JP4462914B2 - Video decoding device - Google Patents

Description

  The present invention decodes block-divided orthogonally transformed / quantized / encoded encoded data, and reduces block distortion and mosquito noise generated in a decoded image obtained by inverse quantization / inverse orthogonal transformation. The present invention relates to a moving picture decoding apparatus.

  In recent years, video communication services such as videophones and video conferences are promising as services that effectively utilize ISDN (Integrated Services Digital Network) and GSTN (General Switched Telephony Network), and efficient transmission of such moving images is expected. Research on high-efficiency coding aimed at has been actively conducted. These studies use the statistical properties of images to reduce the amount of information by removing the redundancy contained in the images. As such an encoding method, a hybrid encoding method combining motion compensation prediction and discrete cosine transform is well known.

  On the decoding side, post-filters that reduce block distortion and mosquito noise that are characteristic of hybrid coding schemes that combine motion-compensated prediction and discrete cosine transform are implemented on decoded images to improve image quality. .

  Hereinafter, a conventional example of a moving picture decoding apparatus using motion compensation prediction and two-dimensional orthogonal transformation will be described with reference to FIG. In FIG. 7, 51 is a variable length decoding unit, 52 is an inverse quantization unit, 53 is an inverse orthogonal transform unit, 54 is an adder, 55 is a frame memory unit, 56 is a motion compensation prediction unit, 57 is a switch, and 58 is a post. The filter part is shown.

  An encoded bit stream that is encoded information is input to the variable length decoding unit 51. The encoded bit stream is an encoding mode that represents an orthogonal transform coefficient of an image generated by a moving image encoding device or an encoding state, that is, whether or not encoding is performed, motion vectors, interframe / intraframe prediction information, etc. Are variable-length encoded and multiplexed.

  The variable length decoding unit 51 decodes the encoded bitstream to obtain quantized orthogonal transform coefficients, motion vectors, and encoding mode information such as interframe / intraframe prediction. The inverse quantization unit 52 performs inverse quantization on the quantized orthogonal transform coefficients to obtain orthogonal transform coefficients, and the inverse orthogonal transform unit 53 performs two-dimensional inverse orthogonal transform to obtain a difference image.

  When the encoding mode is inter-frame prediction, the output of the switch 57 becomes the output of the motion compensation prediction unit 56, and the difference image is decoded by the adder 54 in the frame memory unit 55. Based on the image, the decoded value of the current frame is accumulated in the frame memory unit 55 by adding the motion compensated prediction value from the motion compensation prediction unit 56.

  When the block is intra-frame prediction, the switch 57 is disconnected from the motion compensation prediction unit 56 and nothing is added to the output from the inverse orthogonal transform unit 53, so that the output of the inverse orthogonal transform unit 53 results. Is stored as a decoded image in the frame memory unit 55 as it is.

  The post filter unit 58 applies a post filter for reducing block distortion and mosquito noise to the decoded image of the current frame, which is an output from the adder 54, and outputs a decoded image with reduced noise.

  In the conventional method described above, post-filtering is always performed on the decoded image, and filtering is performed even on portions that do not require post-filter processing, such as still image portions that do not change from the previous frame. There is a disadvantage of wasted processing time. This leads to an increase in power consumption such as when hardware is used.

  In addition, since the coding attribute of the block is not referred to, filtering suitable for the generated noise cannot be performed, so that there is a disadvantage that not only the processing amount is increased but also the image quality is deteriorated due to filtering.

  In view of such a problem, the present invention provides a block distortion reduction filter and a mosquito noise reduction filter based on whether or not the current block is encoded, a motion vector, interframe / intraframe prediction information, and the state of a transform coefficient. ON / OFF control and block replacement with a decoded image that has been postfiltered in the previous frame are adaptively performed, and a decoded image with reduced noise due to the postfilter can be obtained while reducing the processing amount. The present invention provides a moving picture decoding apparatus capable of improving image quality.

