JP2641773B2 - Vector quantization coding device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vector quantization coding apparatus, and in particular, generates an input vector by grouping and blocking a plurality of video signal sequences, and codes the vector. The present invention relates to a vector quantization encoding device that performs quantization by comparing with an output vector on a book.

[Prior Art] FIG. 3 is a conceptual diagram of vector quantization. First, a basic concept of vector quantization will be described with reference to FIG.

Now, the input signal sequence is grouped every k samples,
It is assumed that a k-dimensional input vector is formed.

The input vector is represented as a point in a k-dimensional signal space as shown in FIG. In this k-dimensional signal space,
A set (codebook) of output vectors determined according to the probability distribution of the input vectors is prepared in advance. In vector quantization, an output vector which is the shortest distance (minimum distortion) from an input vector in a k-dimensional signal space is calculated by:
This is done by searching from the set of output vectors and replacing the input vector with its output vector.

The above can be mathematically defined as follows.

The input vector is [x ₁ , x ₂ ,..., X _k ], the k-dimensional signal space is R ^k (∈R ^k ), and the division of R ^k is P = ｛p ₁ , p ₂ ,.
p _N}, each subspace p representative points _i (e.g., center of gravity) the output vector is _i, the output vector _i set (codebook) and _{C = {1, 2, ......} , When _N}, k The dimension N-level vector quantization Q can be defined as follows.

The division P of Q () = _i , if∈p _i R ^k is a subset of a power set of R ^k that satisfies the following equation.

In addition, vector quantization is performed by coding C as shown below.
And decoding D in series.

C: ⇒ _i , if∈p _i (i: index) D: _i ⇒ _i Q: C · D In other words, on the encoding side of vector quantization, the index of the selected output vector is used as encoding information. By providing the same codebook as the encoding side on the transmitting or recording and decoding side, it is possible to reproduce the output vector from the received index.

As described above, in vector quantization, k
The arrangement of output vectors in a dimensional signal space has a large effect on coding efficiency. A codebook, which is a set of output vectors for this purpose, is obtained by grouping vectors that are close to each other by a suitable method in a training sequence (a set of input vectors for generating a codebook), and forming a k-dimensional code. The signal space can be generated by dividing the signal space and using a representative point of each division as an output vector.

Hereinafter, encoding and decoding of a conventional vector quantization method will be described with reference to the drawings.

FIG. 4 is a block diagram of a conventional quantization coding circuit, and FIG. 5 is a block diagram of a conventional quantization decoding circuit. 4 and 5, 402 is an A / D conversion circuit, 404 and 505 are frame memories, 405 is a vector formation unit, 406 is a training sequence generation unit, 407 is a codebook generation unit, and 408 and 502 are codebooks. A memory, 411 is a vector quantizer, 501 is an inverse vector quantizer, 504 is an image forming unit, and 507 is a D / A converter.

The quantization encoding circuit shown in FIG. 4 is a circuit for vector-quantizing image data of a moving image and transmitting or recording a quantization index and a codebook. The quantization decoding circuit shown in FIG. , A quantization index, and a codebook to reproduce a vector and output a moving image.

In FIG. 4, an input video signal 401 is output from an A / D conversion circuit 40.
According to 2, the analog video signal is converted into digital image data 403, and the frame memory group 404 stores and holds n frames. The image data 403 initially read from the frame memory group 404 is captured by the vector forming unit 405 every k samples, and the input vector 410
Formed. This input vector 410 is stored in the training sequence generation unit 406 for n frames.

The codebook generator 407 generates a codebook corresponding to the probability distribution of the input image data for n frames held in the training sequence generator 406, and stores the codebook in the codebook memory 408. Let it. The code book data 409 stored and held in the code book memory 408 is transmitted to a transmission medium or recorded on a recording medium.

Next, the image data 403 is read again from the frame memory group 404, and an input vector 410 is formed. And
This input vector 410 is obtained by a vector quantizer 411.
It is used for searching for an output vector that is the shortest distance (having the minimum distortion) on the codebook memory 408, and the index of the searched output vector is the quantization index data 41.
As 2, it is transmitted to a transmission medium or recorded on a recording medium.

Next, the operation of the vector quantization decoding circuit shown in FIG. 5 will be described.

First, the read or transmitted codebook data 409 is stored and held in the codebook memory 502. Next, an output vector corresponding to the read or transmitted quantization index data 412 is determined by the inverse vector quantizer 501. The selected output vector 503 is output to the image data 50 by the image forming unit 504.
6 is formed and stored in the frame memory 505. The image data 506 is converted into an analog video signal by the D / A conversion circuit 507, and is output as an output video signal 508.

