JP2006211152A - Device and method for coding image and decoding image, and programs for coding and decoding image - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To suppress quality deterioration of an image obtained by decoding without losing a high-frequency component that an original image, such as a moving picture and a still image, has originally. <P>SOLUTION: A filter coefficient calculator 150 performs the filter coefficient arithmetic processing of a two-dimensional FIR filter 250 on the side of a moving picture decoding device on the basis of a decoded image inputted via an adder 140 and an input moving picture that is the original image, and obtains a filter coefficient for minimizing the coding error between the input moving picture and the decoded image for outputting to a variable-length coder 180. The variable-length coder 180 performs variable-length coding to the coded image from a coder 120, the filter coefficient from the filter coefficient calculator 150, a motion vector from a motion detector 170, or the like for outputting to a transmitter 190. The transmitter 190 transmits a moving picture bit stream to which the variable-length coder 180 performs variable-length coding to the moving picture decoding device. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to video encoding methods such as terrestrial digital broadcasting, DVD, and moving image distribution on the Internet, such as MPEG-1 encoding, MPEG-2 encoding, motion JPEG encoding methods, Coding apparatus, image decoding apparatus, image coding method, image decoding method, image coding employing a decoding system for decoding moving images and still images encoded by still image coding schemes such as JPEG2000 The present invention relates to a computer program and an image decoding program.

  In an encoding scheme such as MPEG-2, efficient compression is realized by performing motion compensation and discrete cosine transform (DCT) in units of blocks and quantizing the DCT coefficients. However, when encoding is performed at a low bit rate, each block is DCTed independently, so that a gradation discontinuity called block distortion occurs in the block boundary portion of the decoded image. Further, in the vicinity of the edge, there is a problem that high-frequency noise called ringing is generated due to the lack of high-frequency components accompanying quantization of DCT coefficients, and the quality of the decoded image is deteriorated.

In order to solve such a problem, various methods for reducing deterioration by applying a low-pass filter to a decoded image have been proposed (see Non-Patent Documents 1 to 4).
R. Rosenholtz and A. Zakhor “Iterative procedures for reduction of blocking effects in transform image coding,” IEEE Trans. Circuits Syst. Video Technol., Vol. 2, no. 1, pp. 91-95, Mar. 1992. Y. Yang, NP Galatsanos and AK Katsaggelos, “Projection-based spatially adaptive reconstruction of block-transform cmpressed images,” IEEE Trans. Image Processing, vol. 4, no. 7, pp. 896-908, Dec. 1993. H. Peak, R. Kim and S. Lee “On the POCS-based postprocessing technique to reduce the blocking artifacts in transform coded images,” IEEE Trans. Circuits Syst. Video Technol., Vol. 8, no. 3, pp 358-367, Jun. 1998. Tadashi Kasezawa, “Reduction of quantization noise in transform-coded images using a directional adaptive filter,” IEICE Transactions (D-II), vol.J87-D-II, no. 2, pp. 595- 607, Feb. 2004.

  However, since these methods perform filtering only on the decoding side, the high frequency components inherent in the original image may be lost, and the quality of the image obtained by decoding may deteriorate.

  The present invention has been made in view of the above points, and an image encoding device capable of preventing deterioration in quality of a decoded image as much as possible without losing high-frequency components inherent in an original image such as a still image or a moving image. An object is to provide an image decoding apparatus, an image encoding method, an image decoding method, an image encoding program, and an image decoding program.

  In order to solve the above problems, in the image encoding device of the present invention, an input image is encoded to generate an encoded image, and the encoded image is locally decoded to generate a decoded image. An image encoding device that transmits to an image decoding device, a filter coefficient calculation unit that calculates a filter coefficient of a filter of the image decoding device based on the input image and the decoded image, and the filter coefficient as the code A transmission unit that transmits the converted image to the image decoding apparatus side.

  In particular, in the image encoding device, the filter is a two-dimensional FIR filter, and the filter coefficient of the filter is a mean square between the output image after the filtering process by the two-dimensional FIR filter and the input image. This is the value that minimizes the error.

  In the image encoding device, the input image is a moving image.

  An image decoding apparatus according to the present invention is an image decoding apparatus that decodes an encoded image encoded by an image encoding apparatus to generate an output image, the input image input to the image encoding apparatus, and the A reception unit that receives the filter coefficient calculated based on the decoded image locally decoded by the image encoding device, the encoded image, and a decoding unit that decodes the encoded image and generates a decoded image; And a filter that filters the decoded image generated by the decoding unit using the filter coefficient to produce an output image.

