JPH0750835A - Picture coder - Google Patents

Abstract

PURPOSE:To realize coding with an excellent compression efficiency by devising the coder such that distortion is not protruded up to the outside of a block even when the distortion is produced in the block. CONSTITUTION:A sub band divider 11 applies sub band division to an original picture into a low frequency picture and plural high frequency pictures, an orthogonal transformation device 12 divides a low frequency picture into blocks of a 1st size to subject to an orthogonal transformation. The block dividor 13 divides a high frequency picture into blocks of a 1st size. The block of the 1st size subject to orthogonal transformation and a block of the 1st size of the high frequency picture are synthesized to form a block of a 2nd size to make hybrid transformation. A frame memory 21 rearranges hybrid transformation coefficient to generate a low frequency picture and some high frequency picture, an inverse orthogonal transformation device 22 divides the low frequency picture into blocks of the 1st size, applies inverse orthogonal transformation to the blocks, and a sub band synthesizer 23 synthesizes the entire picture through sub bands to apply inverse hybrid transformation.

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, and more particularly to a motion-compensated prediction code which is a sub-block division of an image, an orthogonal transformation of only a low frequency band, and an overlapping block transformation realized by rearrangement. The present invention relates to an image encoding device adapted to be applied to encoding. For example, it is applied to high-efficiency encoding of image data in the field of digital image processing.

[0002]

2. Description of the Related Art Conventionally, there has been proposed a method of compressing a digital moving image by using techniques such as interframe prediction, orthogonal transformation, quantization, variable length coding and the like. In recent years, in particular, as a new conversion, a conversion whose conversion base spans blocks (overlapping block conversion) has been actively studied. Attempts have also been made to apply this to image coding. For example, (1) "Image Coding Using Motion Compensation Filter Bang Structure" (Norazawa et al., 1992, Image Coding Symposium)
8-5, pp.277-279) is an application of overlapped block transform to motion-compensated interframe predictive coding. (2) "A Study on Image Coding by Hybrid Transform"
(Katada Soga 2 people, 1992, Image Coding Symposium 6-
4, pp.217-220) is an example of DCT (discrete cosine transform) and overlapped block transform LW.
A hybrid transform in which T (overlapping wavelet transform) is combined is applied to intraframe coding.

[0003]

As described above, in the conventional method (1), since conversion using subband division is performed as overlapping block conversion, as shown in FIGS. 3 (a) to 3 (d),
The low-frequency transform base largely extends outside the block. The broken line indicates the boundary of the block of interest. For this reason,
When a large distortion occurs in a certain block in the image, the distortion extends beyond the block. Such distortion causes a large visual deterioration. Further, in the conventional method (2), the conversion bases are as shown in FIGS. 4A to 4D, and the low-frequency conversion bases are less likely to be outside the block. The broken lines indicate the boundaries of the blocks of interest, as in FIG. However, the intra-frame coding has a problem that sufficient compression efficiency cannot be obtained.

The present invention has been made in view of the above circumstances, and an image coding apparatus having a high compression efficiency that does not extend beyond a block even if a block is distorted. It is intended to be provided.

[0005]

In order to achieve the above-mentioned object, the present invention sub-bands an original image into a low-frequency image and a plurality of high-frequency images, and the low-frequency image is divided into first and second sub-band images. The high-frequency image is divided into blocks each having a first size, the high-frequency image is divided into blocks each having a first size, and the orthogonal-transformed block having the first size and the high-frequency image are first divided. Hybrid conversion means for performing a hybrid conversion by synthesizing blocks of a size 1 to form a block of a second size, and rearranging the hybrid conversion coefficients to create a low-frequency image and a plurality of high-frequency images Then, the low-frequency image is divided into blocks of a first size, subjected to inverse orthogonal transform, and the entire image is subjected to sub-band synthesis to perform an inverse hybrid transform, and a hybrid transform coefficient is coded. To do Is obtained is characterized by comprising a video encoding means.

[0006]

The hybrid transform means for performing the hybrid transform transforms an image by a transform base with less overlap between blocks to obtain hybrid transform coefficients, and the inverse hybrid transform means for performing the inverse hybrid transform is a transform with less overlap between blocks. The restored image is obtained by inversely transforming the image according to the basis, and the moving image coding means performs motion compensation predictive coding using the restored image as a reference image.