The basic technical means of the present invention is a video decoding apparatus that decodes an image encoded for each block, and performs motion compensation prediction when the prediction encoding mode of the block is inter-frame prediction. Means and filter means for filtering block boundaries of the decoded image;
Filter control means for controlling the filter means; first image storage means for storing an image before being filtered by the filter means; and second image storage means for storing an image filtered by the filter means. The motion compensation prediction unit generates an image in which motion compensation is performed according to a motion vector acquired at the time of image decoding with respect to an image of a previous frame stored in the first image storage unit, and the filter unit The filter control means includes a case where a filter is not implemented, a case where a first filter is implemented, and a case where a second filter stronger than the first filter is implemented, wherein the filter control means When the mode is intra-frame prediction, select the case where the second filter is performed, and the predictive coding mode of the block is inter-frame prediction. When the first filter is performed when orthogonal transform coefficients exist in the block, and when the prediction coding mode of the block is inter-frame prediction, and orthogonal in the block The case where no filter is performed when no transform coefficient is present is selected, and the filter unit filters the block boundary of the decoded image according to the control of the filter control unit. An image in which motion compensation is performed in accordance with a motion vector acquired at the time of image decoding with respect to the image of the previous frame stored in the image storage means is output.

  According to the present invention, since the on / off control of the block distortion reduction filter and the mosquito noise reduction filter is adaptively performed based on the state of the transform coefficient of the current block, the noise can be reduced by an appropriate filter while reducing the processing amount. Can be obtained. Further, when the embodiment according to the present invention is implemented as hardware, the reduction of the processing amount has the effect of reducing the power consumption.

  An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an embodiment of the present invention. A variable length decoding unit 1 that performs variable length decoding of an encoded bit stream, and an orthogonal transform coefficient that is quantized and connected to the variable length decoding unit 1 is inversely quantized. An inverse quantization unit 2 to be converted, an inverse orthogonal transformation unit 3 connected to the inverse quantization unit 2 to obtain a difference image by two-dimensional inverse orthogonal transformation of orthogonal transformation coefficients, the inverse orthogonal transformation unit 3 and the switch 7 An adder 4 connected to obtain the decoded image by adding the difference image obtained by the inverse orthogonal transform unit 3 and the motion compensated prediction value, and the decoded image obtained by the adder 4 connected to the adder 4 A first frame memory unit 5 for storing the motion, a variable length decoding unit 1, a motion compensation prediction unit 6 connected to the first frame memory unit 5 to output a motion compensation prediction value, and the variable length decoding unit 1 And a switch 7 connected to the motion compensation prediction unit 6 for switching the output of the motion compensation prediction value, and the variable A post-filter processing content control unit 8 connected to the decoding unit 1 to control the operation of the post-filter, and a post-filter unit 9 connected to the adder 4 and the post-filter processing content control unit 8 to perform post-filtering on the decoded image A second frame memory unit 10 connected to the post filter processing content control unit 8 and the output signal selection unit 11 to store the post-filtered image, the post filter unit 9 and the second frame memory unit 10 and an output signal selection unit 11 that is connected to the post-filter processing content control unit 8 and controls switching of the output of the post-filtered image.

  The operation of the moving picture decoding apparatus having the above configuration is as follows.

  The variable length decoding unit 1 performs DCT on an image by a moving image encoding apparatus and quantizes the DCT coefficient, information on whether encoding is performed, intra-frame / inter-frame prediction A coded bitstream generated by variable-length coding of a prediction coding mode, motion vectors, and the like, which are information of the information, for example, ITU-T Recommendation H.264 An encoded bitstream according to H.263 is input.

  Then, by performing variable length decoding here, discrete cosine transform (DCT) coefficients quantized for each block, motion vectors, interframe / intraframe prediction information (predictive coding mode), blocks that are not coded ( Encoding information such as whether or not encoding has been performed) and quantization step size.