  The image encoding method of the present invention generates an encoded image by encoding an input image, generates a decoded image by locally decoding the encoded image, and transmits the encoded image to an image decoding apparatus. An image encoding method for calculating a filter coefficient of a filter of the image decoding device based on the input image and the decoded image, and the filter coefficient together with the encoded image in the image decoding device To the side.

  In particular, in the image encoding method, the filter coefficient of the filter is a value that minimizes the mean square error between the output image after the filter processing by the filter and the input image.

  The image decoding method of the present invention is an image decoding method for decoding an encoded image encoded by an image encoding device into an output image, the input image being input to the image encoding device And a filter coefficient calculated based on the decoded image locally decoded by the image encoding device and the encoded image, and the decoded image is generated by decoding the encoded image, and the filter coefficient The filtered image is filtered to obtain an output image.

  The image encoding program of the present invention generates an encoded image by encoding an input image, generates a decoded image by locally decoding the encoded image, and transmits the encoded image to an image decoding device. An image encoding program for causing a computer to execute a process of transmitting, calculating a filter coefficient of a filter on the image decoding device side based on the input image and the decoded image, and calculating the filter coefficient as the filter coefficient Transmitting the encoded image together with the encoded image to the image decoding apparatus side.

  Also, the image decoding program of the present invention is an image decoding for causing a computer to execute a process for causing a computer to execute a process for decoding an encoded image encoded by an image encoding device into an output image. A filter coefficient calculated based on an input image input to the image encoding device and a decoded image locally decoded by the image encoding device, and the encoded image; The computer is caused to decode the encoded image to generate a decoded image, and filter the decoded image with the filter coefficient to obtain an output image.

  As described above, according to the image encoding device, the image encoding method, and the image encoding program of the present invention, the filter coefficients of the two-dimensional FIR filter of the image decoding device are calculated based on the input image and the decoded image. While calculating and transmitting the filter coefficient together with the encoded image to the image decoding apparatus side, the image decoding apparatus, the image decoding method, and the image decoding program of the present invention include the input image and the image code input to the image encoding apparatus. The filter coefficient calculated based on the decoded image locally decoded by the encoding device and the encoded image are input, the encoded image is decoded to generate a decoded image, and the decoded image is converted into a two-dimensional FIR by the filter coefficient. Since the output image is processed by filtering, the original image such as a still image or moving image that is the input image is originally compared to the conventional method that performs filtering on the decoding side alone. One high-frequency component is not lost, as much as possible it is possible to prevent deterioration of the quality of the output image.

  Since MPEG-1 and MPEG-2 use motion compensation and DCT in units of blocks, degradation such as block distortion and ringing occurs in the decoded image when encoding is performed at a low bit rate. As a typical technique for reducing such deterioration, the techniques of Non-Patent Documents 1 to 4 can be cited. However, since these methods perform low-pass processing on the decoding side, high frequency components inherent in the original image may be lost.

  Therefore, in the embodiment of the present invention, a filter coefficient of a two-dimensional FIR filter (Finite Impulse Response Filter) that minimizes an encoding error between an original image and a decoded image on the decoding side is designed on the encoding side. A method is proposed in which filter coefficients are added to an encoded image together with encoding parameters such as motion vectors and transmitted to the decoding side. By using the proposed method, it is possible to obtain a high-quality image as compared with the conventional method in which filtering is performed on the decoding side alone. Embodiments of an image encoding device, an image decoding device, an image encoding method, an image decoding method, an image encoding program, and an image decoding program according to the present invention will be described below in detail with reference to the drawings. . In the present embodiment, the MPEG-2 moving image encoding method will be described as an example. However, the present invention is not limited to MPEG-1, MPEG-4, H.264, or the like. Of course, other moving image coding systems such as H.264 / AVC may be used, and of course, still picture coding systems such as JPEG and JPEG2000 are also applicable.

  FIG. 1 is a diagram showing a configuration of an embodiment of a moving image encoding apparatus according to the present invention.

  In the figure, a moving image encoding apparatus 100 includes a subtractor 110, an encoding unit 120, a decoding unit 130, an adder 140, a filter coefficient calculation unit 150, a frame memory 160, a motion detection unit 170, A variable length encoding unit 180 is included.