The above-described vector quantization method according to the related art transmits or records a quantization index and a code book as coding information, and thus has very high coding efficiency.

As a conventional technique related to the above-described vector quantization, for example, a technique described in IEICE IE87-48, pp. 47-54 “Vector quantization and its application to image coding” and the like are known. I have.

[Problems to be Solved by the Invention] In the above-described conventional technology, in order to make the content of the codebook match the probability distribution of the input image data, the input image data is divided into n frames, and a codebook is generated for each frame. To perform vector quantization.

In general, a moving image includes various images such as a rapidly moving image, an image with a fine pattern, and a scene change image, and the image characteristics are changed irregularly.

According to the above-described conventional technology, image data is divided into n frames and quantized, regardless of whether or not the above-described irregularly changing image exists in the n frames. Image quality may deteriorate.
There is a problem that the image quality cannot be kept constant.

SUMMARY OF THE INVENTION An object of the present invention is to solve the above-mentioned problems of the related art, and to eliminate a deterioration in image quality that has occurred conventionally when image characteristics change, and to provide a high-quality moving image in accordance with the change in image characteristics. It is an object of the present invention to provide a vector quantization encoding device capable of outputting.

[Means for Solving the Problems] According to the present invention, in a vector quantization coding apparatus for coding an input video signal, a scene change detection means for detecting a scene change from an input video signal; When a scene change detection signal is output from the scene change detection means, a code book for generating a code book from input video signals of a plurality of frames after the frame where the scene change is detected is generated by the code book generation means. This is achieved by including codebook storage means for storing the codebook stored in the memory, and quantization means for quantizing the input video signal using the codebook stored in the codebook storage means.

In addition, the object is to provide a codebook generating means, comprising, from an input video signal, a transition image detecting means for detecting a transition image period, which is a video period in which movement is intense and a picture period in which a picture is fine, instead of the scene change detecting means. This is achieved by generating a codebook from input video signals of a plurality of frames during the transition image period when the transition image detection signal is output from the transition image detection means.

Further, the object is to provide a vector quantization coding apparatus for coding an input video signal, which evaluates the number of significant frames having a large difference from the previous frame from the input video signal to determine a codebook size. Means, codebook generating means for generating a codebook according to the codebook size information from the codebook size determining means, codebook storage means for storing the codebook generated by the codebook generating means, A quantization encoding unit that quantizes the input video signal using a codebook stored in the codebook storage unit.

[Operation] According to the present invention, first, the characteristics of the input image data read from the frame memory, that is, an image with a sharp movement, an image with a fine pattern, a scene change, and the like are detected.
Be evaluated. Then, based on the evaluation result, the following operation is performed, thereby achieving vector quantization without deterioration in image quality.

When a scene change occurs, a codebook is generated up to the previous frame as a training sequence,
After the scene change, the codebook is generated in the subsequent frames. Then, in the case where images with abrupt movements, images with fine patterns, and the like are continuous, a codebook is generated using a frame sequence in which these are continued.

In addition, when a large number of images with abrupt motion, images with fine patterns, scene changes, and the like are present in the evaluated image data, the number of output vectors of the codebook is increased (the size of the codebook is increased).

Therefore, according to the present invention, since a codebook is generated in consideration of the characteristics of input image data, it is possible to suppress deterioration of image quality and output a high-quality image regardless of image characteristics. Become.

[Embodiment] Hereinafter, an embodiment of a vector quantization coding apparatus according to the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the first embodiment of the present invention. In FIG. 1, reference numeral 105 denotes an image characteristic evaluation unit, reference numeral 106 denotes a scene change detection signal, reference numeral 107 denotes a transition image detection signal, and other symbols are the same as those in FIG.

The first embodiment of the present invention shown in FIG. 1 is configured by adding an image characteristic evaluation unit 105 that receives image data 403 and evaluates image characteristics to the conventional technology described with reference to FIG.

In the first embodiment of the present invention, an input input video signal 401 is digitized by an A / D conversion circuit 402 and converted into image data 403, and image data for n frames is stored in a frame memory group 404. It is stored and stored. The image characteristic evaluation unit 105 stores the n held in the frame memory group 404
The characteristics of the image for the frame are evaluated. If a scene change is present in the image data of the n frames, a scene change detection signal 106 is output. If the images appear continuously, a transition image detection signal 107 is output. Then, these signals 106 and 107 are provided to training sequence generating section 406.