  In the inverse quantization unit 2, the quantized DCT coefficient obtained in the variable length decoding unit 1 is inversely quantized according to the quantization step size to obtain an inversely quantized DCT coefficient.

  The inverse orthogonal transform unit 3 performs a two-dimensional inverse DCT on the inversely quantized DCT coefficient to obtain a difference image.

  In the adder 4, when the block is inter-frame prediction, the motion compensated prediction value calculated based on the motion vector of the block from the decoded image of the previous frame stored in the first frame memory unit 5, or the block In the case of intra-frame prediction, the switch 7 switches the output so that the motion compensation prediction value is zero, and adds the difference image obtained by the inverse orthogonal transform unit 3 to obtain a decoded image.

  The first frame memory unit 5 stores the decoded image obtained by the adder 4. The motion compensation prediction unit 6 obtains a motion compensation prediction value from the decoded image of the previous frame in the first frame memory unit 5 and the motion vector that is information from the variable length decoding unit 1.

  The switch 7 switches the input to the adder 4 according to the information of the predictive coding mode decoded by the variable length decoding unit 1.

  The output from the adder 4 is output to the first frame memory unit 5 and also to the post filter unit 9.

  The post filter processing content control unit 8 controls the operations of the post filter unit 9, the second frame memory unit 10, and the output signal selection unit 11 so as to reduce the processing amount and perform the optimum noise reduction processing. Do. As shown in FIG. 2, the following processing is determined according to the coding mode of the current block, which is the coding information output from the variable length decoding unit 1, and the state of the DCT coefficient.

  The encoding mode can be easily determined by bits representing the encoding mode present in the bitstream. Regarding the means for determining the state (presence / absence / content) of the DCT coefficient, if there is a bit flag indicating the state of the DCT coefficient in the bit stream, the determination may be made based on the bit flag. If such a bit flag does not exist in the bit stream, the state of the DCT coefficient of each block is determined based on the variable length decoded data itself.

(1) When the block is not coded Neither the block distortion reduction filter nor the mosquito noise reduction filter is implemented, and the position of the current block of the decoded image in which the post filter of the previous frame of the second frame memory unit 10 is implemented Is output.

(2) When no DCT coefficient is present in a block in inter-frame prediction Neither the block distortion reduction filter nor the mosquito noise reduction filter is implemented, and decoding in which post-filtering of the previous frame in the second frame memory unit 10 is implemented An image at a position where motion compensation has been performed on the image according to the motion vector is output.

(3) When only DC component of DCT coefficient exists in the block in inter-frame prediction, only the block distortion reduction filter is performed on the output from the adder 4, and this image is output.

(4) When DC and AC components of DCT coefficients exist in a block in inter-frame prediction, both a block distortion reduction filter and a mosquito noise reduction filter are implemented on the output from the adder 4.

(5) When only the DC component of the DCT coefficient exists in the block in intra-frame prediction For the output from the adder 4, only the block distortion reduction filter is executed with a stronger filter strength than in the case of inter-frame prediction. This image is output.

(6) When DC and DC components of DCT coefficients exist in the block in intra-frame prediction The block distortion reduction filter and mosquito with a stronger filter strength than the inter-frame prediction for the output from the adder 4 Both noise reduction filters are implemented and this image is output.

  The reason why the six cases are divided as described above is as follows. First, when the current block is an unencoded block, it is the same as a post-filtered decoded image at the same position as the current block in the previous frame in the second frame memory unit 10. Therefore, if the current block is replaced with this, it is not necessary to perform a new post filter, and the amount of processing is reduced.

  Further, when the current block is inter-frame prediction and no DCT coefficient exists in the block, the current block in the previous frame in the second frame memory unit 10 is post-filtered at a position that has been motion compensated by the motion vector. This is the same as the decoded image. Therefore, if the current block is replaced with this, it is not necessary to perform a new post filter, and the amount of processing is reduced.

