JPH07336687A - Picture encoding device - Google Patents

Abstract

PURPOSE:To attain the reduction of a circuit scale and the shortening of processing time in a picture encoding device based on residual encoding in a frequency domain. CONSTITUTION:The band picture (input picture-F) of every frequency band is obtained by a picture converter 102, and the residual o.f this from a motion compensated predictive picture-F is obtained. Further, a central value is substituted for this residual (quantizer 104), and the sum of this residual and the motion compensated predictive picture-F is stored in a picture memory 107. In a succeeding frame, a motion estimator 108 determines a motion vector on the basis of the band picture-F of the picture memory 107 and an input picture-F, and a motion compensator 109 generates the motion compensated picture-F by using this, and outputs it.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding apparatus for transmitting and storing an image with a small coding amount.

[0002]

2. Description of the Related Art As a conventional image coding apparatus, for example, CCITT Recommendation H.264 is recommended. 261 (Prior Art Example 1) is disclosed in H.261. The image size handled by H. 261 is 144 x 180 pixels vertically and horizontally, and the encoding amount is 6 per second.
4 kilobits or more. This conventional example 1 uses a discrete cosine transform (DCT) in block units for image conversion,
In addition, the amount of information is reduced by motion-compensated interframe prediction. However, the conversion in block units causes the following problems.

At (A1) block boundary, a boundary not existing in the original image is perceived. This problem (A1) is generally called block distortion. As a solution to this problem, a technique using subband decomposition for image conversion is disclosed in IEICE Technical Report IE91-82 (conventional example 2). This configuration is shown in FIG.

In FIG. 6, 601 is an image input terminal, and 603.
Is a differentiator, and 604 is a quantizer, which replaces an input value in a certain range with one representative value, as represented by the input / output characteristic of FIG. Reference numeral 606 is an adder, 607 is an image memory for storing a predicted image, and 608 is a motion estimator for obtaining a motion vector for each block by the block matching method. For the block matching method, for example, Nikkan Kogyo Shimbun "Multidimensional signal processing of TV image" by Takahiko Fukibe, Section 6.3.2 can be used.

Numeral 609 is a motion compensator for generating a motion-compensated predicted image from a predicted image based on a motion vector, and a description of a block-based compensation method is given in the magazine "Interface".
Aug.'92, p125, section 1.2. Output terminals 610 and 611 respectively output the residual and the motion vector to the encoder at the subsequent stage. Reference numerals 602 and 612 are image converters, and 613 is an image inverse converter.

The structures of the image converters 602 and 612 and the image inverse converter 613 will be described with reference to FIGS. 7 and 8. In FIG. 7, 701 is an input terminal, 702 is a horizontal filter H0, 703 is a horizontal filter H1, 704, 705 is a 2: 1 down-sampling circuit that thins out the number of horizontal pixels every other pixel, and 706 and 708 are vertical filters H0, 707 and 709. Is a vertical filter H1, 710,711,71
2,713 is a 2: 1 down-sampling circuit for thinning out the number of vertical pixels every other pixel, and 714,715,716,717 are output terminals.

In FIG. 8, 801, 802, 803, 804 are input terminals, 805, 806, 807, 808 are 1: 2 upsampling circuits for inserting pixels having a luminance value of zero every other pixel in the vertical direction, 809, 811 are vertical filters G0, 810, 812 are vertical filters. G1, 813, 814, 819 are adder circuits, and 815, 816 insert pixels having a luminance value of zero every other horizontal pixel 1: 2.
Up-sampling circuit, 817 is horizontal filter G0, 818
Is a vertical filter G1, and 821 is an output terminal.

The image conversion circuit 602, which is configured as described above,
The operations of 612 and the image inverse conversion circuit 613 will be described below. The structure of each filter is expressed in the Z conversion format (Equation 1) to (Equation 8).
Shown in. Z _h ^-1 and Z _v ^-1 are delay operators of 1 horizontal pixel and 1 vertical pixel, respectively.