The training sequence generation unit 406 stores and holds the image data converted to the input vector by the vector forming unit 405 up to that frame up to the frame where the scene change detection signal 106 and the transition image detection signal 107 are generated as a training sequence. . The codebook generator 407 generates a codebook based on the training sequence based on the image data stored and held in the training sequence generator 406, and causes the codebook memory 408 to store and hold the codebook. This code book is recorded or transmitted as code book data 409.

After the generation of the codebook described above is completed, the vector quantizer 411 uses the image data up to the frame as the training sequence as an input vector, and
5 and vector-quantizes this image data. The quantized data is quantized index data
Recorded as 412 or transmitted.

If no scene change or transition image is detected in the n-frame image data, a codebook is generated from the n-frame image data and vector quantization is performed as in the case of the related art.

FIG. 6 is a block diagram showing a specific configuration of the image characteristic evaluation unit 105 in the first embodiment of the present invention described above.
Hereinafter, this will be described. In FIG. 6, 604 is a frame memory, 606 and 610 are threshold setting circuits, 607 and 611 are magnitude comparison circuits, 608 is an accumulation circuit, 612 is a scene change detection signal generation unit, and 613 is a transition image detection signal generation unit. is there.

The image characteristic evaluation unit 105 shown in FIG. 6 compares the significant pixel number calculation unit with the number of significant pixels output from the significant pixel number calculation unit and a predetermined threshold, and based on the result, a scene change detection signal. And a circuit for generating a transition image detection signal.

In the image characteristic evaluation unit 105, the difference between the input image data 403 and the image data (the initial state is a data value “0”) which is the information of the previous frame stored and held in the frame memory 604 is calculated. Thus, frame difference data 605 is obtained for each pixel. The size comparison circuit 607 compares the frame difference data 605 with a threshold value preset in the threshold value setting circuit 606. If the frame difference data 605 is larger than the threshold value, the pixel is regarded as a significant pixel and the accumulation circuit 608 is determined. Is accumulated for each frame. The accumulation circuit 608 outputs the accumulation result as the number of significant pixels 609.

FIG. 7 is a graph showing an example of the variation of the number of significant pixels 609 for each frame.

The number of significant pixels 609 for each frame is determined by a threshold setting circuit 610.
Are compared by the magnitude comparison circuit 611 with the threshold a for detecting a scene change and the threshold b for detecting a transition image. The scene change detection signal generation unit 612 includes:
If the number of significant pixels 609 is larger than the threshold a, the scene change detection signal 106 is output, and the transition image detection signal generation unit 613 outputs
If the number of significant pixels 609 is larger than the threshold value b, a transition image detection signal 107 is output.

8 and 9 show the image characteristic evaluation unit 105 as described above.
FIG. 1 is a diagram for explaining the operation of the embodiment of the present invention shown in FIG. 1 with reference to FIG. 1. Hereinafter, with reference to these drawings, the operation when a scene change detection signal 106 and a transition image detection signal 107 are output Will be described.

First, n frames of image data 403 are loaded into the frame memory 404, and the image characteristic evaluation unit 105 obtains and evaluates the number of significant pixels of the image data. As a result, for example, as shown in FIG. 8, if a scene change has occurred in the l-th frame, the image characteristic evaluation unit 105
A scene change detection signal 106 is output.

As a result, the codebook generating unit 407 generates a codebook using the training sequence up to the frame before the scene change, that is, up to the (l-1) th frame, and the vector quantizer 411 adds the codebook to this codebook. Based on this, vector quantization is performed up to (l-1) frames.

Next, the image data of (n + 1) th frame to (l-1) th frame is taken in while the image data of 1st frame to nth frame is left as it is, and the same encoding processing as described above is executed on this image data. .

The operation when the transition image detection signal 107 is output will be described with reference to FIG.

First, n frames of image data 403 are loaded into the frame memory 404, and the image characteristic evaluation unit 105 obtains and evaluates the number of significant pixels of the image data. As a result, for example, as shown in FIG.
If a transition image has occurred in (m + 1) frames up to (m) frame, the image characteristic evaluation unit 105 outputs a transition image detection signal 107.

As a result, the codebook generation unit 407 generates a codebook using the training sequence up to the frame where the transition image has occurred, that is, up to the (l−1) th frame, and the vector quantizer 411 generates Based on this, vector quantization is performed up to (l-1) frames.

Next, the codebook generation unit 407 generates image data where a transition image has occurred, that is, from l frames to (l +
A codebook is generated using up to m) frames as a training sequence, and the vector quantizer 411 performs vector quantization on frames from 1 to (lm) based on the codebook.

Thereafter, the image data of (n + 1) th frame to (l + m) th frame is taken in while the image data of (l + m + 1) th frame to nth frame is not changed, and the same encoding processing as described above is executed on this image data. .

According to the first embodiment of the present invention described above, the characteristics of the input image can be evaluated by the image characteristic evaluation unit 105, and the number of frames in forming a training sequence can be made variable. Since a codebook suitable for data can be generated, vector quantization according to a change in an input image can be performed, and a high-quality image can be output.

FIG. 2 is a block diagram showing the configuration of the second embodiment of the present invention. In FIG. 2, reference numeral 206 denotes a codebook size control signal, and other symbols are the same as those in FIG.

In the second embodiment shown in FIG.
401 is converted into image data 403 by an A / D conversion circuit 402, and image data for n frames is stored in a frame memory group 404.
Is stored. The image characteristic evaluation unit 105 evaluates the characteristics of the n frames of images, and if there are many images in the n frames that cause image quality degradation, such as scene change images and images with sharp movements, code A codebook size control signal 206 for variably controlling the size of the book is output to the codebook generating unit 407.

The training sequence generation unit 406 sets the n-frame image data converted into a vector by the vector generation unit 405 as a training sequence. Then, the code book generation unit 407
Generates a codebook having a codebook size suitable for input image data based on the training sequence and the above-described codebook size control signal 206, and causes the codebook memory 408 to store the codebook.
This code book is stored or transmitted as code book data 409.

After generating the above codebook, the vector quantum circuit
The vector 411 quantizes the image data of n frames converted into the vector 410 by the vector forming unit 405, and outputs the result as quantization index data 412. This quantized index data 412 is recorded or transmitted elsewhere.

FIG. 10 is a block diagram showing a specific configuration of the image characteristic evaluation section used in the second embodiment of the present invention. Tenth
In the figure, 1003 is a frame memory, 1005 and 1009 are threshold setting circuits, 1006 and 1010 are magnitude comparison circuits, 1007 is an accumulation circuit, 1011 is a significant frame number calculation circuit, 1012 is a codebook size determination circuit, The reference numerals are the same as in FIG.

In the tenth embodiment of the present invention shown in FIG. 10, the significant pixel number calculator has the same configuration as that of the embodiment of FIG. 1 described in FIG.

In the image characteristic evaluation unit 105, the difference between the input image data 403 and the image data (the initial state is a data value “0”) which is information of the previous frame stored and held in the frame memory 1003 is calculated. Thus, it is set as frame difference data 1004 for each pixel. The size comparison circuit 1007 compares the frame difference data 1004 with a threshold value preset in the threshold value setting circuit 1005, and if the frame difference data 1004 is larger than the threshold value, the pixel is regarded as a significant pixel and the accumulation circuit
Let 1007 accumulate that number for each frame. Accumulation circuit 1007
Outputs the accumulation result as the number of significant pixels 1008.

FIG. 11 is a graph showing an example of a change in the number of significant pixels 1008 of this frame.

The number of significant pixels 1008 for each frame is determined by the threshold setting circuit 10
The threshold value set to 09 is compared with the size comparison circuit 1010, and if the number of significant pixels 1008 is larger than the threshold value set in the threshold value setting circuit 1009, the frame is regarded as a significant frame, and the significant frame number calculation circuit At 1011, the number is accumulated. The codebook size determination circuit 1012 evaluates the number of significant frames, determines a codebook size suitable for quantizing n frames of image data, and outputs a codebook size control signal 206.

According to the above-described second embodiment of the present invention, when the characteristics of the input image are evaluated by the image characteristic evaluation unit 105 and the codebook is generated, the codebook size is made variable and the codebook size is varied. Since a codebook having a size can be generated, it is possible to encode an image signal in accordance with a change in an input image, and to output a high-quality image.

Although the above-described embodiments of the present invention both relate to an encoding circuit, the decoding circuit can use the circuit according to the prior art described with reference to FIG.

[Effects of the Invention] As described above, according to the present invention, switching of the number of frames at the time of training sequence generation and codebook evaluation are performed by evaluating characteristics of an input image such as scene change and intensity of motion. Since it is possible to switch the codebook size at the time of generation, etc., it is possible to perform vector quantization encoding of image data using a codebook according to the characteristics of the input image, and output a high-quality image. be able to.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment of the present invention, FIG. 2 is a block diagram showing a configuration of a second embodiment of the present invention, FIG. 3 is a conceptual diagram of vector quantization, and FIG. FIG. 4 is a block diagram of a conventional quantization coding circuit, FIG. 5 is a block diagram of a conventional quantization decoding circuit, and FIG. 6 is an image characteristic evaluation unit in the first embodiment of the present invention described above.
FIG. 7 is a block diagram showing a specific configuration of 105, FIG. 7 is a graph showing an example of the variation of the number of significant pixels 609 for each frame, and FIG.
FIG. 9 is a diagram for explaining the operation of the embodiment of the present invention shown in FIG. 1 having the image characteristic evaluation unit 105 as described above.
FIG. 10 is a block diagram showing a specific configuration of an image characteristic evaluation unit used in the second embodiment of the present invention, and FIG. 11 is a graph showing an example of a change in the number of significant pixels 1008 for each frame. 105: image characteristic evaluation unit, 106: scene change detection signal, 107: transition image detection signal, 206: codebook size control signal, 402: A / D conversion circuit, 404, 505, 604, 1
003: Frame memory, 405: Vector forming unit, 406
... Training sequence generator, 407 Codebook generator 408, 502 Codebook memory 411 Vector quantizer 501 Inverse vector quantizer 504 Image forming unit 507 … D / A converter, 606, 610, 1005, 1009…
... Threshold setting circuit, 607, 611, 1006, 1010... Magnitude comparison circuit, 608, 1007... Accumulation circuit, 612... Scene change detection signal generation unit, 613.
…… Significant frame number calculation circuit, 1012 …… Code book size determination circuit.

Continuation of the front page (56) References IEEE TRANSACTIONS ON COMMUNICATIONS Vol. COM [34] 7 (1986) P.703-710.

Claims (6)

(57) [Claims] 1. A vector quantization coding apparatus for coding an input video signal, wherein a scene change detection means for detecting a scene change from the input video signal, and a scene change detection signal is output from the scene change detection means. A code book generating unit configured to generate a code book from input video signals of a plurality of frames after a frame in which a scene change is detected; a code book storing unit configured to store a code book generated by the code book generating unit; A vector quantization encoding apparatus comprising: a quantization unit that performs quantization of the input video signal using a codebook stored in a codebook storage unit. 2. The apparatus according to claim 1, wherein the scene change detecting means includes a previous frame storing means for storing a video signal of a previous frame, and a difference between each pixel between the input video signal and the previous frame stored in the previous frame storing means. Calculating means for calculating, a significant pixel number calculating means for counting pixels in which the difference value output from the calculating means is larger than a first predetermined threshold, and a significant pixel number output from the significant pixel number calculating means. 2. A vector quantization coding apparatus according to claim 1, further comprising: a sea change detection signal generating means for judging that a scene change has occurred when the threshold value is larger than a second predetermined threshold and outputting a scene change detection signal. . 3. A vector quantization encoding apparatus for encoding an input video signal, comprising: a transition image detection means for detecting, from the input video signal, a transition image period which is a video period in which movement is intense and a video period in which a picture is fine; When a transition image detection signal is output from the transition image detection unit, a code book generation unit that generates a code book from input video signals of a plurality of frames during the transition image period, and a code book generated by the code book generation unit. A vector quantization coding apparatus comprising: a codebook storage means for storing; and a quantization means for quantizing the input video signal using a codebook stored in the codebook storage means. 4. The transition image detection means includes: a previous frame storage means for storing a video signal of a previous frame; and a difference between each pixel between the input video signal and the previous frame stored in the previous frame storage means. Calculating means for calculating, a significant pixel number calculating means for counting pixels in which the difference value output from the calculating means is larger than a first predetermined threshold, and a significant pixel number output from the significant pixel number calculating means. , Third
4. A vector quantum according to claim 3, further comprising: a transition image detection signal generating means for judging the period as a transition image period and outputting a transition image detection signal when a frame period larger than a predetermined threshold is generated. Coding device. 5. A vector quantization coding apparatus for coding an input video signal, wherein a codebook size determining means for evaluating a number of significant frames having a large difference from a previous frame from the input video signal to determine a codebook size. A codebook generating means for generating a codebook according to the codebook size information from the codebook size determining means; a codebook storing means for storing a codebook generated by the codebook generating means; A vector quantization encoding device comprising: a quantization encoding unit that quantizes the input video signal using a codebook stored in a codebook storage unit. 6. The codebook size determining means includes: a previous frame storing means for storing a video signal of a previous frame; and a difference for each pixel between an input video signal and a previous frame stored in the previous frame storing means. Calculating means for calculating
Significant pixel number calculating means for counting the pixels for which the difference value output from the calculating means is greater than a first predetermined threshold; and the number of significant pixels output from the significant pixel number calculating means being a fourth predetermined threshold value. Significant frame number calculating means for counting a larger number of frames, Codebook size calculating means for determining a codebook size according to the number of significant frames output from the significant frame number calculating means, and outputting codebook size information. The vector quantization coding apparatus according to claim 5, further comprising: