JP4231386B2 - Resolution scalable decoding method and apparatus - Google Patents

Description

  The present invention relates to a resolution scalable decoding method and apparatus, and more particularly, to a resolution scalable decoding method and apparatus for efficiently transmitting and storing moving images.

As an international standard for image coding,
(1) MPEG for moving images (for example, see Non-Patent Document 1):
(2) JPEG2000 for still images (for example, see Non-Patent Document 2):
Is well known. MPEG is a technique using motion compensation and DCT, and achieves high coding efficiency by efficiently removing inter-frame correlation and frame correlation. On the other hand, JPEG2000 is a technique that uses wavelet exchange and embedded entropy coding called EBCOT, and does not use inter-frame correlation. It has various effective functions such as SNR scalability. MotionJPEG2000, which can be applied to moving images, has also been proposed and has the same functions as JPEG2000.

The scalability used in JPEG2000 is called an embedded type, and the encoder only encodes once, and there is no need to regenerate compressed data according to individual resolutions. Decoded images of various resolutions and SNRs can be obtained from a single compressed file, leading to reduction in file capacity and reduction in calculation amount. FIG. 21 shows the resolution scalability function of JPEG2000. When the resolution of the original image is K × L, the decoder can restore an image having a resolution of K / 2 ⁿ × L / 2 ⁿ . However, normally, the resolution of the image required on the decoding side is not limited to 1/2 ⁿ times the size of the original image. Therefore, an embedded coding method that can be restored at a resolution of a more general rational number (N / M times the original image, where M and N are positive integers) has been studied.

  As an example of this method, the inventors of the present invention can decode with an M-band division filter bank, an N-band synthesis file bank, and an embedded entropy coding method called EBCOT with a rational multiple of spatial resolution. Encoding methods and decoding methods have also been proposed. However, there is no compatibility with the international standard JPEG2000, and it is necessary to newly install an encoder and a decoder.

The JPEG2000 (PartI) compatible model is compatible with JPEG2000 in both encoder and decoder, and can decode a spatial resolution image (rational number times (N / ^2D times)) (for example, see Non-Patent Document 3). The system configuration is shown in Fig. 22. In this method, after obtaining a 1/2 ^U multi-resolution image by JPEG2000, a decimation filter (N / 2 ^U by filtering) is output to the resolution image output immediately above the required resolution. In this method, a resolution image of 1/2 ^U (where 1/2 ^U > N / 2 ^D ) is decoded once by a JPEG2000 decoder, so that N / 2 ^D times becomes to transmitting unnecessary high-frequency components in resolution image, is not preferable from the viewpoint of coding efficiency. for example, when the ^{N / 2 D = 5/8} , once JPEG2000 decoder Thus from decoding the same resolution image as the original image, it is necessary to transmit the data of the entire band.
"MPEG" edited by the Video Media Society, Ohmsha, April 1996 ISO / IEC1544-1 JPEG2000 Part I: Core coding system, 200-12-15 Tanabe Shu, Hiroshi Watanabe, Hideyoshi Tominaga, “Examination of Optimal Resolution Conversion in Motion JPEG2000”, Audio Visual Decoding Information Processing 40-4, (2003, 3, 7)

The present invention proposes a decoding method capable of decoding at a resolution of N / ^2D times which is excellent in encoding efficiency while maintaining compatibility with JPEG2000 (Part II) for both the encoder and the decoder. JPEG2000 (PartI) compatible model (Tanabe Shu, Hiroshi Watanabe, Hideyoshi Tominaga, “Examination of Optimal Resolution Conversion Method in Motion JPEG2000”, Audio Visual Decoding Information Processing 40-4, (2003, 3, 7)) 1/2 ^{U (here, 1/2 U> N /} 2 D) by the decoder from decoding the resolution images, will be N / ^{2 D} times unwanted high frequency component resolution images also transmitted, This is not preferable from the viewpoint of encoding efficiency.

The present invention has been made in view of the above points, and a resolution scalable decoding method capable of realizing N / ^2D times resolution conversion by transmitting only frequency components necessary for N / ^2D times resolution images. And an apparatus.

  FIG. 1 is a diagram for explaining the principle of the present invention.