[0009]

[Equation 1]

[0010]

[Equation 2]

[0011]

[Equation 3]

[0012]

[Equation 4]

[0013]

[Equation 5]

[0014]

[Equation 6]

[0015]

[Equation 7]

[0016]

[Equation 8]

Among the filters, the one labeled H0 at the end of the name corresponds to the low bandpass filter, and the one labeled H1 corresponds to the high bandpass filter. In the image converter of this embodiment,
With the above filter configuration (Fig. 7), LL, LH,
Images are output for each of the four frequency bands HL and HH. Each of these is called a band image. Further, hereinafter, in order to represent an image in such a frequency domain, -F is added to the end. For example, write band image-F. Further, as shown in FIG. 8, the image inverse converter generates an image in the spatial domain based on the input LL to HH band image-F, and outputs this.

The filter coefficient shown in the present embodiment is the same as that in the document "IEEE International Conference Acoustic Speech Signal Processing" (D. Gall, A. Tabatani: "Su.
b-band Coding of Digital Im
ages Using Symmetric Short
Kernel Filters and Arithme
tic Coding Technologies ", Pro
c. IEEE IntConf. Acoustic Sp
ech Signal Processing, pp.
761-764, April, 1988). If there is no quantization error, the image is completely reconstructed through subband division and subband synthesis.
That is, by inputting the band image-F obtained by the image converter to the image inverse converter, the input image of the original image converter can be obtained.

The operation of the conventional example 2 including the above components will be described with reference to FIG. In the conventional examples and examples described in this specification, in order to easily understand the operation of the actual device, the image is composed of 144 × 176 pixels in the vertical and horizontal directions, and the block for performing the correlation calculation is 8 pixels in the vertical direction and 8 pixels in the horizontal direction. It is composed of. The output of the image converter is the band image-F, and the signal line for transmitting this component image is represented by a thick line.

First, the motion estimator 608 obtains a motion vector in block units based on the predicted image which is the processing result in the previous frame recorded in the image memory 607 and the input image from the input terminal 601. This motion vector is output from the terminal 611 and also sent to the motion compensator 609. The predicted image in the image memory 607 is converted into a motion-compensated predicted image in the motion compensator 609 and further into a motion-compensated predicted image-F in the image converter 612 based on this motion vector. The difference between the motion-compensated predicted image-F and the input image-F output from the image converter 602 is calculated by the differentiator 603, and further converted into a representative value by the quantizer 604. The residual image-F composed of these representative values is output from the terminal 610, and the adder 606 outputs the residual image-F.
The sum of the two outputs is calculated. This is the predicted image-F and matches the input image-F except for the quantization error. This is converted into a predicted image using the image inverse converter 613 and recorded in the image memory 607.

The above operation is performed for each frame of the input image, and the residual image-F and the motion vector are output from the terminals 610 and 611, respectively. Image coding can be performed by passing these outputs through a coder in the subsequent stage to obtain a code. Further, the image decoding apparatus is composed of a group of constituent elements 614, which decodes the above codes to obtain a residual image-F and a motion vector, and inputs these to output a predicted image. In addition,
The operation of the image decoding device is the same as the operation of the image encoding device, and therefore its explanation is omitted.

With the above configuration, the problem (A1) can be solved by using the subband decomposition in which the data used for conversion is not closed in the block. In addition,
In the second conventional example, the difference from the motion-compensated predicted image -F is calculated in the frequency domain after image conversion. This is to solve the following problem (A2).

First, as a predicted image-F, in addition to motion-compensated inter-frame prediction (inter), intra-frame prediction (intra) is added, and this is expanded so as to be switched for each local area. Then, in the case of difference in the spatial domain, (intra) and (i
There is a problem (A2) that a signal level step is generated at the boundary of the local area of (nter), and this is subband decomposed to deteriorate the coding efficiency. In the conventional example 2, this problem (A
2) is solved by taking the difference in the frequency domain after image conversion. By the way, since the motion-compensated inter-frame prediction used in Conventional Examples 1 and 2 utilizes temporal continuity, the following problems occur.

(A3) The part appearing from behind the moving body is
Since there is no time continuity, the prediction error increases and the amount of coding in the subsequent stage increases. This will be referred to as the uncover problem hereinafter. As a solution to this problem (A3), there is one disclosed in JP-A-61-1114677 (Prior art example 3).