  FIG. 2 is a diagram showing the configuration of the embodiment of the moving picture decoding apparatus according to the present invention.

  In addition, the moving picture decoding apparatus 200 includes a receiving unit 210, a variable length decoding unit 220, a decoding unit 230, an adder 240, a two-dimensional FIR filter 250, and a frame memory 260.

  Next, the operation will be described.

  First, in the moving image coding apparatus 100, when the moving image input to the apparatus 100 is subjected to intra-frame (intra) coding, the inter-frame (inter) coding is performed without using the subtracter 110. Is output to the encoding unit 120 via the subtractor 110.

  The encoding unit 120 performs an orthogonal transform such as DCT and wavelet transform and an encoding process such as a quantization process, generates an encoded image, and outputs the encoded image to the variable length encoding unit 180 and the decoding unit 130.

  In decoding section 130, the encoded image is subjected to processing reverse to that of encoding section 120, that is, local decoding processing such as inverse orthogonal transformation and inverse quantization processing, and output to adder 140.

  In the adder 140, when the subtracter 110 does not operate, that is, when the encoding unit 120 performs intra-frame encoding of the input moving image as it is, the decoded image from the decoding unit 130 is directly used as the filter coefficient calculation unit 150 and the frame memory. To 160. On the other hand, when the subtracter 110 is operated, that is, the encoding unit 120 calculates an error image between the input moving image and the predicted image from the frame memory 160 based on the motion vector detected by the motion detection unit 170 between frames (interlaces). In the case of encoding, the adder 140 adds the decoded image from the decoding unit 130 and the predicted image, and outputs the result to the filter coefficient calculation unit 150 and the frame memory 160.

  The filter coefficient calculation unit 150 calculates the filter coefficient of the two-dimensional FIR filter 250 on the moving image decoding apparatus 200 side, as will be described later, based on the decoded image input via the adder 140 and the input moving image that is the original image. Processing is performed to obtain a filter coefficient that minimizes the coding error in the input moving image that is the original image and the decoded image on the decoding device side, and the filter coefficient is output to the variable length coding unit 180.

  Here, the filter coefficient calculation process in the filter coefficient calculation unit 150 will be described.

  First, the filter coefficient calculation unit 150 performs filter coefficient calculation processing based on the decoded image input from the adder 140 and the input moving image that is the original image, and calculates an encoding error between the original image and the decoded image. The filter coefficient to be minimized is calculated for each frame of the input moving image.

That is, the filter coefficient calculation unit 150 quantizes the filter coefficient that minimizes the encoding error obtained as described above into the floating-point number r expressed by the following equation (1) in order to suppress the code amount.

However, in the above formula (1). n _f represents the number of bits in the mantissa, n _e represents the number of bits in the exponent, and e _o represents the number of offsets in the mantissa.

  The filter coefficient calculation unit 150 outputs the filter coefficients quantized in this way to the variable length coding unit 180, and transmits them to the decoding device 200 together with the coded parameters such as the coded image and the motion vector.

  The frame memory 160 stores the decoded image from the adder 140 as a reference image.

  The motion detection unit 170 performs motion compensation prediction based on the input moving image and the reference image stored in the frame memory 160 every time the input moving image is input and interframe coding is performed, and the input A motion picture and a predicted picture with little error are obtained and output from the frame memory 160 to the subtractor 110, and a motion vector at that time is detected and output to the variable length coding unit 180.

  In the variable length coding unit 180, coding parameters such as a coded image from the coding unit 120, a filter coefficient from the filter coefficient calculation unit 150, and a motion vector from the motion detection unit 170 are converted into Huffman coding and arithmetic code. The data is variable length encoded by the conversion or the like and output to the transmitter 190.

  The transmission unit 190 transmits the moving image bitstream variable-length encoded by the variable-length encoding unit 180 to the moving image decoding apparatus 200 shown in FIG.

  Next, the operation in the video decoding device 200 shown in FIG. 2 will be described.

  In the moving picture decoding apparatus 200, the receiving unit 210 receives the moving picture bit stream from the moving picture encoding apparatus 100 shown in FIG. 1 and outputs it to the variable length decoding unit 220.

