JPH0327688A - Decoder for vector quantizing code - Google Patents

Abstract

PURPOSE:To surely attain up-dating of a code book by providing a memory storing one of patterns generated at the same time. CONSTITUTION:A picture data of one pattern obtained by adding an average data and a difference data of two patterns is obtained through an adder 51 and inputted directly to a selector 72. Similarly, a picture data of the 2nd pattern obtained by subtracting the difference data from the average data of the two patterns is obtained through a subtractor 52 and the data is stored once in a memory 6. A selector 72 selects the input from the adder 51 on the display of the 1st pattern and sends it to an interpolation filter 71. The selector 72 selects the picture data of the 2nd pattern stored in the memory 6 and sends it to the filter 71. Code book RAMs 21, 22 are not used for the coding during this period and only the quantity required to decode the succeeding picture pair is up-dated.

Description

[Detailed Description of the Invention] [Field of Industrial Application] The present invention relates to vector quantization for decoding and reproducing moving image data compressed and recorded on a storage medium such as a CD-ROM using vector quantization. The present invention relates to a code decoding device.

[Conventional technology]

One of the compression encoding methods for moving images is a method using vector quantization. This method has the advantage that the decoding process is simpler than the encoding process, and the decoding device can be implemented on a relatively small scale.

FIG. 8 is a configuration diagram of a moving image decoding device that decodes compression-encoded image data using this vector quantization. In FIG. 8, 1 is a buffer memory for temporarily storing codes extracted from a storage medium such as a CD-ROM, 2 is a codebook ROM for decoding vector quantized codes, 3 is a D/A converter for converting decoded image data into an analog video signal.

Further, the address output from the address generation circuit 4 is supplied to the buffer memory 1 and the codebook ROM 2, and is used when decoding the vector quantized code.

By the way, vector quantization is a method of assigning one code (called an index) to an image block consisting of m pixels x n pixels (m and n are natural numbers), but here, we will assign one code (called an index) as shown in Figure 10. Assume that each index corresponds to an image block of 2 pixels x 21g. This correspondence table between index and image data is called a codebook.

In the decoding device having the configuration shown in Fig. 8, an image block of m pixels x niji1l pixels is regarded as being composed of n lines of m pixels, and one index is decoded in n steps.

Specifically, as shown in Figure 9 (in an example where m=n=2), the same address is given to the buffer memory n times.

Accordingly, the same index is given n times from the buffer memory to the next-stage food book ROM as an upper address. Further, the codebook ROM is given horizontal pixel addresses and line addresses within the image block as lower addresses. As horizontal pixel addresses, m addresses are given every n times of the same index given by the buffer memory.

On the other hand, a different address is given to the line address each time.

When such an address is given to the buffer memory and codebook and the index is decoded in several lines, 6 of the image data of a specific line on the image block corresponding to each index is extracted and outputted in a delayed manner. can do. This decoded image data is directly transferred to the D/A
Since it is possible to input the signal into a converter, convert it into a video signal, and display it on a display, a moving image decoding device can be realized with a simple circuit W5 structure.

Incidentally, in compressing and encoding m-images, frame correlation of a plurality of screens is often used in order to improve the compression efficiency.

W, 311 is a 'wJ image using frame correlation, which is composed of a plurality of continuous still images, but in this example, each image is divided into two and sequentially encoded.

Hereinafter, a coding unit consisting of these two still images will be referred to as an image pair. This image pair is binarized and then added and g. are calculated, resulting in average data and difference data, respectively. These two types of data are then vector quantized to generate an index and a food book. Hereinafter, the index and codebook created from average data will be referred to as index A and codebook A, and the index and codebook created from differential data will be referred to as index D and codebook D. Furthermore, in the vector quantization here, it is assumed that an appropriate index is assigned to an image block consisting of 2ilki x 2 pixels.

FIG. 2 shows the components of a decoding device for decoding compressed data encoded using frame correlation in this manner.

In Figure 2, 11 and 12 are storage media (\
& -1 in the bar for temporarily storing the extracted code.
::! buffer memories 21 and 22 are codebook RAMs for decoding vector quantization codes, 71 is an arithmetic unit that operates on decoded data output from the codebooks RAMs 21 and 22, and 71 is a one-dimensional complementation filter, 3 is a D/A to convert its output into an analog signal
It is a converter. Further, the address output from the address generation circuit 4 is supplied to the buffer memories 11 and 12, the codebooks RAM 21 and n, and the memory 73.

Data a1 input to buffer memories 11 and 12
and a2 are the index A and index D mentioned above, respectively. After the index for one image pair is stored in both buffer memories, the index is output to the codebooks RAM 21 and 22 according to the address e supplied from the address generation circuit 4. Codebooks R A M 21 and 22 have the contents of the codebook person and codebook D written therein, respectively, and the index can be read by outputting the indexes output from the buffer memories 11 and 12 and the image data originating from the address. decryption is performed.