[0007]

Embodiments will be described below with reference to the drawings. FIG. 1 is a block diagram for explaining an embodiment of an image coding apparatus according to the present invention, in which 1 is a hybrid converter, 2 is a hybrid inverse converter, 11 is a subband divider, and 12 is orthogonal. Converter, 13 is a block divider, 1
4, 19 is an adder, 15 is a mode selector, 16 and 20 are switches, 17 is a quantizer, 18 is an inverse quantizer, 21 is a frame memory, 22 is an inverse orthogonal transformer, and 23 is a subband synthesizer. , 24 is a motion compensation predictor, and 25 is a motion vector detector.

The hybrid converter 1 comprises a sub-band divider 11, an orthogonal transformer 12, and a block divider 13. The hybrid inverse converter 2 includes a frame memory 2
1 and an inverse orthogonal transformer 22 and a subband synthesizer 23. The sub-band divider 11 sub-bands the original image into a low-pass image and a plurality of high-pass images, and the orthogonal transformer 12 divides the low-pass image into blocks of a first size. And perform orthogonal transformation. The block divider 13 divides the plurality of high frequency images into blocks having a first size.
Hybrid conversion is performed by synthesizing the block of the first size subjected to the orthogonal transform and the block of the first size of the high-frequency image to form the block of the second size.

The frame memory 21 rearranges the hybrid transform coefficients to create a low-frequency image and a plurality of high-frequency images, and an inverse orthogonal transformer 22 converts the low-frequency image into blocks of a first size. Divide and perform inverse orthogonal transform. The subband synthesizer 23 synthesizes the entire image with subbands. In this way, inverse hybrid transformation is performed to create a restored image, and the restored image is used for motion compensation prediction. The motion compensation predictor 24 is a moving picture prediction unit for predicting a hybrid transform coefficient.

FIGS. 2A and 2B are diagrams for explaining the hybrid conversion. Here, an example is shown in which the image size is 24 pixels × 24 pixels, and this is subjected to 8 × 8 hybrid conversion. In the figure, the image is first divided into a low-frequency image (LL) of 12 pixels × 12 pixels and three high-frequency images (LH, HL, HH) of 12 pixels × 12 pixels by a subband filter. Then each image is 4
It is divided into blocks of pixels × 4 pixels, and each block of the low-frequency image (LL) is two-dimensionally orthogonally transformed. As the orthogonal transform, DCT, discrete Fourier transform, Hadamard transform, KL transform or the like is used. Finally, the low-frequency image (L
L) and high-frequency images (LH, HL, HH) respectively 4
Pixel x 4 pixels are collected to obtain a block of hybrid transform coefficients consisting of 8 pixels x 8 pixels.

In addition to the above-described hybrid conversion, the "coding method" of Japanese Patent Application No. 4-68749 may be used. That is, the one-dimensional signal is multi-resolution divided, the low-frequency component of the divided signal is divided into blocks of size n / 2, and the conversion basis of the low-frequency component is kept at n / 2 points so that it remains within the block. Encoding that performs conversion and collects n / 2 points of wide-area components of multi-resolution division that belong to the same position with the conversion coefficient of n / 2 points and the input signal to form an n block in size and perform n-point hybrid conversion Is the way. In this case, the hybrid transform is configured by a one-dimensional hybrid transform in the horizontal direction and a one-dimensional hybrid transform in the vertical direction.

Next, the operation of the encoder will be described with reference to FIG. First, the image to be encoded is subband-divided by the subband divider 11. The orthogonal transform unit 12 orthogonally transforms the low-frequency image of the subband-divided image. At this time, as the orthogonal transform, the transform base is within D.
CT or the like is used. The block divider 13 constitutes a block of hybrid transform coefficients. These functions are
This is as described with reference to FIG. The adder 14 takes the difference between the hybrid transform coefficient of the original image output from the block divider 13 and the hybrid transform coefficient of the reference image output from the motion compensation predictor 24. The mode selector 15 compares the hybrid transform coefficient output from the block divider 13 with the difference output from the adder 14, selects which one to encode, and switches the switches 16 and 20. . That is, when the differential signal is selected, the switches 16 and 20 are switched downward, and when not, the switches 16 and 20 are switched upward.