  Further, when the current block is inter-frame prediction and only the DC component of the DCT coefficient exists in the block, only block distortion is generated, and therefore only the block distortion reduction filter may be implemented. Therefore, the amount of processing is reduced by not performing the mosquito noise reduction filter.

  If the current block is inter-frame prediction and both DC and AC components of the DCT coefficient exist in the block, both block distortion and mosquito noise may be generated. Both noise reduction filters need to be implemented.

  Further, when the current block is intra prediction and only the DC component of the DCT coefficient exists in the block, only the block distortion is generated. Therefore, only the block distortion reduction filter may be implemented. Therefore, the amount of processing is reduced by not performing the mosquito noise reduction filter. However, since the intra-frame prediction is compressed higher than the inter-frame prediction, the image quality is deteriorated. Therefore, the filtering process is performed with a stronger filter strength than that in the inter-frame prediction.

  Also, if the current block has intra-frame prediction and DC and AC components of DCT coefficients are present in the block, both block distortion and mosquito noise may occur, so the block distortion reduction filter and mosquito noise Both reduction filters need to be implemented. However, since the intra-frame prediction is compressed higher than the inter-frame prediction, the image quality is deteriorated. Therefore, the filtering process is performed with a stronger filter strength than that in the inter-frame prediction.

  An example of the above processing will be described with reference to FIG. Originally, the decoded image is output in units of blocks from the adder 4, but in FIG. 3A, the output content of the adder 4 is shown in a form in which blocks are arranged in the same manner as the image to be restored for explanation. . Also, the contents of the second frame memory in FIG. 3B are images on which the post-filter for the previous frame has been performed.

  In FIG. 3, since block 1 is exactly the same as the previous frame, information indicating whether or not encoding is performed is a block that is not encoded. Therefore, this block may use the block indicated by block 2 of the image on which the post-filter of the previous frame stored in the second frame memory is used, and it is necessary to perform block distortion and mosquito noise reduction processing. There is no.

  Since block 3 is the same as block 4 indicated by the motion vector, it has a motion vector in the inter-frame prediction and no DCT coefficient. Therefore, the block indicated by block 4 may be used, and it is not necessary to perform block distortion and mosquito noise reduction processing.

  Further, since the block 5 is an image that does not exist at all in the previous frame, DC and AC components of the DCT coefficient exist in the block in intra-frame prediction. Therefore, block distortion and mosquito noise reduction processing is performed on the output from the adder 4. The post filter unit 9 reduces noise generated in the decoded image by the adder 4.

  FIG. 4 is a diagram for explaining the configuration of the post filter unit 9. The post filter processing content control unit 8 is connected to the post filter processing content control unit 8, and the block distortion reduction filter unit 22, the switch 23, the mosquito noise reduction filter unit 24, and the switch 25. A filter switching control unit 21 that controls the operation of the adder 4, a block distortion reduction filter unit 22 that is connected to the adder 4 and the filter switching control unit 21 and performs a filter that reduces block distortion, and the adder 4 and the filter switching A switch 23 connected to the control unit 21 and the block distortion reduction filter unit 22 to switch between the output of the block distortion reduction filter and the output of the adder 4, and connected to the filter switching control unit 21 and the switch 23 A mosquito reduction filter section 24 that performs a filter for reducing the mosquito noise, the filter switching control section 21, the switch 23, and the module Quito noise reduction or connect outputs of mosquito noise reduction filter in the filter unit 24, constituted by switcher 25 to output switching whether the output of the switching unit 23.

  The filter switching control unit 21 controls the post filtering operation as follows in accordance with the processing determined by the post filter processing content control unit 8.

(1) When neither a block distortion reduction filter nor a mosquito noise reduction filter is implemented The block distortion reduction filter unit 22 is turned off so that the output of the adder 4 becomes the output of the switch 23 as it is. Further, the mosquito noise reduction filter unit 24 is turned off so that the output of the switch 23 becomes the output of the switch 25 as it is.