Now, the data decoded using the codebooks RAM 21 and 22 are input to the arithmetic unit 5. By the way, when the image data of the first image of the original is x and the image data of the second image is y, their average X = ( x + y) / 2, the difference y
= (x-y)/2, then x+y=x, C1 and difference data C2 are added to obtain the image data of the first screen. Furthermore, when displaying the second screen, the average data C1 and the difference data C2 of the two screens constituting the image pair are subtracted to obtain image data for the second screen.

The output from the arithmetic unit 5 is corrected in the horizontal direction by the Kikuma filter 71, and then converted into an analog video signal d by the D/A converter 3 and displayed on the display.

[Problem to be solved by the invention]

Image data changes significantly. On the other hand, if vector quantization is performed with a fixed codebook, distortion due to vector doji conversion will become noticeable depending on the screen unless the codebook size is increased at the expense of encoding efficiency. Therefore, a method is generally used to update the codebook according to changes in image data, but in the decoding method described above, the codebook is used to decode the index while the screen is displayed. , as shown in FIG. 6, the codebook can only be changed within the vertical retrace period after one screen has been displayed. Therefore, when a scene change occurs, for example, there are too many vectors to change and the codebook may not be able to be updated. For example, if the image block size is 2 pixels x 2 pixels, the index at type is 256
In this case, in order to rewrite the entire codebook within the vertical retrace period, it takes 63.5 μsec (1 line plan) x 15 (vertical Number of lines during retrace period) = 93
Data of 256 x 4 pixels must be rewritten in 5 μsec. This is at least 0.5μs
This means that it is necessary to access the codebook using ec. In reality, since various screen controls are performed during the vertical retrace period, it is difficult to perform this reversal using an inexpensive 8-bit microcomputer or the like.

An object of the present invention is to provide a video decoding device that decodes a compression code using frame correlation, in which image data of multiple screens are mutually calculated, converted into data of number a class, and then each of the data is vector quantized. To provide a means for reliably updating a large number of vectors in a codebook even if it becomes necessary to update a large number of vectors in a codebook due to a scene change or the like.

[Means to solve the problem]

In order to achieve the above object, in the present invention, in a moving image decoding device that decodes a compression code using frame correlation, among several screens of image data obtained by further calculating vector data obtained by decoding an index, At least two screens are created at the same time, and the screen is configured to have a memory for storing at least one of the simultaneously created screens.

[Effect]

In the present invention, while one screen is being created and displayed, image data for another screen is also created and stored in the memory. In this way, there is no need to decode the index using the codebook during the period when image data is output from this memory, so the entire period can be used for updating the codebook. This makes it possible to reliably update the codebook significantly even if the image data changes significantly due to a scene change or the like.

〔Example〕

Hereinafter, a central embodiment of the present invention will be described using the drawings.

FIG. 1 shows the configuration of a moving image decoding device. This device converts pressure image data encoded by vector quantization into C
The data is read from a storage medium such as a D-ROM, decoded, and then displayed on a display.

In FIG. 1, 11 and 12 are buffer memories for temporarily storing codes read from a storage medium, 21 and 22 are codebook RAMs for decoding vector quantization codes, and 51 is a codebook. R
A M An adder for adding together the decoded data output from 21 and 22, 52 a similar subtracter, 73 a memory for storing the subtracted data, 72 an adder 51
71 is a one-dimensional interpolation filter for smoothing the image; and 3 is a D/A converter for converting the output into an analog signal. Furthermore, the address output from the address generation circuit 4 is transmitted to the buffer memories 11 and 12 and the codebook R.
It is supplied to AM 21 and 22 and memory 6.

The data a1 and a2 input to the buffer memos IJ11 and l2 are the index A and index D described above, respectively. After both buffer memories have stored indexes for l image pairs, they output the indexes to the codebook RAMs 21 and 22 in accordance with the address e supplied from the address generation circuit 4. Codebook R A M 21 and 22 each contain codebook A
and the contents of the food book D are omitted, and the index can be decoded by outputting predetermined image data according to the index output from the buffer memories 11 and l2 and the address f supplied from the address generation circuit 4. is executed.

Here, some additional information will be given regarding the addresses supplied to the buffer memory and food book RAM.

As shown in FIG. 4, the address output by the address generation circuit 4 is matched to the screen display on the display. For example, 256 pixels in the horizontal direction and 240 lines in the vertical direction are displayed within the display period of one screen. Access minutes. Also,#!
As shown in Figure 5, Buffer Memo IJ 11 and 12
128 x 120 indexes are stored in each of the codebooks RAM 21 and 22, and 1024 pixel data corresponding to 256 types of indexes are stored in each of the codebooks RAM 21 and 22.

