JPH0813141B2 - Still image coding method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Outline] The present invention is an encoding method in which a still image is hierarchically quantized and transmitted to gradually reproduce a fine screen. When or after vector quantization of the interpolation error signal with the prediction playback screen, the edge pixel of the image is extracted from the playback screen up to the previous stage, and only the edge pixel is scalar-quantized to the scalar quantization error signal. By doing so, pixels having a large amplitude due to edge pixels are eliminated and vector quantization matching is made more optimal.

[Industrial applications]

The present invention relates to a still image encoding method, and particularly to sub-sampling an original image, quantizing it, and transmitting it after variable-length encoding, predicting a reproduction surface from the next-stage screen, and sub-sample interpolation error with the next-stage screen. The present invention relates to a still image encoding method in which a signal is quantized, variable encoded, and transmitted for a predetermined number of steps.

When using image transmission for image database search and still image transmission in video conference, etc., thinned out, that is,
If hierarchical transmission is possible so that high resolution images can be gradually reproduced from subsampled low resolution images,
In the image database search, unnecessary images can be discarded at an early stage, so required images can be searched at high speed, and still image transmission allows an overview of the images to be seen at an early stage. , Such hierarchical coding is necessary.

[Conventional technology]

FIG. 8 shows a schematic configuration of a conventional still image coding system, in which the upper half of the figure shows the transmitting side and the lower half shows the receiving side.

In FIG. 8, 11 and 14 are subsamplers on the transmission side,
19 is a subsampler on the receiving side, 12 and 21 are scalar quantizers on the transmitting side and the receiving side, 13 and 17 are entropy coders (variable length encoders) and entropy decoders (variable length) on the transmitting side and the receiving side, respectively. Decoder), 15
Reference numerals 20 and 20 denote frame memories for storing playback screens on the transmitting side and the receiving side, respectively, and 16 and 18 denote interpolators for reproducing the playback screens on the transmitting side and the receiving side, respectively.
The subsampler 19 and the frame memory 20 form a predictor.

FIG. 9 is for explaining the operation of the sub-sampling performed in FIG. 8. In the figure, (1) to (6) respectively represent the entire image at the first to sixth transmission stages of the image. Part of the block (8 × 8 pixels) is shown, ○ indicates that the pixel is newly transmitted (pixel at the position to be subsampled), and ● (hatched black circle) indicates that the pixel has already been transmitted. .

Further, FIG. 10 is a diagram for explaining an interpolation operation for screen reproduction.

Next, the operation of this conventional example will be described. First, in the screen transmitted in the first stage, the still original image is subsampled by the subsampler 11 on the transmission side in FIG. 8 as shown in FIG. 9 (1). , Quantized by the scalar quantizer 12, variable-length coded by the entropy coder 13, and sent to the transmission line L. In this second stage, no signal is generated from the subsampler 19 of the predictor.

The receiving side of the data from the transmission line L performs the reverse operation to the transmitting side. That is, it is decoded by the entropy decoder 17,
Inverse quantization is performed by the scalar quantizer 21 to obtain screen data indicated by ● in FIG. 10 (1), and the interpolator 18 is used in FIG. 10 (1)-
In the order shown in (3), the entire frame screen is filled up to obtain a reproduction screen and the reproduction screen is stored in the frame memory 20.

On the other hand, on the transmitting side as well, the data quantized by the scalar quantizer 12 is sent to the interpolator 16 and the same interpolation operation as that performed by the interpolator 18 on the receiving side is performed to perform the same interpolation as shown in FIG. 10 (3). A similar playback screen shown in is stored in the frame memory 15.

In the screen transmitted to the second stage, as shown in FIG. 9 (2), the sub-sampler 11 sub-samples ◯ pixels,
Unlike the first stage, this time in sync with this subsampler
Also in 14, the same ◯ pixel of the screen data previously stored in the frame memory 15 is subsampled.

Therefore, in the scalar quantizer 12, the interpolation error signals of both sub-samplers 11 and 4 are scalar-quantized and transmitted through the entropy coder 13. The output signal from the sub-supplier 14 is a predicted reproduction screen. The smaller the error signal, the more the transmission amount sent from the entropy decoder 13 can be reduced.

In the case of this second-stage screen, the screen already stored in the frame memory 20 on the receiving side is also subsampled by the subsampler 19 in synchronization with the subsamplers 11 and 14 on the transmitting side, and the scalar quantizer 21 is used. Interpolator 18 in addition to the output of
Is interpolated in the order shown in Fig. 10 (2) and (3)
The entire frame screen that is similar to the screen of the tier is reproduced and stored in the frame memory 20.

On the transmitting side, the same operation as this is performed by adding both outputs of the subsampler 14 and the scalar quantizer 12 and interpolating by the interpolator 16, and stored in the frame memory 15.

Thereafter, in the same manner, the still screens from the third stage to the sixth stage are subsampled as shown in FIGS. 9 (3) to 9 (6). In this case, the third stage is interpolated in each order of Fig. 10 (4) and (5), the fourth stage is Fig. 10 (3), the fifth stage is Fig. 10 (5), and the sixth stage is interpolated. The transmission process for a set of still screens (six solid lines) is completed.

[Problems to be solved by the invention]

In such a conventional still image encoding method, scalar quantization is performed on all sub-sampled images of both a large distortion portion and a small distortion portion of the still image before transmission, resulting in a large amount of information and a time required for transmission. There was a problem that it would be long.

On the other hand, it is conceivable to use a vector quantization method as a highly efficient encoding method for encoding and transmitting image information at a low bit rate.

That is, as shown in FIG. 11, a fixed vector quantization codebook 30 (consisting of code groups) prepared in advance is prepared, and the vector quantizer 31 selects from the codebook 30 according to the error signal. In this method, the one with the smallest error is selected (matching), and only that index is transmitted to reduce the transmission amount.

However, since the codebook is fixed in this method as well, there is no choice but to select an inappropriate codebook in which the error cannot be reduced, and there is a problem that the distortion peculiar to vector quantization is involved. Since there are usually edge pixels in the image, the amplitude of the edge pixels is large, so that the amplitude level of the entire error signal becomes high, and there is a problem that proper matching cannot be performed.

Therefore, an object of the present invention is to realize a still image coding method that can reduce the transmission time and reduce distortion by reducing the amount of transmission information even in a screen including edge pixels.

[Means for solving problems]

FIG. 1 is a diagram conceptually showing a still image encoding system according to the present invention for achieving the above object.
Sub-sampler 11 sub-samples, quantizer 1 quantizes and variable-length code with entropy code 13 and transmits, while a reproduction screen is created from the next screen with predictor 2 and synchronized with the first sub-sampler 11. An improved still image coding method in which the reproduction screen is subsampled by the second subsampler 14 and the interpolation error signal with the subsample signal of the next screen is quantized by the quantizer 1 for a predetermined number of steps. Is.

That is, in the present invention, the quantizer 1 is the vector quantizer 3 that performs block coding on the interpolation error signal and matches it with the fixed codebook 4, and further extracts edge pixels from the reproduction screen to obtain the interpolation error signal. An edge correction means 5 for generating a scalar quantized signal only for the edge pixels, and a subtractor 6 for subtracting the scalar quantized signal from the interpolation error signal or the output of the vector quantizer 3 to generate a scalar quantized error signal. , Are provided.

[Action]

In FIG. 1, scalar quantization is applied to interpolation error signals of an original image subsampled and a reproduction prediction screen subsampled. In this case, in the scalar quantization, the edge correction means 5 detects the edge pixel of the reproduction screen predicted by the predictor 2, and scalar-quantizes only this edge pixel to obtain the interpolation error signal or the output from the vector quantizer output. The scalar quantization value is subtracted and sent to the vector quantizer 3 or the transmission line.

 On the receiving side, the reverse operation of this transmitting side is performed.

In this way, the amount of transmission can be reduced by flattening the large amplitude due to the edge pixels of the screen, and the distortion due to vector quantization is also reduced.

〔Example〕

An embodiment of a still image encoding system according to the present invention will be described below.

FIG. 2 shows an embodiment of the still image encoding system of the present invention conceptually shown in FIG. 1. In FIG. 2, the predictor 2 shown in FIG. Interpolator 16 as in the example
Further, the edge correction means 5 includes an edge extractor 51 for extracting edge pixels from the reproduction screen, a third sub-sampler 52 for sub-sampling the edge extraction screen, and It is composed of a control circuit 53 that selects an interpolation error signal only for edge pixels in the subsampled screen, and a scalar quantizer 54 that scalar-quantizes the selected interpolation error signal.

On the other hand, as shown in FIG. 3, the receiving side is replaced by the vector quantizer 3a, the codebook 4a, the edge extractor 51a, as in the transmitting side, instead of the scalar quantizer in the conventional example.
Subsampler 52a, control circuit 53a, and scalar quantizer
54a forms a closed loop.

Next, the operation of the transmitting side (encoding side) in the embodiment shown in FIG. 2 will be described with reference to the codebook change diagram shown in FIG. 4 and the flowcharts shown in FIGS. 5 and 6. .

First, the sub-sampler 11 first sub-samples and outputs the first screen of FIG. 9 (1). In this case, the vector quantizer 3 uses the index of the input block signal (this is the entropy coder 13). (May be created) as it is, or a fixed load prepared in the codebook 4 is fully searched to detect (match) the code with the smallest error, and the index is sent to the transmission line. At this time, the frame memory is
The screen predicting the screen of the receiving side is stored in 15.

At this time, the edge information (pixels) in the screen is extracted from the first stage screen information stored in the frame memory 15 by the edge extractor 51 including a Laplacian filter.

That is, as shown in FIG. 5, when the reproduction screen of the frame memory 15 is a mask of 512 × 512, the threshold is D, and the filter is 3 × 3 (step S1 in FIG. 5), the pixel X of the screen is displayed.
(I, J) filters from I, J = 2 (the same)
S2). This is because the upper leftmost pixel of I and J = 1 cannot be used.

Next, in order to raise the edge pixels, the filter M
Multiply each pixel X (I, J) by (3 × 3) (S
3). Then, it is checked whether or not the absolute value of each pixel X (I, J) is equal to or larger than a predetermined threshold D (at step S4). If the absolute value exceeds the threshold D, the pixel X (I, J) is set to "0". If it does not exceed D, the pixel X (I, J) is set to "1" (S5, S6). That is, the pixel data is set to "0" only in the area where it can be identified as the edge portion.

This operation is executed for 512 x 512 (S7 to S10 in the same), 2
The value edge information is stored in the edge extractor 51 (see S1
1).

Figure 4 (1) shows subsampler 11 and subsampler 14.
The interpolation error signal is generated at the time of sub-sampling the second screen of Fig. 9 (2). In the figure, the pixels marked with ○ are sub-samplers 11 and 14 After being sampled, the interpolation error shown in FIG. 4 (1) is input to the subtractor 6. In order to make the figure easier to see, only the signal level of the sub-sampling points (appropriately selected) will be shown below.

The interpolation error shown in FIG. 4 (1) is input to the control circuit 53 of the edge correction means 5. During this period, the third sub-sampler 52 sub-samples the binary screen information stored in the edge extractor 51 in synchronization with the first and second sub-samplers 11 and 14 and outputs it to the control circuit 53. There is.

Therefore, the control circuit 53 sends the interpolation error of only the edge pixel among the pixels sub-sampled by the sub-sampler 52 to the scalar quantizer 54 for the scalar quantization.

The scalar-quantized interpolation error is the other input of the subtractor 6, and further produces an error (which becomes a scalar quantization error) from the interpolation error of the subsamplers 11 and 14.

In this way, the reason why the interpolation error is further made into the scalar quantization error is that the edge pixel on the screen remains in the interpolation error because the amplitude level is high, and if this is directly vector-quantized, it is dragged to the edge pixel level. This causes the matching to shift.

In this way, in the vector quantizer 3 to which the scalar quantization error is input, the signal level at the sub-sampling point in FIG. 4 (1) is repacked in a software conceptual manner as shown in FIG. 4 blocks, and referring to the data of codebook 4, the vector quantization code with the smallest error is selected (matching) as shown in FIG. 4 (3).
To do.

The vector quantization code determined in this way is used for the screen processing of the third stage. The vector quantization code of FIG. 4 (3) is also an appropriately selected number.

Here, if the interpolation error at the third stage is assumed to be the value shown in FIG. 4 (4) (not particularly related to the values in FIG. 4 (1) to (3)), it is indicated by a triangle mark. The location is determined as an edge pixel by the edge extractor 51 from the reproduction screen of the second stage. The pixels marked with Δ are sub-sampled by the sub-sampler 52 to be scalar-quantized by the scalar quantizer 54. Four
When subtracted by the subtracter 6 from the interpolated pixel in FIG. 4 (4), it is input to the vector quantizer 3 in the block form on the left side in FIG. 4 (6) as information of small amplitude as shown in FIG. 4 (5). Then, matching is performed and the data is output in the form on the right side of FIG. This output vector is sent to the receiving side in the form of an index as described above.

FIG. 6 shows the operation procedure of the control circuit 53, the scalar quantizer 54, and the subtracter 6, and shows binary edge information X (51
By sub-sampling 2, 512), if the sub-sampling point is “0”, it is determined to be an edge pixel and the interpolated pixel F ^* (512, 512) is scalar-quantized to further scale the interpolated pixel. The procedure for obtaining the quantization error F ^** (512, 512) is shown.

 Similarly, the screen processing of the fourth to sixth stages is executed.

Thus, 6 to complete a set of still image transmissions.
The data of the output code of the vector quantizer 3 becomes smaller every time the screen transmission is performed in stages, and the amplitude can be made smaller.

Note that the sub-sampling operation on the receiving side shown in FIG.

In the above embodiments, the method using only the vector quantizer as the quantizer 1 has been described, but the hybrid quantization that employs not only vector quantization but also scalar quantization is shown in FIGS. 7 (1) and (2). ). In FIG. 7, only the quantizer 1 is shown.

Further, the subtractor 6 is inserted before the vector quantizer 3, but the same effect can be obtained by inserting it after the vector quantizer 3.

Further, the step S4 of FIG. 5 may be executed by the control circuit 53.

Further, although the entropy coder 13 performs variable length coding on the output of the vector quantizer 3, it is more preferable that the output of the scalar quantizer 54 is also variable length coded.

〔The invention's effect〕

As described above, according to the still image encoding method of the present invention, pixels on the interpolation error corresponding to the edge pixels on the transmission screen are scalar-quantized to the vector quantizer on the next-stage screen. Since it is subtracted from the interpolation error or the matching output code from the vector quantization, the matching in the vector quantizer becomes more appropriate, the transmission amount is small, and the distortion caused by the vector quantization can be reduced. There is.

[Brief description of drawings]

FIG. 1 is a block diagram showing the principle of a still image coding system according to the present invention, FIG. 2 is a block diagram showing an embodiment of the still image coding system shown in FIG. 1, and FIG. FIG. 4 is a block diagram showing an embodiment of the receiving side corresponding to the transmitting side of the figure, FIG. 4 is a diagram for explaining the selection process of the codebook used in the present invention, and FIG. 5 is a book. FIG. 6 is a flow chart showing the operation of the edge extractor used in the present invention, FIG. 6 is a flow chart for explaining the procedure in the control circuit, the scalar quantizer, and the subtracter used in the present invention, FIG. 7 (1) and (2) is a block diagram showing another embodiment of the present invention, FIG. 8 is a block diagram showing an example of a conventional still image encoding system, and FIGS. 9 (1) to 9 (6) are screen transmissions. Figures for explaining the method of sub-sampling in a step, and Figures 10 (1) to (5) are supplementary figures. How diagram for explaining of FIG. 11 is a block diagram, in the case of a combination of a vector quantization and fixed codebook. In FIGS. 1, 2, and 7, 1 is a quantizer, 2 is a predictor, 3 is a vector quantizer, 4 is a codebook, 5 is an edge correction means, 6 is a subtractor, 11, 14 Indicates a subsampler, and 8 indicates a scalar quantizer, respectively. In the drawings, the same reference numerals indicate the same or corresponding parts.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Kiichi Matsuda 1015 Kamiodanaka, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited (72) Inventor Toshitaka Tsuda 1015, Kamedotachu, Nakahara-ku, Kawasaki City, Kanagawa Prefecture, Fujitsu Limited

Claims (3)

[Claims] 1. An original image is sub-sampled by a first sub-sampler (11), quantized by a quantizer (1) and variable-length coded by an entropy coder (13) for transmission, and a reproduced screen is displayed from the next-stage screen. The reproduction screen is subsampled by the second subsampler (14) in synchronization with the first subsampler (11) by the predictor (2), and the interpolation error signal with the subsample signal of the next-stage screen is obtained. In a still image coding method in which quantization in a quantizer (1) is repeated a predetermined number of times, the quantizer (1) block-codes the interpolation error signal to perform matching with a fixed codebook (4). A vector quantizer (3), which further comprises edge correction means (5) for extracting edge pixels from the reproduction screen and generating a signal in which only the edge pixels of the interpolation error signal are scalar-quantized. A subtracter (6) for subtracting the scalar quantized signal from the interpolation error signal or the output of the vector quantizer (3) to generate a scalar quantized error signal; Method. 2. The edge correction means (5), an edge extractor (51) for extracting edge pixels from the reproduction screen, and a third sub-sampler (52) for sub-sampling the edge extraction screen, It is composed of a control circuit (53) for selecting the interpolation error signal only for edge pixels in the subsampled screen, and a scalar quantizer (54) for scalar quantizing the selected interpolation error signal. The still image encoding method according to claim 1. 3. A quantizer (1) for performing a combination of vector quantization by a vector quantizer (3) and a scalar quantizer (8) by a scalar quantizer (8). The still image encoding method according to the first item of the above.