[0025]

However, even if the above-mentioned conventional technique is used, the following problems still exist. (B1) In the conventional example 2, since the block matching method is used for motion estimation, an image inverse converter for obtaining a predicted image in the spatial domain is required. In addition, an image converter for the motion-compensated prediction image is required to perform the difference in the frequency domain.

As described above, in the second conventional example, a total of three image converters and inverse converters are required. The image converter and the inverse converter require a lot of units in hardware configuration, which affects power consumption and price. Further, it requires a lot of processing time, which is a rate-determining part of the real-time operation. For these reasons, it is better to have as few such converters as possible.
Further, when the conventional example 2 is applied to coding of about 20 kilobits per second (referred to as ultra-low bit rate coding), the following problems occur. (B2) Due to the limitation of the coding amount, among the residuals in the frequency domain However, the high band component cannot be transmitted and the image quality deteriorates. This is described, for example, in CCITT Recommendation H.264. 261 is caused by the above operation. When the limit of the coding amount is about to be exceeded, the quantization width is increased in advance. As a result, it is possible to reduce information by ignoring small fluctuations in the value, and it is possible to satisfy the limitation of the coding amount. However, as a result, all small fluctuations in the value in the high band component are replaced with a representative value of zero, and as a result, the high band component cannot be transmitted. For this reason,
An edge composed of the high band component in the previous frame remains in the predicted image, causing image deterioration. This state is shown in FIGS. 9 and 10.

Consider a case where a square 902 in an input image 901 is moved to the left as shown in FIG. By applying image conversion to the input image 901, LL, LH, H for each frequency band
Obtain L, HH image-F (903). Let us consider the case where this is predicted image-F and the residual with this is sent in the next frame. Moreover,
It is assumed that the residual can be transmitted only in the LL band image due to the limitation of the coding amount. Then, the image of the result of adding the residuals-F is 904
As described above, all band images other than LL are the same as the prediction signal-F903. When this image-F904 is inversely transformed, as shown by 905, it becomes an image with deterioration in vertical edge portions. This occurs because the high frequency in the HL band is relatively displaced from the LL band. In FIG. 10, reference numeral 906 shows the luminance distribution on the cross section ab at this time. A portion 907 surrounded by a rectangular dotted line is a portion where image deterioration occurs.

Furthermore, when the background prediction method shown in Conventional Example 3 is introduced into the first embodiment (described later) which solves the problem (B1), the following problems occur. (B3) In the image in each frequency band after the image conversion, the difference between the input image and the background image is smaller than that in the image before conversion, so that it is difficult to set the threshold value.

The present invention is to solve the above-mentioned problems, and in an image coding apparatus by residual coding in the frequency domain,
The purpose is to reduce the circuit scale and processing time.

[0030]

In order to achieve the above object, a first aspect of the invention is to provide an image conversion means for decomposing an image into a plurality of images for each frequency band, and an image memory for storing the image for each frequency band. A motion estimation means for calculating a motion vector in a local area based on the image for each frequency band output by the image conversion means and those output by the image memory; and the image memory based on the motion vector. Is provided with a motion compensation means for performing motion compensation on each image for each frequency band output for each local region, and outputting this as a motion-compensated prediction image, and a motion vector is calculated based on the image for each frequency band. Based on the configuration, a motion-compensated prediction image is obtained for each frequency band.

In the second invention, an image conversion means for decomposing an image into a plurality of images for each frequency band, an image memory for storing the image for each frequency band, and each frequency band output by the image conversion means. Of the image and subtraction means for obtaining a residual between them output from the image memory, quantizing means for converting the residual into a representative value, an image for each frequency band composed of the representative value, and the image memory And an addition unit for obtaining the sum of them to be output, an image for each frequency band composed of the sum, and a selection unit for selecting one of them having a value of 0 and outputting this to the image memory,
The image output by the selector is changed for each frequency band based on the quantization width change signal from the encoder that encodes the image for each frequency band made up of the representative value.

Further, a third aspect of the invention is an image conversion means for decomposing an image into a plurality of images for each frequency band, an image memory for storing an image for each frequency band, and each frequency band output by the image conversion means. Means for obtaining a residual difference between the image and those output by the image memory, a quantizing means for converting the residual into a representative value, an image for each frequency band composed of the representative value, and an output of the image memory The addition means for obtaining the sum with them, the image for each frequency band composed of the sum and those output from the image memory are input, and the total sum of the absolute values of the differences between these inputs in the local region is calculated for each frequency band. The configuration is provided with a control means for selecting one of two inputs as an image to be output to the image memory by using the sum of the two.

[0033]

In the first aspect of the invention, the motion estimating means estimates the motion vector using the image for each frequency band. Further, the motion compensation means uses this motion vector to perform motion compensation on the image for each frequency band output from the image memory. Since the motion compensation image is obtained in this way, only the image in the frequency domain can be processed. That is, as in Conventional Example 1,
It does not require an image inverse transforming means from the frequency domain to the spatial domain or an image transforming means for transforming the motion-compensated image from the spatial domain to the frequency domain, and the processing time is shortened accordingly.
It is possible to reduce the device cost when implementing hardware.

According to the second aspect of the invention, when the quantization width of a certain frequency band is increased by the quantization width changing signal from the encoder at the subsequent stage, an image having a value of 0 is output instead of the output of the adding means. Output as a band image. As a result, the quantization width is large, and the image in the high frequency band output by the quantizing unit has only the value 0, and the image in the high frequency band in the output of the adding unit becomes the same as the previous value. It is possible to avoid deterioration.

According to the third aspect of the invention, the background / moving object identification necessary for updating the background image for each frequency band stored in the image memory is identified by the background image and the predicted image output from the adding means. Judgment is made based on the sum of absolute differences for each frequency band. As a result, it is possible to avoid the difficulty of setting the threshold value, which occurs when the threshold value is set in the image for each frequency band and the identification is performed for each frequency band.

[0036]

Embodiments of the present invention will be described below with reference to the drawings. A first embodiment of the present invention will be described with reference to FIGS. FIG. 1 is a configuration diagram of the first embodiment.

In FIG. 1, 101 is an input terminal, 102 is an image converter, 103 is a subtractor, 104 is a quantizer, 105 is an inverse quantizer, and 106 is an adder. It is the same as the component. Further, 107 is an image memory for recording one frame of four images of each frequency band LL, LH, HL, HH after image conversion, 108 is a motion estimator, and 109 is a motion compensator. Also, 110 and 110 are output terminals, which are connected to the encoder in the subsequent stage. Here, the operation of the motion estimator 108 will be described with reference to FIG.

In FIG. 2, the movement amount of the block that maximizes the correlation in the local block is calculated as the motion vector based on the converted image. In FIG. 2, 201 is a correlation calculation circuit, 202 is a motion vector storage memory, and the above are the components of the motion estimator 108. 210 is a band image output from the image memory 107 (prediction image-F in FIG. 1), 212
Is a band image of the input image (input image-F).

Next, the operation will be described. The correlation calculation circuit 201 is composed of two processes. First, the first process will be described. The LL band image of the band image 212 is divided into 64. This divided one element consists of 9 pixels vertically and 11 pixels horizontally, and this is called a block. The upper left coordinates of the block are represented by (Equation 9).

[0040]

[Equation 9]

Similarly, the LH, HL, and HH band images are also divided into 64,
Based on (Equation 10), a group 213 of four blocks corresponding in position is extracted from the LL to HH band images.

[0042]

[Equation 10]

Next, the block group 211 having the upper left coordinates represented by (Equation 9) is extracted from the band image 210 based on (Equation 10). The correlation value of these block groups is defined by (Equation 11), and the displacement (Δj, Δ
i) is calculated.

[0044]

[Equation 11]

The above is the first processing. Next, the correlation calculation circuit 201 and the second processing will be described. This is a band image 210
The position of the block to be extracted from is changed by the amount represented by (Equation 12) and the correlation operation (Equation 11) is performed to obtain the displacement (ξ, η)
Is to seek.

[0046]

[Equation 12]

Along with this, a pixel value at the intermediate point of the pixel points is required, and this is obtained by bilinear linear complementation (Equation 13).

[0048]

[Equation 13]

Block group 201 obtained as described above
And the correlation value of the block group 213 (Equation 14) is minimized.
Find the displacement (ξ, η) in subpixel units.

[0050]

[Equation 14]

A motion vector based on the displacement in pixel units and subpixel units obtained by the above two-step processing
(Δj + ξ, Δi + η) is calculated, and this is recorded in the motion vector storage memory using (j0, i0) as an index.

The above operation in the correlation calculation circuit 201 and the motion vector storage memory 202 is executed for each block (j0, i0). The above is the operation of the motion estimator 108.

The processing procedure of the first embodiment having such components will be described below. Input terminal 101
The input image of is decomposed into four LL to HH band images by the image converter. Also, the LL to HH band image and the LL to HH band image recorded in the image memory 107 are input, and the motion estimator calculates a motion vector for each block. Using this motion vector and the LL to HH images in the image memory 107, the motion compensator 109 performs motion compensation in block units, and outputs this as a motion-compensated predicted image-F. At this time, motion compensation in units of sub-pixels is performed by complementation obtained by (Equation 13).

Motion-compensated prediction image obtained as described above
-F and the input image -F output by the image converter 102,
The subtractor 103 obtains the pixel value difference for each position for each of the LL to HH band images. This difference is replaced by a representative value in the quantizer 104 for each pixel value according to the conversion characteristic (FIG. 3), and is output from the output terminal 110. Further, the motion vector output from the motion estimator 108 is output from the output terminal 111. The above two are passed to the encoder at the subsequent stage. Also the quantizer
The representative value after the difference output by 104 is added to the motion-compensated prediction image-F by the adder 106, and this is recorded in the image memory 107 as the prediction image-F. This predicted image-F is input to the motion estimator 108 and the motion compensated image 109 in the next frame.

By the above operation, the motion-compensated predicted image-F is generated for the input image for each frame based on the predicted image-F in the previous frame, and the residual difference between this and the input image-F is calculated. Output. The image decoding apparatus is a dotted rectangle 112 in FIG.
Of image components, and an image inverse converter that forms a pair with the image converter 102, receives the representative value of the residual and the motion vector output from the decoder, obtains a motion-compensated prediction signal, and inverse-transforms this. What is done is output as a decoding result. The image encoding can be realized as described above.

The image converter 102 may be a block unit Fourier transform or a wavelet transform, in addition to the subband transform. Further, the image shift in the spatial domain can be performed by using only the phase component without changing the absolute value of the image (complex number) in the frequency domain. The image converter 102, the motion estimator 108, and the motion compensator 109 in this case will be described. The image converter 102 divides the input image (144 vertical pixels, 176 horizontal pixels) into 18 vertical blocks and 22 horizontal blocks, and this block (composed of vertical 8 horizontal 8 pixels) The Fourier transform coefficients a (q, p) and b (q, p) are converted by the equation 16).

[0057]

[Equation 15]

[0058]

[Equation 16]

The motion estimator 108 uses the Fourier transform coefficients a1 and b1 output from the image converter 102 and those a2 and b2 stored in the image memory 107, and the motion vector (u, v) is calculated by (Equation 17). To calculate. The above motion vector is calculated for each block.

[0060]

[Equation 17]

The motion compensator 109 outputs Fourier transform coefficients a3 and b3 based on (Equation 18) and (Equation 19).

[0062]

[Equation 18]

[0063]

[Formula 19]

Of the sine component and cosine component, only the cosine component may be encoded and transmitted, and the sine component may be restored by the Hilbert transform on the decoder side. This can further reduce the information to be transmitted. Further, a window function may be applied to the data in the block at the time of Fourier transform.

Next, the second embodiment of the present invention will be described with reference to FIG.
Will be explained. The input terminal 101, the image converter 102, the subtractor 103 that takes the difference, the adder 106, the image memory 107, the motion estimator 108, the motion compensator 109, and the output terminals 110 and 111 are the same as the constituent elements of the first embodiment. It is a thing. 212 is a control terminal that receives a signal for controlling the quantization width based on the encoding amount in the encoder in the subsequent stage, and 204 is based on this control signal,
A quantizer having a function of changing the quantization width for each of the band images LL to HH, and a selector 213 for selecting a signal based on a control input.

The operation of the second embodiment composed of the above components will be described only in the part different from that of the first embodiment.
When the coding amount is about to exceed the limit, the latter-stage encoder sends the quantization width wLL, wLH,
Output wHL and wHH. This control signal is first sent to the quantizer 2
It is transmitted to 04, and the quantization width for each band image is changed.
In addition, in FIG. 3 showing the conversion characteristic of the quantizer, the quantization width is 4
Is the case. Also, the selector 213 selects the terminal a when the control signal is equal to the maximum value of the signal, and the terminal b otherwise.
Is to be selected. The maximum value of the signal is the maximum value of the pixel value of the band image that can be handled by this device. For example, the input of the quantizer is a digital signal and the signal line is 8
In the case of a book, the signal takes a value from -128 to 127, so the maximum value of the signal is 127.

With the above configuration, when the quantization width w of a certain high-band image becomes large and becomes equal to the maximum value of the signal, the output of the quantizer 204 becomes zero and the terminal b of the selector 213 is shown in FIG. A motion-compensated prediction image -F that is not corrected by the residual output from the subtractor 103 is input to the. At this time, the selector 213 does not select this, but selects the value 0 of the terminal a. In this way, the high-band image can be replaced by an image consisting of the value 0, instead of the signal which has not been modified by the residual, as explained in the section of the operation. The selector terminal a may be selected when the quantization width is equal to or larger than a certain value.

Next, a third embodiment of the present invention will be described with reference to FIG. An input terminal 101, an image converter 102, a subtractor 103 that takes a difference, a quantizer 104, an adder 106, an image memory 107, a motion estimator 108, a motion compensator 109, and an output terminal 11.
0 and 111 are the same constituent elements as in the first embodiment.

Further, reference numeral 512 is a group of constituent elements for generating the motion-compensated prediction image-F. 513 is an image memory for storing the background image-F consisting of four band images, 514 is a controller for generating this background image-F from the predicted image-F,
The above two are the constituent elements 51 for generating the background image-F.
Is 1. 515 is a selector that outputs one of the background image-F and the motion-compensated predicted image-F, and 516 is a controller that outputs a selection signal. Like the conventional example 3, the predicted image-F, that is, the input image- Whether F is closer to the background image-F or the motion compensation prediction image-F is recorded for each pixel as a selection result. Then, as a control signal for a pixel to the selector 515,
A control signal for selecting the one having the highest selection frequency out of the selection results in the vicinity of the pixel is output.

Here, the operation of the component group 511 will be described. Background image output from the image memory 513-Ffb and adder 10
Based on the predicted image-Ffp output by 6, the controller 514
The background image-Ffb is updated based on (Equation 14).

The above processing is performed for each pixel of each band image. Note that e (j0, i0) is the sum of the residuals in the block B (j0, i0), which is further added for each band image.
That is, e (j0, i0) is the block B (j0, i0)
It represents the size of the residual in the image area before conversion. (Equation 14)
Determines that the region with a large e (j0, i0) is a moving object, and that the predicted image-Ffp at this point, that is, the signal representing the moving object is not included in the background image-Ffb, The background image-Ffb with is the next background image-F. On the other hand, when e (j0, i0) is small, the prediction signal fp is used and the background image -Ffb is changed by the update coefficient α.

The background image-F is generated as described above. The background image-F generation method shown here is based on the IPSJ Transactions Vol.28 No.4 "Moveable Object Image Extraction Technology",
The method disclosed in Apr.'87 is applied to band images LL to HH.

The operation of the third embodiment constructed as above will be described below. Similar to the first embodiment,
The input image of the input terminal 101 is decomposed into a band image (input image-F) by the image converter 102.

The subtractor 103 calculates the residual between this input image-F and the band image-F output by the selector 515. Further, the difference for each band image-F is replaced with a representative value in the quantizer 104 (input / output characteristic FIG. 3) and output from the output terminal 110. Further, the sum of the band image-F including the representative value and the band image-F output by the selector 515 is calculated by the adder 106. This is called prediction image-F. Using this predicted image-F, the constituent element group 514 generates the background image-F by the operation described above. Furthermore, using the predicted image-F and the band image-F output from the image converter 102, the component group 512 outputs the motion-compensated predicted image-F by the operation described in the first embodiment.

The above background image-F and motion compensation prediction image-F
Is selected by the selector 515 according to the operation of the controller 516. The representative value of the residual between the selected image-F and the input image-F is output from the output terminal 110, and the predicted image-F is obtained by adding the residual to the selected image-F.

As described above, the residual image-F is output from the output terminal 110 and the motion vector is output from the output terminal 111. These are encoded and transmitted by the encoder in the subsequent stage. Further, the image decoder is composed of the component group 517, decodes the transmission data, obtains the residual image-F and the motion vector, and based on these, the prediction image output from the adder 106 based on the above-mentioned operation. -F is obtained, and a reproduced image is obtained by passing this through an image inverse converter that is paired with the image converter 102.

[0077]

As described above, by using the image coding apparatus of the first invention, the number of image converters and inverse converters can be reduced, and the processing time can be shortened when software is implemented by a general-purpose processor. You can Further, when hardware is realized, a device is unnecessary, and the device cost and power consumption can be reduced accordingly.

Further, in the image coding apparatus of the second invention,
It is possible to suppress image quality deterioration when high band frequencies cannot be transmitted by ultra-low bit rate encoding.

Further, in the image coding apparatus of the third invention,
It is possible to easily set the threshold value when the background image is generated, and to stably generate the background as compared with the case where the background image is generated for each frequency band image.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an image coding apparatus according to a first embodiment of the present invention.

FIG. 2 is a diagram showing an operation of a motion estimator in the first embodiment.

FIG. 3 is a diagram showing the operation of a quantizer.

FIG. 4 is a block diagram showing a configuration of an image coding apparatus according to a second embodiment of the present invention.

FIG. 5 is a block diagram showing a configuration of an image coding apparatus according to a third embodiment of the present invention.

FIG. 6 is a block diagram showing the configuration of an image inverse conversion circuit in Conventional Example 1.

FIG. 7 is a block diagram showing a configuration of an image converter.

FIG. 8 is a block diagram showing the configuration of an image inverse converter.

FIG. 9 is a diagram showing a task (B2).

FIG. 10 is a diagram showing a task (B2).

[Explanation of symbols]

 102 image converter 103 subtractor 104 quantizer 106 adder 107 image memory 108 motion estimator 109 motion compensator 204 quantizer 212 output terminal 213 selector 513 image memory 514 controller 515 selector 516 controller

Claims (3)

[Claims] 1. An image conversion unit that decomposes an image into a plurality of images for each frequency band, an image memory that stores an image for each frequency band, an image for each frequency band output by the image conversion unit, and the image memory. Motion estimation means for calculating a motion vector in a local area based on those output by
An image for each frequency band is provided based on the motion vector, and motion compensation means for performing motion compensation for each local region on the image for each frequency band output from the image memory and outputting this as a motion-compensated predicted image. An image coding apparatus, characterized in that a motion vector is calculated based on the above, and a motion compensation prediction image for each frequency band is obtained based on this. 2. An image conversion means for decomposing an image into a plurality of images for each frequency band, an image memory for storing the image for each frequency band, an image for each frequency band output by the image conversion means, and the image memory. Of subtracting means for obtaining the residual with them, the quantizing means for converting the residual into a representative value, the sum of the image for each frequency band consisting of the representative value and those output by the image memory. For each frequency band composed of the representative value, there is provided an addition means for obtaining, an image for each frequency band composed of the sum, and a selection means for selecting one of those having a value of 0 and outputting this to the image memory. An image encoding device for changing the image output from the selector for each frequency band based on a quantization width change signal from an encoder for encoding the image. 3. An image conversion means for decomposing an image into a plurality of images for each frequency band, an image memory for storing an image for each frequency band, an image for each frequency band output by the image conversion means, and the image memory. Of subtracting means for obtaining the residual with them, the quantizing means for converting the residual into a representative value, the sum of the image for each frequency band consisting of the representative value and those output by the image memory. The addition means to be obtained, the image for each frequency band composed of the sum and those output from the image memory are used as inputs, and the sum of the absolute values of the differences between these inputs in the local region is used for each frequency band. An image coding apparatus, comprising: a control unit that selects one of two inputs as an image to be output to the image memory.