? The address generation circuit 4 includes buffer memories 11 and 12.
On the other hand, among the addresses output from the address generation circuit 4, the address e is obtained by removing bits indicating even and odd from among the addresses in the horizontal direction and the vertical direction. supply each. Therefore, the same indexes are given to the codebooks R A M 21 and 22 as addresses bl and b2 when the address generation circuit 4 accesses the odd lines on the display screen, even when the even lines are accessed. , this distinction between odd and even lines is determined by the codebook R A M
This is done by the lower address given to 21 and 22. That is, the two addresses (addresses indicating even-odd in the horizontal and vertical directions) that were not given to the buffer memory are supplied as the lower address f of both codebook RAMs. By supplying such addresses, in this embodiment, one index is decoded in two lines. In other words, when displaying odd lines, the decoded image data of the odd lines on the disordered image block corresponding to each index. ``C, 2:1 only'' is output continuously, and when even-numbered lines are displayed, only the decoded image data of the even-numbered lines on the image block corresponding to each index are continuously output.

Now, the data decoded by the codebooks R A M 21 and 22 are sent to the adder 51, and at the same time, the data is sent to the subtracter 52.
is input. What is obtained through the adder 51 is the image data of the first screen obtained by adding the average data cl of the two screens and the difference data c2, and is directly sent to the selector 72.
is input. Similarly, what is obtained through the subtracter 52 is one image data of the second screen obtained by subtracting the difference data c2 from the average data c1 of the two screens, but this image data is temporarily stored in the memory. It is stored in 6.

The selector 72 is used to switch between the first image data and the second image data. When displaying the first image, the selector 72 selects the input from the adder 5l and selects the input from the next stage filter 71.
send to. After the image is enlarged horizontally to 512 pixels by this interpolation filter 71, Dfi. ．． ) A transformer 3
It becomes an analog video signal d and is displayed on the display.

When displaying the second screen, the selector 72 selects the image data of the second screen stored in the memory 6 and sends it to the next-stage O interpolation filter 71. During this time, codebook RAM21
and 22 are not used for encoding and are only updated in order to decode the next image pair.

FIG. 7 shows the timing for index decoding and codebook updating described above.

In this way, in the central embodiment, the existence of the memory 6 makes it possible to secure sufficient time for updating the codebooks 21 and 22.

That is, while the first screen is being displayed, the decoding of the second screen is also completed, and the entire period during which the second screen is being displayed can be used for updating the codebook. Therefore, scene changes etc. i[! lI1i! ．． Even if the data changes significantly, it is possible to reliably update the codebook significantly.

The embodiment described above decodes a code obtained by vector quantizing the data obtained by calculating the image data of two screens, but the proposed rule is based on the number of screens set to two. The present invention can also be applied to vector quantization by performing calculations such as orthogonal transformation on three or more image data.

〔Effect of the invention〕

As described above, in the present invention, at least two of the several screens of image data obtained by further calculating the vector data obtained by decoding the index are simultaneously created, and at least one of the simultaneously created screens is is stored in memory. In this way, there is no need to use the codebook to decode the index during the period when image data is being output from this memory, so the entire period can be used for updating the codebook. As a result, even if the image data changes significantly due to a scene change, etc., it is possible to reliably update the codebook significantly, and as a result, the displayed image and music will be updated.

[Brief explanation of drawings]

Fig. 1 is a block diagram of a moving picture decoding device to which the present invention is applied, Fig. 2 is a block diagram of a conventional moving picture decoding device, and Fig. 3 shows the decoding process of pressure mIIIb image data decoded in this embodiment. 4 is a diagram showing the address generation specifications of the address generation circuit in this embodiment. FIG. 5 is a diagram showing the data structure in the buffer memory and codebook in this embodiment. 6 and 7 are diagrams showing the timing of index decoding and codebook update in the conventional decoding device of the present invention, respectively, and FIG. 8 is a diagram showing the conventional decoding device that decodes data encoded by vector quantization. Figure 5g9 is a diagram showing the general configuration, Figure 5g9 is a diagram showing the address supply to the buffer memory and codebook of the decoding device in Figure 8, and Figure 10 is a diagram showing the pixel configuration of the image block corresponding to each index. . l...Buffer memory 2...Code block 3...D/A converter b.
...Index d...Analog video signal C...Decoded vector address = O Fig. 3 No. 4 - (25GX240) -1 No. ? figure

Claims (1)

[Claims] (1) In a vector quantization code decoding device having a vector storage means for storing vectors to be output corresponding to indexes assigned to vectors that are a plurality of data sets, the vector quantization code is A calculation means having a plurality of storage means, for calculating a plurality of types of vectors outputted by the plurality of vector storage means to obtain a plurality of types of decoding results, and a calculation means for obtaining a plurality of types of decoding results, and a part of the plurality of types of decoding results. a decoding result storage means for storing the type; a switching means for selecting and outputting one type from among the decoding results stored in the decoding result storage means and the remaining decoding results not stored in the decoding result storage means; A vector quantization code decoding device comprising: