JP2000287211A - Image compressor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain an image compressor where degradation in the image quality can be prevented. SOLUTION: A data converter 10 converts coded data in compliance with the motion JPEG into coded data in compliance with the MPEG. The data converter 10 is provided with a motion compensation prediction section 60 that preliminarily detect a motion in a prescribed area prior to a motion retrieval (motion detection) by an extended motion compensation prediction section 62 of an MPEG encode section 30 provided to the data converter and an object area for motion detection by the extended motion compensation prediction section 62 is set on the basis of the result of motion detection by the motion compensation prediction section 60.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image compressing apparatus, and more particularly, to detecting a motion of moving image data to be compressed and generating information indicating a motion state based on the result of the motion detection. The present invention also relates to an image compression device for compressing the moving image data based on information indicating the state of the movement.

[0002]

2. Description of the Related Art In recent years, a moving image compression method (encoding method) has been proposed.
MPEG (Moving Picture coding Experts Grou
p) and Motion JPEG (Joint Photographic codi)
ng Experts Group) is widely used.

For example, as shown in FIG. 5A, a moving image is composed of three images 80A, 80B, and 80C representing the movement of a person, and when the image 80B is composed based on the image 80A, the figure As shown in FIG.
0A and a pixel (in this case, a pixel indicating a person) moving between the image 80A and the image 80B are detected (hereinafter, referred to as motion detection), and the moving pixel in the image 80A is shifted by this motion amount (vector amount). Do it.

[0004] Conversely, an image 80B can be constructed if the image 80A has information indicating the amount of the movement. MPEG uses this to compress image data.

In the MPEG, the information indicating the amount of motion is referred to as motion vector data. The generation of the motion vector data is performed in an area of an image 80B corresponding to a predetermined area of the image 80A and in the vicinity thereof. This is performed by searching for a pixel group that is substantially the same as the pixel group in the predetermined area at 80A.

[0006] In MPEG, the signal-to-noise ratio (Sign
To improve the al to noise ratio (SNR), the image is divided into macroblocks of 16x16 pixels, and 0.5 pixels (= average between pixels) are prepared for each macroblock. Thus, the accuracy of the motion vector data is improved.

[0007] On the other hand, motion JPEG is a method for realizing moving image display by continuously reproducing color still images compressed based on JPEG, which is an international standard for color still image coding. It is easier to digitize a video than an encoding method having a simple compression algorithm, and it is easy to perform processing for each frame of an image in an editing operation or the like. However, the compression ratio is not so high.

For this reason, there is an increasing demand for a data converter for converting encoded data based on Motion JPEG into encoded data based on MPEG having a higher compression rate.
This type of conventional data conversion device is generally configured as shown in FIG.

That is, the conventional data converter 90 shown in FIG. 1 includes a variable length decoding unit 22, an inverse quantization unit 24, and an inverse D
JPEG decoding section 20 including CT section 26
And the rearrangement unit 32, the operation unit 34, the DCT unit 36, the quantization unit 38, the variable length encoding unit 40, and the Huffman table 4.
2. MPEG encoding unit 30 including inverse quantization unit 44, inverse DCT unit 46, operation unit 48, motion compensation prediction unit 50, buffer 52, and bit rate control unit 54
And

[0010] In the data conversion device 90, motion JPEG encoded data is decoded by the JPEG decoding unit 20 and output as decoded image data.
It had been converted to PEG encoded data.

[0011]

However, in the conventional data converter, the JPEG decoding unit and the MPE
Because the G encoding unit was configured independently,
Due to the restriction of encoding by the EG, there is a problem that a moving image having a fast motion cannot be predicted with high accuracy, and the image quality deteriorates. This will be specifically described as follows.

It takes an enormous amount of time to encode moving image data in real time using MPEG. That is, for example, if the difference S between the image 80A and the image 80B in FIG. 5 is obtained by an absolute value, the difference S is obtained by the following equation (1).

S = (BA) ＾ 2 (1) Here, B indicates a pixel in a predetermined area of the image 80B, and A indicates a pixel corresponding to the pixel B of the image 80A. That is, the number of operations here is two.

The image 80A is divided into macroblocks of 16 × 16 pixels, and the range of motion search is up, down, left, and right.
When two pixels are set, the number of pixels to be searched is 65/2 (= (+
32-(− 32) +1) × (+32 − (− 32) +
1)), and the image signal is 720 pixels × 480 lines ×
In the case of 30 frames, 720 pixels × 480 lines are 45 × 30 macroblocks, so the amount of calculation required per second is about 87.6 billion times (= 2 × 65 ＾ 2 × 45 × 30 ×
30 × 16 × 16). Needless to say, this value increases as the search range is increased.

As described above, since a large amount of time is required for encoding by MPEG, the search range is usually limited to about ± 128 pixels. Therefore, a moving image having a large amount of motion cannot be completely searched and the data amount increases or the SNR increases. Is deteriorated (image quality is deteriorated).

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and has as its object to provide an image compression apparatus capable of preventing a decrease in image quality.

[0017]

According to a first aspect of the present invention, there is provided an image compression apparatus, wherein a motion detecting means detects a motion of moving image data to be compressed by a motion detecting means. An image compression apparatus that generates information indicating a state of motion based on the motion information, and compresses the moving image data based on the information indicating the state of motion. Preliminary motion detection means for performing preliminary motion detection in a predetermined area prior to detection is provided, and a target area for motion detection by the motion detection means is set based on a result of motion detection by the preliminary motion detection means. You.