(2) When implementing only the block distortion reduction filter: The block distortion reduction filter section 22 is turned on so that the output of the block distortion reduction filter section 22 becomes the output of the switch 23. Further, the mosquito noise reduction filter unit 24 is turned off so that the output of the switch 23 becomes the output of the switch 25 as it is.

(3) When implementing both the block distortion reduction filter and the mosquito noise reduction filter: The block distortion reduction filter section 22 is turned on so that the output of the block distortion reduction filter section 22 becomes the output of the switch 23. Further, the mosquito noise reduction filter unit 24 is turned on so that the output of the mosquito noise reduction filter unit 24 becomes the output of the switch 25.

  The block distortion reduction filter unit 22 performs the following processing when implementing the block distortion reduction filter according to the control of the filter switching control unit 21. The purpose of this is to remove the discontinuity of the block boundary. Therefore, as shown in FIG. 5, two pixels x0 and x1 sandwiching the block boundary of the decoded image that is the output of the adder 4 are smoothly connected. Like that. At this time, in order to prevent over-compensation of a portion with a large level difference, nonlinear processing as shown in Expression (1) is performed, and x0 is replaced with the output pixel value y0. Here, q is a parameter for controlling the filter strength.

  Further, when the control of the filter switching control unit 21 turns off the block distortion reduction filter unit 22, the above processing is not performed.

  When the control of the filter switching control unit 21 turns on the block distortion reduction filter unit 22, the switching unit 23 switches so that the output of the block distortion reduction filter unit 22, that is, a decoded image with reduced block distortion is output. Is done. Further, when the control of the filter switching control unit 21 turns off the block distortion reduction filter unit 22, switching is performed so that the output of the adder 4, that is, the decoded image that has not been filtered is output as it is.

  The mosquito noise reduction filter unit 24 performs the following processing when implementing the mosquito noise reduction filter according to the control of the filter switching control unit 21. Here, as shown in FIG. 6, an output pixel yi, j is obtained using a filter shown in Expression (2) by a 3 × 3 pixel window centered on the pixel of interest xi, j and is replaced with xi, j. Here, ε is a parameter for controlling the filter strength.

  Further, when the control of the filter switching control unit 21 turns off the mosquito noise reduction filter unit 24, the above processing is not performed.

  In the switch 25, when the control of the filter switching control unit 21 turns on the mosquito noise reduction filter unit 24, switching is performed so that the output of the mosquito noise reduction filter unit 24, that is, the decoded image with reduced mosquito noise is output. Is done. When the control of the filter switching control unit 21 turns off the mosquito noise reduction filter unit 24, the output of the switch 23, that is, the decoded image that has not been subjected to the filtering process, or the decoded image that has undergone only the block distortion reduction process is output. Switching is performed as follows.

  The second frame memory unit 10 receives the output of the output signal selection unit 11 and stores it as a decoded image of the previous frame in which noise is reduced. When the current block is a block that is not encoded by the control from the post filter processing content control unit 8, the image of the previous frame corresponding to the position is input to the output signal selection unit 11 in units of blocks. In addition, when there is no DCT coefficient in a block in inter-frame prediction, an image at a position where motion compensation has been performed according to a motion vector is input to the output signal selection unit 11.

  In the output signal selection unit 11, switching is performed so that either the output from the post filter unit 9 or the output from the second frame memory unit 10 is output according to the control of the post filter processing content control unit 8. . When the current block is an unencoded block, or when there is no DCT coefficient in the block in inter-frame prediction, switching is performed so that the input from the second frame memory unit 10 is output. In addition, when only the DC component of the DCT coefficient exists in the block in the inter-frame prediction, or when the DC component and the AC component of the DCT coefficient exist in the block in the inter-frame prediction, or the DCT coefficient in the block in the intra-frame prediction. When there is only a direct current component, or when there are direct current and alternating current components of DCT coefficients in the block in intra-frame prediction, switching is performed so that the input from the post filter unit 9 is output.

That is, on / off control of the block distortion reduction filter and the mosquito noise reduction filter based on the presence / absence of encoding of the current block , motion vector , interframe / intraframe prediction information , and the state of the transform coefficient, of performing a replacement block by decoding image post filter is applied adaptively, it is possible to obtain a reduced decoded image noise due to post filter also while reducing the processing amount. Further, when the present embodiment is implemented as hardware, the reduction of the processing amount also has an effect of reducing power consumption.

As is apparent from the above description, according to the present invention, since the on / off control of the block distortion reduction filter and the mosquito noise reduction filter is adaptively performed based on the presence / absence of the transform coefficient of the current block, the processing amount is reduced. While being reduced, a decoded image with reduced noise can be obtained by an appropriate filter. Further, when the present embodiment is implemented as hardware, the reduction of the processing amount also has an effect of reducing power consumption.

It is a block diagram which shows the structure of one Embodiment in this invention. It is explanatory drawing which shows the process of the post filter process content control part in this invention. It is a figure explaining an example of the post filter process in the present invention. It is a figure which shows the detail of a structure of the post filter part in this invention. It is a figure explaining the structure of the block distortion reduction filter in this invention. It is a figure explaining the structure of the mosquito noise reduction filter in this invention. It is a block diagram which shows the structure of the conventional moving image decoding apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Variable length decoding part, 2 ... Inverse quantization part, 3 ... Inverse orthogonal transformation part, 4 ... Adder, 5 ... 1st frame memory part, 6 ... Motion compensation prediction part, 7 ... Switch, 8 ... Post filter Processing content control unit, 9 ... post filter unit, 10 ... second frame memory unit, 11 ... output signal selection unit.

Claims (6)

A video decoding device for decoding an image encoded for each block,
Motion-compensated prediction means for performing motion-compensated prediction when the prediction encoding mode of the block is inter-frame prediction;
Filter means for filtering the block boundaries of the decoded image;
Filter control means for controlling the filter means ;
First image storage means for storing an image before being filtered by the filter means;
Second image storage means for storing the image filtered by the filter means
The motion compensation prediction unit generates an image in which motion compensation is performed according to a motion vector acquired at the time of image decoding with respect to the image of the previous frame stored in the first image storage unit,
The filter means includes a case where no filter is implemented, a case where a first filter is implemented, and a case where a second filter stronger than the first filter is implemented,
The filter control means selects a case where the second filter is performed when the prediction encoding mode of the block is intra-frame prediction,
When the prediction coding mode of the block is inter-frame prediction, and when an orthogonal transform coefficient is present in the block, the case of performing the first filter is selected,
When the prediction encoding mode of the block is inter-frame prediction, and when there is no orthogonal transform coefficient in the block, select a case where no filter is performed,
The filter means filters a block boundary of the decoded image according to the control of the filter control means ;
If it does not implement the filter, and wherein Rukoto output an image motion compensation is performed according to the motion vector obtained at the time of image decoding for the image of the second image storing means before stored in the frame A moving picture decoding apparatus.   2. The moving picture decoding apparatus according to claim 1, wherein when the orthogonal transform coefficient is only a direct current component, the filter means controls to perform only filtering for reducing block distortion.   2. The moving picture decoding according to claim 1, wherein when the orthogonal transform coefficient has an AC component, the filtering means controls to perform filtering for reducing block distortion and filtering for reducing mosquito noise. apparatus.   The moving picture decoding apparatus according to claim 1, wherein the filter means changes a pixel value sandwiching the block boundary to obtain a filtered pixel value in order to smoothly connect the block boundaries. 2. The moving picture decoding apparatus according to claim 1, wherein the filtering unit obtains an output pixel value of the filtering based on a pixel value sandwiching a block boundary.   The filter means uses a pixel value in a 3 × 3 pixel window centered on a target pixel to be filtered, and replaces a value calculated based on a difference between the target pixel and its surrounding pixels with the target pixel. The moving picture decoding apparatus according to claim 1.

Images