  The variable length decoding unit 220 performs variable length decoding processing opposite to that of the variable length encoding unit 180, and the encoded image, the motion vector, and the filter coefficient are respectively obtained by the decoding unit 230, the frame memory 260, and 2 Output to the dimension FIR filter 250.

  The decoding unit 230 generates a decoded image by performing inverse quantization and inverse orthogonal transformation on the encoded image, and outputs the decoded image to the adder 240.

  In the frame memory 260, when the decoded image decoded by the decoding unit 230 is an encoded image obtained by inter-frame encoding, a prediction image is generated based on the motion vector from the variable length decoding unit 220, Output to the adder 240. In addition, when the decoded image decoded by the decoding unit 230 is an encoded image obtained by intra-frame encoding, since the prediction encoding is not performed, the predicted image is not output.

  The adder 240 adds the decoded image from the decoding unit 230 and the predicted image from the frame memory 260 when the decoded image decoded by the decoding unit 230 is an inter-frame encoded image. The image is restored and output to the two-dimensional FIR filter 250 and also output to the frame memory 260 and stored as a reference image. Note that when the decoded image decoded by the decoding unit 230 is an intra-frame encoded image, it is not predictive-encoded, so the decoded image from the decoding unit 230 is directly used as a two-dimensional FIR filter. Output to 250.

  In the two-dimensional FIR filter 250, the two-dimensional FIR filter configured by the filter coefficient transmitted from the filter coefficient calculation unit 150 in the moving image coding apparatus 100 is applied to the decoded image from the adder 240, for example, for each frame. The encoding error between the original image and the decoded image is minimized to obtain an output image.

  As described above, according to the present embodiment, the encoding error between the input moving image on the moving image encoding device 100 side that is the original image and the decoded image on the moving image decoding device 200 side is designed to be minimal. A filter coefficient is designed, and this filter is applied to the decoded image by the two-dimensional FIR filter 250 of the image decoding apparatus 200, so that a high-quality image can be obtained as compared with the conventional method in which the filter coefficient is designed and filtered on the decoding side alone. Can be obtained.

  Next, the two-dimensional FIR filter 250 in the present embodiment will be described.

  FIG. 3 shows an example of the two-dimensional FIR filter 250 in the present embodiment.

The two-dimensional FIR filter 250 is, for example, a two-dimensional FIR filter of 5 × 5 taps, and the number of taps is set as K ₁ = 2 and K ₂ = 2 as described below.

  In this embodiment, the filter coefficient calculation unit 150 on the moving image coding apparatus 100 side outputs the output image from the two-dimensional FIR filter 250 applied to the decoded image on the moving image decoding apparatus 200 side, and the moving image code. The two-dimensional FIR filter 250 is designed so that the mean square error with the original image that is the input moving image on the conversion apparatus 100 side is minimized. The design method will be described below.

The output image y ^ (n ₁ , n ₂ ) of the two-dimensional FIR filter 250 applied to the decoded image as shown in FIG. 3 is defined by the following equation.

However, u (n ₁ , n ₂ ) (n ₁ ε {0,..., N ₁ −1}, n ₂ ε {0,..., N ₂ −1}) is encoded by the encoding unit 120 and It is a decoded image of N ₁ × N ₂ pixels obtained by the decoding unit 130, or the encoding unit 120 and the decoding unit 230. f (k ₁ , k ₂ ) (k ₁ ε {−k ₁ ,..., k ₁ }, k ₂ ε {−k ₂ ,..., k ₂ }) is a filter coefficient.

Also, the mean square error E (K ₁ , K ₂ ) between the output image y ^ (n ₁ , n ₂ ) of the two-dimensional FIR filter and the original image y (n ₁ , n ₂ ) is expressed by the following equation: (3) It is defined as follows.

The filter coefficients f (k ₁ , k ₂ ) of the two-dimensional FIR filter 250 used in this embodiment are output images y ^ (n ₁ , n ₂ ) from the two-dimensional FIR filter 250 defined by the above equation (3). ) And the original image y (n ₁ , n ₂ ) to minimize the mean square error E (K ₁ , K ₂ ), the following equation (4) is satisfied.

Here, the filter coefficient f (k ₁ , k ₂ ) is defined as the following expression (5) using the vector f.