The quantizer 17 quantizes the signal from the switch 16, and the dequantizer 18 dequantizes the quantized signal. The adder 19 adds the signal output from the dequantizer 18 and the signal from the switch 20. The frame memory 21 stores the restored hybrid transform coefficients output from the adder 19 while rearranging them into a low-frequency image (LL) and a high-frequency image (LH, HL, HH). The inverse orthogonal transformer 22 inversely transforms the low-frequency image (LL) portion of the signal stored in the frame memory of the frame memory 21. The subband synthesizer 23 subband synthesizes the output of the inverse orthogonal transformer 22 to create a reference image. The functions of the frame memory 21, the inverse orthogonal transformer 22, and the subband synthesizer 23 are exactly the reverse of the hybrid transform process shown in FIG.

The motion vector detector 25 detects a motion vector from the original image and the reference image output from the subband synthesizer 23. The motion compensation predictor 24 uses the motion vector detected by the motion vector detector 25 and the subband synthesizer 2
A block for prediction is cut out on the basis of the reference image output from 3 and this is hybrid-converted and supplied to the adder 14. The encoded data is mode selection information output from the mode selector 15, a quantization index output from the quantizer 17, and motion vector information output from the motion vector detector 25. These are variable length coded by an entropy code or the like, further error correction coded, and recorded or transmitted.

Although the embodiments have been described above, the present invention can be applied to any moving picture coding other than the above.
For example, an inter-frame differential coding method that does not use motion-compensated prediction and a known document "A Study on MDCT Video Coding Using Overlap Block Motion Compensation" (1991 Image Coding Symposium 7-4 pp.177-178) It is also applicable to a coding method using overlap motion compensation prediction as described in (1). The coding method using the overlap motion compensation prediction is a coding method that reduces block distortion, which is a problem in image coding using the DCT, and includes motion compensation for overlapping blocks and MDCT (Modified DC).
This is a moving image coding system that combines T) and T).

[0016]

As is apparent from the above description, according to the present invention, since moving picture coding is performed using hybrid conversion in which blocks of low-frequency conversion bases have little overlap, distortion occurs in a certain block. Even if the distortion is not spread to the adjacent blocks, it is possible to realize the coding that can obtain a visually good restored image. Also, since inter-frame coding is performed,
Encoding with high compression efficiency can be realized.

[Brief description of drawings]

FIG. 1 is a configuration diagram for explaining an embodiment of an image encoding device according to the present invention.

FIG. 2 is a diagram for explaining a two-dimensional hybrid conversion in the present invention.

FIG. 3 is a diagram showing a low-frequency transform base of a conventional one-dimensional wavelet transform.

FIG. 4 is a diagram showing a low-frequency conversion base of a conventional hybrid conversion.

[Explanation of symbols]

1 ... Hybrid converter, 2 ... Hybrid inverse converter,
11 ... Subband divider, 12 ... Orthogonal transformer, 13 ... Block divider, 14, 19 ... Adder, 15 ... Mode selector, 16, 20 ... Switch, 17 ... Quantizer, 18 ... Inverse quantizer , 21 ... Frame memory, 22 ... Inverse orthogonal transformer, 23 ... Subband synthesizer, 24 ... Motion compensation predictor,
25 ... Motion vector detector.

Claims (1)

[Claims] 1. An original image is sub-band divided into a low-frequency image and a plurality of high-frequency images, and the low-frequency image is divided into blocks of a first size and orthogonally transformed to obtain high-frequency images. Each image is divided into blocks of a first size, and the orthogonally transformed block of the first size and the block of the first size of the high-frequency image are combined to create a block of the second size. A hybrid conversion unit that performs a hybrid conversion by configuring a block and a hybrid conversion coefficient are rearranged to create a low-frequency image and a plurality of high-frequency images, and the low-frequency image is converted into a block of a first size. It is characterized by comprising an inverse hybrid transform means for performing an inverse hybrid transform by dividing and performing an inverse orthogonal transform and sub-band combining the entire image, and a moving image coding means for encoding a hybrid transform coefficient. Image code Device.