The present invention (Claim 1) is a resolution scalable decoding method in which an original image is equally divided into ^2D bands by a wavelet filter, and encoded data is decoded.
Discard the N + 1 band or higher signal of the compressed data encoded by the encoder,
2 ^U band (U is the smallest integer satisfying N <2 ^U ) is synthesized from the low band for signals other than discarded,
N / 2 ^U times resolution conversion is performed, N / 2 ^D times resolution image is generated,
Decimation by a zero-order hold method with an enlargement ratio of N times and a low-frequency wavelet filter is connected in U stages.

The present invention (Claim 2) is a resolution scalable decoding method in which an original image is equally divided into ^2D bands by a wavelet filter, and encoded data is decoded.
Discard the signal of N + 1 band or more of the compressed data encoded by the encoding device,
2 ^U band (U is the smallest integer satisfying N <2 ^U ) is synthesized from the low band for signals other than discarded,
N / 2 ^U times resolution conversion is performed, N / 2 ^D times resolution image is generated,
1: N upsampling, LPF band limiting to (1/2 ^U ) and 2 ^U : 1 downsampling are performed.

  FIG. 2 is a principle configuration diagram of the present invention.

The present invention (Claim 3) is a resolution scalable decoding device that decodes encoded data by equally dividing an original image into ^2D bands by a wavelet filter,
When the compressed data encoded by the encoding means is input, a signal discarding means for discarding a signal of N + 1 band or more,
Band synthesis means for synthesizing 2 ^U bands (U is the smallest integer satisfying N <2 ^U ) from a low frequency band for signals other than those discarded by the signal discarding means;
Resolution conversion means for performing N / 2 ^U times resolution conversion and generating an N / 2 ^D times resolution image after being synthesized by the band synthesizing means,
Decimation by a zero-order hold method with an enlargement ratio of N times and a low-frequency wavelet filter is connected in U stages.

The present invention (Claim 4) is a resolution scalable decoding device that decodes encoded data by equally dividing an original image into ^2D bands by a wavelet filter,
When the compressed data encoded by the encoding means is input, a signal discarding means for discarding a signal of N + 1 band or more,
Band synthesis means for synthesizing 2 ^U bands (U is the smallest integer satisfying N <2 ^U ) from a low frequency band for signals other than those discarded by the signal discarding means;
A resolution converting means for performing N / 2 ^U times resolution conversion after synthesizing by the band synthesizing means, and generating an N / 2 ^D times resolution image;
It has an LPF that limits the bandwidth to 1: N and (1/2 ^U ), and means for performing 2 ^U : 1 downsampling.

  According to the present invention, it is possible to efficiently encode an image and to store it with a small disk capacity. Since it has spatial resolution scalability, it is possible to decode an image with a spatial resolution according to the performance and application of the image display device. When decoding from a low band to an arbitrary band, an image having a spatial resolution lower than that of the original image can be reproduced. When all data is decoded, an image having the same resolution as the original image is reproduced. When it is desired to reproduce an image having a spatial resolution lower than that of the original image according to the performance and application of the image display device, only decoding of the encoded data corresponding to the necessary band is required. The processing time is shorter than when the resolution conversion is performed by reproducing an image having the same resolution as the original image, and only the necessary data needs to be transmitted when transmitting the encoded bit stream, and the transmission rate is also reduced.

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

FIG. 3 shows a resolution scalability function capable of rational number conversion in an embodiment of the present invention. The encoder 100 encodes an image with a resolution of K × L and generates a single compressed data. The decoder 200 extracts a bit stream of a part of the compressed data necessary for N / ^2D double resolution conversion, so that the resolution is a rational multiple of the original image, that is, N / 2 ^D times (however, N, D Is a positive integer).

[First embodiment]
In this embodiment, an encoder / decoder capable of decoding at an arbitrary rational number multiple spatial resolution will be described.

  FIG. 4 shows a basic configuration of the encoder according to the first embodiment of the present invention. The encoder 100 shown in the figure includes an equal band dividing unit 110, a quantizing unit 111, and an embedded entropy encoder EBCOT 112 that enables scalable decoding.

The input original image is divided into ^2D equal bands by the equal band dividing unit 110. For the division, a wavelet filter (eg, wavelet 97 irreversible filter, wavelet 53 reversible filter) defined in JPEG2000 (Part I and Part II) is used. In JPEG2000PartI, wavelet transform is used as band division. In the wavelet transform, band division is performed by a division method called Mallat division shown in FIG. In Mallat division, an input signal can be divided into a plurality of bands by successively dividing in a low frequency direction using a one-dimensional two-division filter. By dividing D times, it is possible to divide an image having a spatial resolution of 1 / ^2D with respect to the original image. This process is performed in the horizontal direction and the vertical direction, respectively. On the decoder 200 side, decoding is performed sequentially from the low frequency side. By decoding ^U times, an image having a spatial resolution of 2 ^U / 2 ^D = 1/2 ⁿ times the original image can be restored.

On the other hand, in JPEG2000 (Part II), an arbitrary band division method can be selected. In the present invention, as shown in FIG. 6, the wavelet filter is applied recursively not only to the low frequency side but also to the high frequency side to divide the image into equal bands. Each has a bandwidth of approximately 2π / ^2D and divides the original image into equally spaced bands. For comparison, FIG. 7 shows an example of image division by Mallat division and equal band division.

  The divided signal is quantized by the quantizing unit 111 and compressed data is generated by the embedded entropy encoder EBCOT 112. The embedded entropy encoder is an entropy encoding method that realizes a function that allows the decoder 200 to freely set the spatial resolution and SNR of a decoded image by simply extracting a part of the compressed data. Note that by using the entropy encoding EBCOT 112, it is possible to control the amount of compressed data without performing quantization, and thus quantization can be omitted.

  FIG. 8 shows a basic configuration of the decoder according to the first embodiment of the present invention.

  The decoder 200 is a scalable decoder capable of decoding with an arbitrary rational multiple of spatial resolution.

  The decoder 200 shown in the figure includes a bit stream extraction unit 220, an embedded entropy EBCOT decoding unit 221, an inverse quantization unit 222, an N-band synthesis unit 223, and a resolution conversion unit 224.

The bit stream extraction unit 220 extracts only data necessary for decoding the N / 2D times spatial resolution from the input compressed data, and the embedded entropy EBCOT decoding unit 221 performs entropy decoding. The EBCOT decoding unit (decoder) decodes the data compressed by the embedded entropy EBCOT encoder 112. Entropy that can freely set the spatial resolution and SNR of the decoded image by extracting only a part of the compressed data. Decryption method is used. When the spatial resolution of the decoded image is N / ^2D times that of the original image, only the data for N bands from the low frequency band side of the equal-band divided filter bank is EBCOT decoded. From the decoded data, the inverse quantization unit 222 decodes signals for N bands from the low frequency band side.

  FIG. 9 illustrates the relationship between the band dividing unit and the band synthesizing unit when the number of divisions is two levels.

In this case, 2 ^D = 4 and N = 3, and the synthesis filter bank (N band synthesis unit 223) synthesizes the three bands on the low frequency side. At this time, the resolution of the synthesized filter bank output image is the same size as the original image, but the spectrum of the original image and the synthesized filter bank output is as shown in FIG. 10, and the synthesized filter bank output signal is in the high frequency band. The signal will be discarded. In the example, although the resolution of the synthesized filter bank output image is the same size as the original image, the data to be transmitted may be three bands on the low frequency side, which is desirable from the viewpoint of coding efficiency. On the other hand, when ^2D = 4 and N = 3, the JPEG2000 (PartI) compatible model (Tanabe Shu, Hiroshi Watanabe, Hideyoshi Tominaga, “Examination of Optimal Resolution Conversion Method in Motion JPEG2000”, Audio Visual Decoding Information Processing 40 −4, (2003, 3, 7)), it is necessary to transmit signals of all four bands, and extra data will be transmitted. In the synthesis filter bank, it is not always necessary to decode up to the same resolution as the original image, and only U stages satisfying the following conditions may be decoded.

Smallest integer value U satisfying U <2 ^U:
For example, when N = 1, 2, U = 1
When N = 3, 4, U = 2
When N = 5-8, U = 3
It can be seen that the number of stages of the synthesis filter bank does not depend on the number of divisions of the encoder 200 (that is, 2 ^D ) and is determined only by N.

  Finally, the output image of the N band synthesizing unit 223 is processed in the resolution converting unit 224, and an N / M times resolution image is obtained. Various methods proposed so far can be used for resolution conversion. Note that the output image of the N-band synthesizing unit 223 is a signal in which a high-frequency component unnecessary for the N / M resolution image is discarded, so that the influence of aliasing that occurs during resolution conversion can be suppressed as much as possible.