とする。 However,

And

By substituting the above equation (2) into the above equation (3) and using f shown in the above equation (5), the filter coefficient satisfying the above equation (4) is expressed by the following equation (7). Is done. The filter coefficient calculation unit 150 of the present embodiment calculates a vector f that is a filter coefficient f (k ₁ , k ₂ ) that satisfies the following equation (7), and the two-dimensional FIR filter 250 of the moving picture decoding apparatus 200. To set.

In the above equation (7), the vector r represents the vector of the correlation function between the decoded image and the original image shown below.

である。 However,

It is.

Further, the matrix R represented by the equation (7) represents a self-covariance matrix of the decoded image shown below.

である。 However,

It is.

In the above formula (11), R (i ₁ , i ₂ ) is defined as follows.

As a result, from the above equations (9) and (12), the two-dimensional FIR filter 250 used in the embodiment of the present invention is processed by the encoding unit 120 and the decoding unit 130, or the encoding unit 120 and the decoding unit 230. It can be seen that the decoded image u (n ₁ , n ₂ ) and the original image y (n ₁ , n ₂ ) that is the input image in the moving image coding apparatus 100 can be designed.

By using the two-dimensional FIR filter 250 that minimizes the coding error between such a decoded image u (n ₁ , n ₂ ) and the original image y (n ₁ , n ₂ ), this embodiment is It is possible to obtain a high-quality output image.

  Next, in order to confirm the effectiveness of the present embodiment, the results of experiments using actual images will be briefly described.

  First, the effectiveness of the proposed method for reducing degradation that occurs in a decoded image is confirmed. Next, at the same bit rate, the quality of the image obtained by the method of this embodiment is compared with that of Non-Patent Document 4 described above.

  The results of experiments are shown in order to confirm the effectiveness of the method of the present embodiment regarding the reduction of deterioration in the decoded image, that is, the block distortion generated in the decoded image and the reduction of ringing.

  As an experiment target image, Crowded Crosswalk (720 × 480 pixels, 90 frames), which is an ITE standard moving image, is used. The method of the present embodiment is applied to the luminance component of this image.

However, in the method of the present embodiment, the parameters for quantizing the filter coefficients in the above equation (1) are n _f = 8, n _e = 7, and e _o = 0.

  FIG. 4 shows an example of the encoding condition in the present embodiment.

  As shown in FIG. 4, the encoding condition in this embodiment is MPEG-2 TM5 (ISO / IEC JTC1 / SC29 / WG11 Test Model Editing Committee, “MPEG-2 video test model 5,” ISO / IECJTC1 / SC29 / WG11 Doc.N0400, 1993.), and the GOP structure is M = 3 and N = 15. The motion vector search range is ± 31 × ± 31 for a unidirectional prediction P picture, ± 15 × ± 15 for a bidirectional prediction B picture between one frame, and ± 31 × ± 31 for a bidirectional prediction B picture between two frames. The motion vector search accuracy is half-pixel accuracy, the quantization scale is nonlinear, and the frame rate is 30 fps. The encoding rate is 2 Mbps.

  As a result of the experiment, block distortion generated by encoding as a result of applying the two-dimensional FIR filter 250 used in the method of the present embodiment to a decoded image in which block distortion or ringing has occurred and deteriorated has occurred. , And ringing has been shown to be reduced.

  Next, in order to show the effectiveness of the method of the present embodiment, the results of comparison experiments regarding the quality of the output image obtained by the method of the present embodiment and the method of Non-Patent Document 4 at the same bit rate will be described. To do.

  As the experiment target image, an ITE standard moving image, such as Crowded Crosswalk and Flower Basket (720 × 480 pixels, 90 frames each) is used. The method of the present embodiment and the method of Non-Patent Document 4 are applied to the luminance components of these images.

  Note that encoding is performed based on the encoding conditions shown in FIG. 4, and the above-described values are also used for the parameters of the method of the present embodiment.

In the method of Non-Patent Document 4, the number of directional low-pass filters R and prototypes Z _N ( _q ) and Z _F ( _q ) used for selecting the low-pass filters are as follows. Similarly, R = 4, Z _N ( _q ) = 6, and Z _F ( _q ) = 0.4375 _q (q is the value of the quantization scale).

The PSNR and the bit rate of the image obtained by the method of the present embodiment and the technique of Non-Patent Document 4 are shown in FIGS. The value of PSNR is calculated by the following equation (13).

However, in the above equation (13), b _max is the maximum luminance value of the original image, and MSE is the luminance value b (n ₁ , n ₂ ) of the original image of N ₁ × N ₂ pixels and the luminance value b of the output image. Using ^ (n ₁ , n ₂ ), the following equation (14) is defined.

  FIGS. 5A and 5B are diagrams showing the PSNR of each decoded image in the Crowded Crosswalk image at a bit rate of 2 to 6 Mbps and 11 to 15 Mbps, respectively.

  In FIG. 5, ◯ is the PSNR of the decoded image before the filter processing by the two-dimensional FIR filter 250 of the present embodiment, Δ is the PSNR of the decoded image by the filter processing according to Non-Patent Document 4, and □ is the present embodiment. The PSNR of the decoded image after the filtering process by the two-dimensional FIR filter 250 is shown.

  From FIG. 5, it can be seen that the PSNR is most improved in the decoded image after filtering by the two-dimensional FIR filter 250 of the present embodiment indicated by □ for both the bit rates of 2 to 6 Mbps and 11 to 15 Mbps.

  FIGS. 6A and 6B are diagrams illustrating the PSNR of each decoded image in the Flower Basket image at a bit rate of 2 to 6 Mbps and 11 to 15 Mbps, respectively.

  As in FIG. 5, in FIG. 6, ◯ is the PSNR of the decoded image before the filter processing by the two-dimensional FIR filter 250 of the present embodiment, Δ is the PSNR of the decoded image by the filter processing according to Non-Patent Document 4, □ indicates the PSNR of the decoded image after the filter processing by the two-dimensional FIR filter 250 of the present embodiment.

  From FIG. 6, as in FIG. 5, both the bit rates of 2 to 6 Mbps and 11 to 15 Mbps have the highest PSNR in the decoded image after filtering by the two-dimensional FIR filter 250 of the present embodiment indicated by □. I can see that

  Therefore, from FIGS. 5 and 6, the method of the present embodiment can obtain a high-quality image at the same bit rate in both the Crowded Crosswalk and the Flower Basket images as compared with the method of Non-Patent Document 4. Can be confirmed.

  Furthermore, from FIG. 5B and FIG. 6B, the method of the present embodiment is more effective than the method of Non-Patent Document 4 at the higher bit rate in the case of 11-15 Mbps than in the case of 2-6 Mbps. It can be seen that indicates a high value.

  This is because the method of Non-Patent Document 4 is used to select one of filters having R + 1 types of directivity on the decoding side and apply it to one bit stream, whereas the method of the present embodiment is used. This is because a higher-quality image is obtained at a high bit rate because the two-dimensional FIR filter 250 that minimizes the coding error is used for each frame (image).

  As described above, the method according to the present embodiment can reduce degradation that has occurred in a decoded image, and can obtain a high-quality image at the same bit rate as compared with the decoded image and the technique of Non-Patent Document 4.

  Thus, in this embodiment, in order to reduce block distortion and ringing that occur in the decoded image, a filter that minimizes the encoding error between the original image and the decoded image is designed on the encoding side. The filter coefficient is multiplexed with an encoded image and an encoding parameter such as a motion vector and transmitted to the moving image decoding apparatus 200. In the moving image decoding apparatus 200, the two-dimensional FIR filter 250 filters the decoded image with the filter coefficient. Compared with the conventional method in which filter processing is performed by setting filter coefficients on the decoding side alone, high-frequency components inherent in the original image that is the input moving image are not lost, and the quality of the output image is prevented as much as possible. be able to.

  In addition, as a result of applying the method of the present embodiment to a plurality of images and performing experiments, it has been confirmed that the method of the present embodiment can reduce deterioration caused by encoding. Furthermore, it was confirmed that the method of the present embodiment can obtain a higher quality image than the conventional method at the same bit rate.

  In the present embodiment, the filter coefficient used in the method of the present embodiment in the experiment is quantized to a 16-bit floating-point value, but the present invention is not limited to this. Of course, the number of bits to be added may be reduced by setting an appropriate value for the number of bits in the exponent part.

  Further, in the present embodiment, the filter coefficient calculation unit 150 of the moving image coding apparatus 100 has been described to detect the filter coefficient for each frame. However, the present invention is not limited to this, and for each GOP or the like. Of course, the filter coefficient may be calculated every predetermined number of frames or every arbitrary frame such as a scene change, and transmitted to the decoding side every predetermined number of frames or every arbitrary frame. In this way, if the filter coefficient is calculated every predetermined number of frames or every arbitrary frame and transmitted to the decoding side, the number of times the filter coefficient is calculated and transmitted is reduced, thereby reducing the processing load and the amount of transmission data. It becomes possible to do.

