, in the third embodiment, video with a high degree of freedom is encoded by controlling the trade-off between prediction efficiency and calculation load according to the size of the input image and the performance of the encoding device. realizable.

  In the third embodiment shown in FIG. 10, the scene change detection unit 27, the feature amount prediction process determination unit 25, and the user input unit 26 are added to the configuration shown in FIG. Further, the scene change detection unit 27, the feature amount prediction process determination unit 25, and the user input unit 26 may be added, so that the activity prediction process can be performed as in the second embodiment.

  Next, a fourth embodiment of the image encoding device and the image encoding method according to the present invention will be described.

  In the fourth embodiment, in the configuration shown in FIG. 1 or FIG. 7, taking the activity prediction process as an example, information other than the variance value is also stored in the feature amount storage memory 19.

  In FIG. 1, FIG. 7, and FIG. 10, if the variance value obtained by Equation 1 is stored in the feature quantity storage memory 19 as it is, a problem arises that the memory capacity increases. Therefore, for example, the variance value Var obtained in units of blocks of 8 × 8 pixels is calculated by Expressions 2 to 4 and converted into a quantization step size QS, and this quantization step size QS is stored in the feature amount storage memory 19. By doing so, the used capacity of the feature amount storage memory 19 is reduced.

  That is, the quantization step size QS calculated for the I or P picture is stored in the feature amount storage memory 19, and the quantization step size QS of the P or B picture is predicted using this quantization step size QS. . By performing such processing, the information stored in the feature amount storage memory 19 becomes the parameter information of the quantization step size QS for each 16 × 16 pixels from the variance value information for each 8 × 8 pixels. The used capacity of the storage memory 19 is greatly reduced.

  Further, a parameter for changing the pixel value quantization step size QS may be stored in the feature amount storage memory 19. For example, when the image area of the encoding target picture is an edge portion, the quantization parameter is reduced to prevent the deterioration of the image quality so that the deterioration of the image quality is not noticeable. At this time, the quantization step size QS shown in the above Equation 4 is changed to a smaller value, but a parameter for determining how much the change is made is stored in the feature amount storage memory 19. For example, when the quantization step size QS calculated in units of sequences, pictures, and frames is used as the reference quantization step size, and the reference quantization step size is ½ times or ３ times, the denominator coefficient 2 , 3 are parameters stored in the feature quantity storage memory 19. That is, the used capacity can be greatly reduced as compared with the case where the quantization step size QS is stored in the feature amount storage memory 19. The parameter may be a binary determination flag indicating whether or not an edge exists in the edge detection unit.

  In addition, the feature value storage memory 19 may store pixel values after performing preprocessing such as cutting high-frequency components on the calculation target image using a preprocessing filter or the like. For example, when the MPEG-2 TM5 quantization control method is used, the dispersion value becomes large at an edge portion that is visually noticeable for deterioration, so that the quantization parameter QP is heavily weighted as shown in Equation 4 above. It will be roughly quantized. For this reason, the deterioration of the edge portion becomes conspicuous. Also, when encoding at a low bit rate, the quantization step size becomes inevitably rough, and there is a problem that the block distortion increases and visually uncomfortable.

  Therefore, when the image region to be encoded is determined to be an edge portion using an edge detector, a process for sharpening the edge portion of the image is performed by applying a mid-high pass filter. The pixel value after applying the filter is stored in the feature amount storage memory 19. Specifically, when edge detection is performed, processing such as Sobel calculation (edge detection by matrix processing) is applied to determine whether the area is an edge, and the area is determined to be an edge. First, a mid-high pass filter is applied to the region to enhance the edge portion of the image. That is, in the I or P picture, the above-described edge detection or filter processing is applied, the pixel value after the filter application is stored in the feature amount storage memory 19, and based on the stored pixel value information after the filter application, By predicting the pixel value after applying the filter for the P or B picture, it is possible to greatly reduce the enormous amount of processing such as the edge detection and the filter operation in the P and B pictures. Note that various filters such as a low-pass filter can be applied to the above-described filter in addition to the medium-high pass filter.

  Although the embodiments of the present invention have been described above, in the first to fourth embodiments, the unit of image input has always been described as a frame. However, the present embodiment is not limited to this, and the above embodiment can be applied even when an input of an interlaced image is assumed and the image input is a field. Moreover, in the calculation of the feature amount, an example in which processing is performed in units of 8 × 8 pixel blocks has been shown. May be performed. Further, instead of the above-mentioned picture unit encoding process, a slice unit may be used. Further, the present invention can be applied to all coding schemes using I, P, and B pictures.

  Further, the present invention can be applied to transmission of moving image information via the Internet or a mobile phone, and recording of moving image information on a recording medium such as an optical disk, a magneto-optical disk, or a flash memory.

It is a block block diagram which shows 1st Embodiment of the image coding apparatus by this invention. It is a flowchart which shows the flow of the process for the setting of the quantization step size in 1st Embodiment shown in FIG. It is a figure which shows one specific example of the calculation method of the motion vector information used for the feature-value prediction process of the P picture in FIG. It is a figure which shows one specific example of the method of calculating | requiring the dispersion value of the calculation process target block X in S104 in FIG. It is a figure which shows one specific example of the calculation method of the motion vector used for the feature-value prediction process of the B picture in step 104 of FIG. It is a figure which shows the other specific example of the calculation method of the motion vector used for the feature-value prediction process of the B picture in step 104 of FIG. It is a block block diagram which shows 2nd Embodiment of the image coding apparatus and image coding method by this invention. It is a figure which shows the intra prediction method of the I picture used in 2nd Embodiment shown in FIG. It is a flowchart which shows the flow of the process for the setting of the quantization step size in the prediction mode in a frame of 2nd Embodiment shown in FIG. It is a block block diagram which shows 3rd Embodiment of the image coding apparatus and image coding method by this invention. It is a flowchart which shows the flow of the encoding process of 3rd Embodiment shown in FIG. It is a block block diagram which shows an example of the conventional image coding apparatus. It is a figure which shows an example of the encoding object block in a macroblock.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Original image memory 2 Subtractor 3 Orthogonal transformation part 4 Quantization part 5 Encoding part 6 Transmission buffer 7 Inverse quantization part 8 Inverse orthogonal transformation part 9 Adder 10 Frame memory 11 Motion detection and motion compensation part 12 Intra-frame prediction part DESCRIPTION OF SYMBOLS 13 Quantization control part 20 Operation process object macroblock 21 Reference image 22 Difference information storage memory 23a, 23b Adjacent block 24a, 24b Arrow which shows a prediction direction 25 Feature-value prediction process determination part 26 User input part 27 Scene change detection part

Claims (15)

An image encoding device that encodes moving image information including a series of pictures,
A feature amount calculating means for calculating a feature amount of a picture to be encoded;
Feature quantity storage means for storing a variance value at the time of calculating the feature quantity;
An image encoding apparatus comprising: a feature amount prediction unit that calculates a feature amount of a picture to be encoded based on the variance value stored in the feature amount storage unit. In claim 1,
The picture for which the feature amount is calculated by the feature amount calculating means is an I picture to be subjected to intra-screen predictive coding,
The pictures for which the feature amount is calculated by the feature amount prediction means are a P picture and a B picture that are inter-screen predictively encoded,
A variance value of the P picture is calculated by the feature amount prediction unit based on the variance value of the already encoded I picture or the P picture stored in the feature amount storage unit. A variance value is stored in the feature quantity storage means,
A variance value of the B picture is calculated by the feature amount prediction unit based on the already encoded feature values of the I picture and the P picture stored in the feature amount storage unit, and stored in the feature amount storage An image encoding apparatus characterized in that storage in the means is prohibited. In claim 2,
The image coding apparatus according to claim 1, wherein the feature amount calculating unit divides the I picture into blocks and generates the variance value for each block based on a pixel value of each pixel of the block. In claim 2,
Means for dividing the I picture into blocks, and for each block, a difference value from a reference image in the I picture is used as a pixel;
The image coding apparatus according to claim 1, wherein the feature amount calculation means generates the variance value for each block based on the pixel value of the difference value. In claim 2, 3 or 4,
The feature amount predicting means divides the P and B pictures into blocks, and for each block, the feature value predicting means is used as a reference block for the block, to the variance value of the block of the I or P picture of the feature amount storage means or its peripheral blocks. An image coding apparatus that generates a variance value based on the image data. In any one of Claims 2-5,
The feature amount calculated by the feature amount calculation means is a quantization step size when the I picture is encoded,
The image coding apparatus according to claim 1, wherein the feature quantity calculated by the feature quantity prediction unit is a quantization step size when the P and B pictures are coded. In claim 5 or 6,
The feature predicting means, for each block to be encoded of the P picture,
Predict a motion vector based on the motion vector of the already encoded block around the block;
Predicting the reference block of the I or P picture based on the motion vector;
An image coding apparatus characterized in that a predicted variance value of the reference block or its neighboring blocks is obtained from the feature amount storage means to obtain a variance value of a block to be encoded. In claim 5, 6 or 7,
The feature predicting means, for each block to be encoded of the B picture,
Predicting a forward prediction motion vector for a forward I or P picture in display order in the B block and a backward prediction motion vector for a backward P picture in display order in the B block;
Predicting the forward reference block in the forward I or P picture based on the forward predicted motion vector, predicting the backward reference block in the backward P picture based on the backward predicted motion vector;
Obtaining a variance value of the block to be encoded by obtaining a variance value of the front reference block or its peripheral block and a variance value of the reference block or its surrounding block from the feature amount storage means. An image encoding device. In any one of Claims 1-8,
A feature amount prediction process determination unit for determining whether or not the feature amount prediction unit predicts a feature amount is provided,
When it is determined by the feature amount prediction processing determination unit that the feature amount prediction unit does not predict the feature amount, the feature amount calculation unit calculates a feature amount, and the variance value at the time of the calculation is used as the feature amount storage unit. An image encoding apparatus characterized by being stored in In claim 9,
A scene change detection means for detecting a scene change of a moving image is provided,
An image coding apparatus characterized by causing a feature amount prediction processing determination unit to determine that a feature amount is not predicted by a feature amount prediction unit for a scene-changed picture. In any one of Claims 1-10,
An image encoding apparatus, wherein information of a quantization step size obtained from the variance value is stored in the feature amount storage means instead of the variance value. In any one of Claims 1-10,
An image encoding apparatus, wherein the feature value storage means stores a pixel value after pre-processing a picture to be encoded. In claim 12,
The image coding apparatus according to claim 1, wherein the preprocessing is processing for enhancing an edge portion of an image. An image encoding method for encoding moving image information composed of a series of I, P, and B pictures,
When encoding the I picture, the feature amount of the I picture is calculated and stored,
When encoding the P and B pictures, the feature amount is calculated based on the stored feature amount, and the calculated feature amount of the P picture is stored and held. Image coding method. In claim 14,
The stored feature amount is a variance value calculated for each divided block of the I, P, and B pictures,
An image encoding method, wherein information for determining a quantization step size at the time of encoding the I, P, and B pictures is calculated from the variance value.

Images