  In a third embodiment to be described later, as a resolution conversion method, an example is shown in which a zero-order hold method with an enlargement ratio of N times and decimation by a low-frequency wavelet filter defined in JPEG2000 are connected in multiple stages.

Since the compressed data is compatible with JPEG2000 (Part II), it can be decoded by the JPEG 2000 Part II decoder shown in the lower part of FIG. However, in that case, the resolution of the image to be decoded is limited to ^2U / ^2D .

[Second Embodiment]
In the present embodiment, a method for realizing N / 2 ^U times resolution conversion will be described.

FIG. 11 is a diagram showing (N / 2 ^U ) double resolution conversion by multi-stage connection of a zero-order hold method and a wavelet filter in the second embodiment of the present invention, and decimation is performed by the zero-order hold method and the wavelet filter. This is an example of 2 ^U multistage connection.

The feature of this method is that a resolution conversion of N / 2 ^U times can be realized by using only the wavelet filter defined by JPEG2000, and it is not necessary to prepare a new decimation filter.

The zero-order hold method is a method of enlarging a signal N times, and is a simple method that can be realized by arranging original signals N times as shown in FIG. This zero-order hold method uses N-times upsampling and a filter H (z) = 1 + z ⁻¹ +... + Z ^{− (n−1)}
Can be interpreted as filtering by Upsampling is an operation of inserting zero-value samples between adjacent signals. In N-times upsampling, the resolution of the original signal is increased to N-times by inserting N−1 zero values. The amplitude characteristic of H (z) shows a gentle low-pass characteristic as shown on the right side of FIG. Therefore, the zero-order hold can be understood as an image enlargement method using a gentle low-pass filter.

  FIG. 13 shows a spectrum characteristic diagram by the zero-order hold method. When a signal having the spectral characteristics shown in the right part of FIG. 10 is input, first, the band is expanded three times by N-times upsampling, and an imaging component is generated in the high band. Since this corresponds to applying a gentle low-pass filter H (z) to the signal, the output has a form in which the high-frequency signal is suppressed.

Next, decimation processing by multistage connection of wavelet filters is performed. First, a low-frequency wavelet filter is applied, and then 2: 1 downsampling is performed. The signal band is halved by the low-frequency wavelet filter, and the resolution is halved by downsampling. At the same time, aliasing components are generated (see FIG. 14). It is possible to reduce the resolution to 1/2 ^U by U-stage cascade connection of 2: 1 decimation processing using this wavelet filter.

[Third Embodiment]
In the present embodiment, as a method for expressing N / 2 ^U times resolution, a method using an LPF that limits the bandwidth to 1/2 ^U times will be described.

FIG. 15 shows a resolution conversion method by LPF processing with 1/2 ^U band limitation in the third embodiment of the present invention.

This method uses an upsampling of 1: N, an LPF that is band limited to (1/2 ^U ), and 2 ^U : 1 downsampling. First, the signal is expanded N times by 1: N upsampling, and then LPF is applied to limit the signal band to 1/2 ^U. Thereafter, 2 ^U : 1 downsampling is performed, and the resolution is reduced to 1/2 ^U.

FIG. 16 shows spectral characteristics obtained by this method. When a signal having a spectral characteristic on the right side in FIG. 10 is input, first, the band is expanded three times by N-times upsampling, and a folded component is generated in the high band. Thereafter, ² U performs LPF that band-limits to ^{1/2 U} fold: by performing 1 down-sampling is bandwidth ^{1/2 U,} it is possible to obtain a reduced image in ^{1/2 U} times. However, since there is no ideal LPF that limits the bandwidth to 1/2 ^U times, the influence of aliasing due to downsampling cannot be avoided.

[Fourth Embodiment]
In the present embodiment, resolution that can be decoded when encoded will be described.

  As an example, a resolution that can be decoded when a 1920 × 1080 HDTV image is encoded using the present invention will be described.

For comparison, FIG. 17 shows an example of encoding with JPEG2000. When the number of division levels is 3, the decodable resolution is 1/2 ⁿ (n = 1, 2, 4) of the original image in the vertical and horizontal directions.
・ 240 × 135
・ 480 × 270
・ 960 × 540
・ 1902 × 1080
It becomes a pixel. For this reason, widely distributed SDTV images cannot be decoded.

On the other hand, FIG. 18 shows the resolution that can be decoded when a 1920 × 1080 HDTV image is encoded according to the present invention. The number of divisions was set to 3 so that the minimum decodable resolution was the same as that of JPEG2000. At this time, the minimum decodable resolution is 240 × 135 pixels, and generally 240 m × 135 n (where n and m are positive integers, n = 1, 2,..., 8, m = 1, 2). ,... 8) A pixel image can be decoded. In this case, it is widely distributed. 720 x 405 (SDTV, black on top and bottom)
・ 1920 × 1080 (HDTV)
Can be easily decoded from a single compressed file.

  The effectiveness of the resolution scalable encoding / decoding described above will be verified below.

In the following simulation, the target image has an image size of 1920 × 1080 (pixels),
Digital cinema text images with 8 bits (pixels / pixel) (dance, lake, party, nobility), “T.Nakachi, T. Sawabe, T. Fujii and T. Fujii,” Multiresolution lossless video codiong using inter / Intra frame adaptive prediction, “IEICE Trans. on Fundamentals, vol. E85-A, no. 8, pp. 1822-1830, Aug. 2002.” and the circular zone plate shown in FIG. 19 were used. The circular zone plate is calculated by the following equation, and the influence of frequency components from the low range to the high range can be examined.

The reference image was created using DCT as shown in FIG. First, a K × L DCT operation is performed on an original image of K × L pixels. Next, an inverse DCT operation of K (N / 2D) × L (N / 2D) is performed on the component of K (N / 2D) × L (N / 2D) from the low range of the obtained DCT coefficient, The result of the brightness adjustment is a reference image to be obtained.

  Table 1 shows PSNR [dB] when N / 2D = 3/4 times resolution conversion is performed. The number of division levels is D = 2, and the analysis filter bank uses (5, 3) reversible filters, and N-times enlargement methods are 1) N-times upsampling, 2) zero-order hold method, and 3) linear interpolation method, respectively. This shows a case where (a) (9, 7) irreversible LPF and (b) (5, 3) reversible LPF are used in 1 / 2D reduction.

From Table 1, it can be seen that N times upsampling is the worst and the linear interpolation method gives the best characteristics when comparing N times upsampling, zero-order hold method, and linear interpolation method. This is because N-fold upsampling has a strong influence of the high-frequency aliasing distortion component, and the zero-order hold method and the linear interpolation method effectively suppress the high-frequency aliasing distortion of the DC component. It is thought that. In addition, when (9,7) irreversible LPF and (5,3) reversible LPF are compared, (9,7) irreversible LPF gives good characteristics, especially in the case of N-times upsampling. Is remarkable. From this, it is considered that the (5, 3) reversible LPF cannot sufficiently suppress the high-frequency aliasing distortion component.

  Table 2 shows PSNR [dB] when N / 2D = 5/8 times resolution conversion is performed.

The number of division levels is D = 3, and the analysis filter bank is a (5, 3) reversible filter. The general tendency is the same as in the case of N / 2d = 3/4 times, except that the zero-order hold method gives better characteristics than the linear interpolation method. This is considered to be the result of discarding more high-frequency signal components during synthesis filter bank synthesis. Under this condition, the high-frequency cutoff characteristic of the low-pass filter is not so important, and it can be said that the characteristic of the zero-order hold method that can hold more low-frequency signal components gives a good characteristic as a result.

  For example, as a method of enlarging N times, in addition to the zero hold method and the upsampling method described in the above embodiment, other image enlargement methods (for example, linear interpolation method, nearest neighbor method, bilinear method, Bicubic method) can be applied.

  The present invention is not limited to the above-described embodiments and examples, and various modifications and applications are possible within the scope of the claims.

  The present invention is applicable to a moving picture decoding technique.

It is a figure for demonstrating the principle of this invention. It is a principle block diagram of this invention. It is a figure which shows the resolution scalability function in which conversion of the rational number multiple in one embodiment of this invention is possible. 1 is a basic configuration diagram of an encoding device according to a first embodiment of the present invention. 5 is an example of Mallat division in the first exemplary embodiment of the present invention. 2 is an example of equal band division using a wavelet filter according to the first embodiment of the present invention. FIG. 7 is a diagram showing comparison of equal band division in the first embodiment of the present invention. 1 is a basic configuration diagram of an encoding device according to a first embodiment of the present invention. FIG. 5 is a diagram (in the case of two stages) showing a relationship between an equal band dividing unit and an N band combining unit in the first embodiment of the present invention. It is a figure which shows the spectrum of the original image and N band synthetic | combination part output in the 1st Embodiment of this invention (when ^2D = 4 and N = 3). It is a figure which shows the (N / ^2U ) double resolution conversion by the multistage connection of the zero-order hold method and the wavelet filter in the 2nd Embodiment of this invention. It is a figure for demonstrating the zero order hold method in the 2nd Embodiment of this invention. It is a figure ( ^2D = 4, N = 3) which shows the spectrum characteristic of the zero order hold method in the 2nd Embodiment of this invention. It is the spectrum ( ^2D = 4, N = 3) of the wavelet multistage connection in the 2nd Embodiment of this invention. It is a figure which shows the resolution conversion method by the (1 / ^2U ) band-limited LPF process in the 3rd Embodiment of this invention. It is a spectrum characteristic ( ^2D = 4, N = 3) of the (1 / ^2U ) band-limited LPF process in the 3rd Embodiment of this invention. This is a resolution that can be decoded when an HDTV image in the case of JPEG2000 is encoded. It is the resolution level which can be decoded when the HDTV image in the 4th Embodiment of this invention is encoded. This is a circular zone plate used for the simulation. It is a figure which shows the production | generation of the reference image in simulation. JPEG2000 resolution scalability. 1 is a system configuration diagram of JPEG2000 (PartI).

Explanation of symbols

100 Encoder 110 Equal Band Division Unit 111 Quantization Unit 112 EBCOT
200 Decoder, Decoder 220 Bitstream Extractor 221 EBCOT Decoder, Signal Discarding Unit 222 Inverse Quantizer 223 N-band Synthesizer, Band Synthesizer 224 Resolution Converter, Resolution Converter 225 EBCOT Decoder 226 Inverse Quantization 227 Band synthesis unit

Claims (4)

In a resolution scalable decoding method, in which an original image is equally divided into ^2D bands by a wavelet filter and encoded data is decoded,
Discard the N + 1 band or higher signal of the compressed data encoded by the encoder,
2 ^U band (U is the smallest integer satisfying N <2 ^U ) is synthesized from the low band for signals other than discarded,
N / 2 ^U times resolution conversion is performed, N / 2 ^D times resolution image is generated,
A resolution scalable decoding method, characterized in that decimation by a zero-order hold method with an enlargement factor of N times and a low-frequency wavelet filter are connected in U stages. In a resolution scalable decoding method, in which an original image is equally divided into ^2D bands by a wavelet filter and encoded data is decoded,
Discard the signal of N + 1 band or more of the compressed data encoded by the encoding device,
2 ^U band (U is the smallest integer satisfying N <2 ^U ) is synthesized from the low band for signals other than discarded,
N / 2 ^U times resolution conversion is performed, N / 2 ^D times resolution image is generated,
A resolution scalable decoding method characterized by performing 1: N upsampling, LPF band-limiting to (1/2 ^U ) and 2 ^U : 1 downsampling. A resolution scalable decoding device for equally dividing an original image into ^2D bands by a wavelet filter and decoding encoded data,
When the compressed data encoded by the encoding means is input, a signal discarding means for discarding a signal of N + 1 band or more,
Band synthesis means for synthesizing 2 ^U bands (U is the smallest integer satisfying N <2 ^U ) from a low frequency band for signals other than those discarded by the signal discarding means;
Resolution conversion means for performing N / 2 ^U times resolution conversion and generating an N / 2 ^D times resolution image after being synthesized by the band synthesizing means,
A resolution scalable decoding apparatus characterized in that decimation by a zero-order hold method with an enlargement ratio of N times and a low-frequency wavelet filter are connected in U stages. A resolution scalable decoding device for equally dividing an original image into ^2D bands by a wavelet filter and decoding encoded data,
When the compressed data encoded by the encoding means is input, a signal discarding means for discarding a signal of N + 1 band or more,
Band synthesis means for synthesizing 2 ^U bands (U is the smallest integer satisfying N <2 ^U ) from a low frequency band for signals other than those discarded by the signal discarding means;
A resolution converting means for performing N / 2 ^U times resolution conversion after synthesizing by the band synthesizing means, and generating an N / 2 ^D times resolution image;
1. A resolution scalable decoding apparatus comprising: an LPF for band-limiting to 1: N and (1/2 ^U ) and means for performing 2 ^U : 1 down-sampling.