  Further, in the present embodiment, the moving image encoding apparatus 100 has been described so that the locally decoded decoded image is not filtered by the two-dimensional FIR filter as shown in FIG. Of course, as shown in the moving image encoding apparatus 300, the decoded image subjected to local decoding may be filtered by a two-dimensional FIR filter 155 and stored in the frame memory 160 as a reference image. In this case, the moving picture decoding apparatus 400 stores the decoding after the filtering process by the two-dimensional FIR filter 250 in the frame memory 260 as a reference image, as shown in FIG. In this way, since motion compensation prediction can be performed using the reference images filtered by the two-dimensional FIR filters 155 and 250, the processing load increases, but the motion compensation prediction becomes more accurate, and the encoded image and code It is possible to reduce the amount of transmission to the decoding side such as the conversion parameters.

  Further, in the present embodiment, filter coefficient calculation section 150 in moving image coding apparatus 100 is provided after adder 140 so that the mean square error between the input moving image that is the original image and the decoded image is minimized. However, the present invention is not limited to this, and the filter coefficient calculation unit 150 is provided after the DCT in the encoding unit 120 and before the quantization process. But of course. In this case, the image to be compared with the DCT coefficient is the inverse DCT coefficient after the inverse DCT in the decoding unit 130 and before the inverse quantization process, so the average square between the DCT coefficient and the inverse DCT coefficient By calculating the filter coefficient so as to minimize the error, it is possible to set the filter coefficient in which the coding error due to the quantization error is reduced among the coding errors.

  Further, in the present embodiment, the two-dimensional FIR filter 250 that is a cyclic filter has been described as an example of a filter for filtering a decoded image on the video decoding device 200 side. However, the present invention is not limited to this, An IIR (Infinite Impulse Response) filter, which is an acyclic filter, may be used, and the filter type may be either asymmetric or symmetric.

  Further, in the present embodiment, when the filter coefficient of the two-dimensional FIR filter 250 is obtained on the moving image coding apparatus 100 side, the output image and the input image after the filtering process by the two-dimensional FIR filter 250 using the arithmetic expression as described above. However, the present invention is not limited to this, and the two-dimensional FIR filter is not limited to this, but is obtained by a technique such as fractal theory, chaos theory, or genetic algorithm (GA). Of course, a value that minimizes the mean square error between the output image after the filtering process by 250 and the input image may be obtained.

  In the present embodiment, the filter coefficient of the two-dimensional FIR filter 250 that minimizes the coding error is calculated for each frame or for each predetermined frame. However, the present invention is not limited to this. Of course, the filter coefficients of the two-dimensional FIR filter 250 may be calculated for each 16 × 16 macroblock, 8 × 8 block, or block of any size.

  Also, in the present embodiment, the moving image encoding apparatus 100 and the moving image decoding apparatus 200 are described as hardware (circuit) as shown in FIGS. 1 and 2, respectively. Not limited to, for example, software, that is, a moving image encoding program and a moving image decoding program for executing functions similar to those of the moving image encoding device 100 and the moving image decoding device 200, such as a CD. Of course, the program may be stored in a memory or a hard disk in a computer via a recording medium or a network, and the CPU in the computer may read and execute the program.

  According to the moving image encoding device, the moving image encoding method, and the moving image encoding program of the present invention, the filter coefficient of the two-dimensional FIR filter of the moving image decoding device is calculated based on the input moving image and the decoded image. Then, the filter coefficient is transmitted to the moving image decoding apparatus side together with the encoded image. On the other hand, according to the moving image decoding apparatus, the moving image decoding method and the moving image decoding program of the present invention, the filter coefficient is input to the moving image encoding apparatus. The filter coefficient calculated based on the input moving image and the decoded image locally decoded by the moving image encoding device and the encoded image are input, the encoded image is decoded to generate a decoded image, and the filter Since the decoded image is subjected to the two-dimensional FIR filter processing by the coefficient to be the output image, the high frequency inherent in the original image which is the input moving image is compared with the conventional method in which the decoding side performs the filtering process alone. There is an advantageous effect that the quality of the output image can be prevented from being deteriorated as much as possible without losing components, and MPEG-1 encoding or MPEG- in terrestrial digital broadcasting, moving image distribution on the DVD or the Internet, etc. 2, MPEG-4, H.264. 261, H.H. 263, H.M. It is useful for various encoding methods and decoding methods such as H.264 / AVC encoding.

The figure which shows the structure of embodiment of the moving image encoder which concerns on this invention. The figure which shows the structure of embodiment of the moving image decoding apparatus which concerns on this invention. The figure which shows an example of the two-dimensional FIR filter (5x5 tap) in this Embodiment The figure which shows the encoding conditions in this Embodiment (A), (b) is a diagram showing the PSNR of each decoded image in a Crowded Crosswalk image at bit rates of 2 to 6 Mbps and 11 to 15 Mbps, respectively. The figure which shows PSNR of each decoded image in the Flower Basket image in the bit rate of 2-6 Mbps and 11-15 Mbps, respectively (a), (b). The figure which shows another structural example of embodiment of the moving image encoder which concerns on this invention. The figure which shows another structural example of embodiment of the moving image decoding apparatus which concerns on this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100,300 Video encoding apparatus 110 Subtractor 120 Encoding part 130 Decoding part 140 Adder 150 Filter coefficient calculation part 160 Frame memory 170 Motion detection part 180 Variable length encoding part 190 Transmission part 200,400 Video decoding apparatus 210 Receiving unit 220 Variable length decoding unit 230 Decoding unit 240 Adder 155, 250 Two-dimensional FIR filter 260 Frame memory

Claims (9)

An image encoding device that encodes an input image to generate an encoded image, locally decodes the encoded image to generate a decoded image, and transmits the encoded image to an image decoding device,
A filter coefficient calculation unit that calculates a filter coefficient of a filter of the image decoding device based on the input image and the decoded image;
A transmission unit that transmits the filter coefficient together with the encoded image to the image decoding device;
An image encoding apparatus comprising: The image encoding device according to claim 1,
The filter is a two-dimensional FIR filter;
The filter coefficient of the filter is a value that minimizes the mean square error between the output image after filtering by the two-dimensional FIR filter and the input image.
An image encoding apparatus characterized by that. The image encoding device according to claim 1,
The image coding apparatus, wherein the input image is a moving image. An image decoding device that decodes an encoded image encoded by an image encoding device to produce an output image,
A reception unit that receives an input image input to the image encoding device, a filter coefficient calculated based on a decoded image locally decoded by the image encoding device, and the encoded image;
A decoding unit that decodes the encoded image to generate a decoded image;
A filter that filters the decoded image generated by the decoding unit according to the filter coefficient into an output image;
An image decoding apparatus comprising: An image encoding method for generating an encoded image by encoding an input image, generating a decoded image by locally decoding the encoded image, and transmitting the encoded image to an image decoding device,
Calculating a filter coefficient of a filter of the image decoding device based on the input image and the decoded image, and transmitting the filter coefficient to the image decoding device side together with the encoded image;
An image encoding method characterized by the above. The image encoding method according to claim 5, comprising:
The filter coefficient of the filter is a value that minimizes the mean square error between the output image after filtering by the filter and the input image.
An image encoding method characterized by the above. An image decoding method for decoding an encoded image encoded by an image encoding device into an output image,
A filter coefficient calculated based on an input image input to the image encoding device and a decoded image locally decoded by the image encoding device and the encoded image are input, and the encoded image is decoded. A decoded image is generated, and the decoded image is filtered by the filter coefficient to obtain an output image.
An image decoding method characterized by the above. Image coding for generating a coded image by coding an input image, generating a decoded image by locally decoding the coded image and transmitting the coded image to an image decoding apparatus. A program for
Based on the input image and the decoded image, the filter coefficient of the filter on the image decoding device side is calculated, and the filter coefficient is transmitted to the image decoding device side together with the encoded image.
An image encoding program which causes a computer to execute the above. An image decoding program for causing a computer to execute a process for causing a computer to execute a process for decoding an encoded image encoded by an image encoding device into an output image,
A filter coefficient calculated based on an input image input to the image encoding device and a decoded image locally decoded by the image encoding device and the encoded image are input, and the encoded image is decoded. A decoded image is generated, and the decoded image is filtered by the filter coefficient to obtain an output image.
An image decoding program that causes a computer to execute the above.

Images