As described above, according to the image compression apparatus of the first aspect, the preliminary motion detection means for performing preliminary motion detection in a predetermined area prior to the motion detection by the motion detection means is provided. Since the target area of the motion detection is set based on the result of the motion detection by the preliminary motion detection means, even if the motion of the moving image to be compressed is fast, it is possible to follow the fast motion, Therefore, the SNR can be improved, and as a result, a decrease in image quality in image compression can be prevented.

According to a second aspect of the present invention, as in the image compression apparatus according to the first aspect, the motion detecting means detects a moving object as a result of the motion detection by the preliminary motion detecting means. It is preferable that the moving object and its surroundings are set as a motion detection target region, and if no moving object is detected, a region outside the predetermined region is set as a motion detection target region.

That is, when a moving object is detected by the preliminary motion detecting means, the moving object and its surroundings are searched in more detail by the motion detecting means. Information that can be generated, and if no moving object is detected by the preliminary motion detecting means, the motion detecting means detects motion in an area outside the predetermined area which is the target of motion detection by the preliminary motion detecting means. Since the detection is performed, it is possible to detect a moving object in a short time and with high accuracy, and as a result, it is possible to prevent a decrease in image quality in image compression.

In the image compression apparatus according to the third aspect,
The information indicating the state of the motion generated based on the result of the motion detection by the motion detecting means is fed back to the preliminary motion detecting means, and the feedback is performed by the preliminary motion detecting means. A motion detection area is set based on the information indicating the state of the motion.

Thus, according to the image compression apparatus of the third aspect, the information indicating the state of the motion generated based on the result of the motion detection by the motion detecting means is fed back to the preliminary motion detecting means, In the motion detecting means, since the area of motion detection is set based on the information indicating the state of the motion fed back, it is possible to improve the probability of detecting a moving object in a moving image,
As a result, a decrease in image quality in image compression can be further suppressed.

In general, it is known that a moving image reproduced by information indicating a motion state obtained by a motion search in a frequency domain has an excellent visual SNR.

Therefore, in the image compression apparatus according to the fourth aspect, in the invention according to any one of the first to third aspects, the preliminary motion detection by the preliminary motion detection means is performed in the frequency domain.

As described above, according to the image compression apparatus of the present invention, since the preliminary motion detection means performs preliminary motion detection in the frequency domain, it can be compared with the case where motion detection is performed in the real space domain. To reduce image quality in image compression.
It can be more suppressed.

[0026]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Referring to the drawings, an embodiment in which an image compression apparatus according to the present invention is applied to a data conversion apparatus for converting encoded data according to motion JPEG into encoded data according to MPEG. This will be described in detail.

[First Embodiment] First, referring to FIG.
The configuration of the data conversion device 10 according to the first embodiment will be described. As shown in the figure, a data conversion device 10 according to the first embodiment decodes encoded data based on motion JPEG and outputs the decoded data as decoded image data.
It includes an EG decoding unit 20 and an MPEG encoding unit 30 that MPEG-decodes the decoded image data output by the JPEG decoding unit 20 and outputs it as encoded data.

The JPEG decoding unit 20 is provided with a variable length decoding unit 22 for Huffman decoding input data, and an output terminal of the variable length decoding unit 22 performs inverse quantization on the input data. It is connected to an input terminal of an inverse discrete cosine transform unit (hereinafter, referred to as an inverse DCT unit) 26 that performs an inverse discrete cosine transform operation on the input data via an inverse quantization unit 24.

On the other hand, the MPEG encoding unit 30 includes a rearrangement unit 32 for rearranging the input image data so that the operation for compressing the input image data into each of I, P, and B pictures of MPEG can be conveniently performed. The input terminal of the rearrangement unit 32 is connected to the output terminal of the inverse DCT unit 26, and one output terminal of the rearrangement unit 32 is connected to the plus input terminal of the operation unit 34.

An output terminal of the arithmetic unit 34 is a discrete cosine transform unit (hereinafter referred to as a DCT unit) that divides the input image data into blocks of 8 × 8 pixels, performs a discrete cosine transform operation for each block, and converts it into a spectrum. Through 36)
The input data (DCT coefficient) is connected to an input terminal of a quantization unit 38 for independently quantizing a DC component (DC component) and an AC component (AC component) to remove a high-frequency component.

The output terminal of the quantization unit 38 is branched into two, and one is connected to the input terminal of a variable length coding unit 40 for reducing the amount of data by Huffman coding the input data based on the Huffman table 42. The other is connected, and the other is an inverse quantization unit 44 that performs inverse quantization on the input data, and an inverse discrete cosine transform unit (hereinafter, referred to as an inverse DCT unit) that performs an inverse discrete cosine transform operation on the input data. ) 46 are sequentially connected to one input terminal of a calculation unit 48.

On the other hand, the other output terminal of the rearranging section 32 is J
It is connected to one input terminal of a motion compensation prediction unit 60 that performs motion compensation prediction on the decoded image data obtained by the PEG decoding unit 20.

The output terminal of the motion compensation prediction unit 60 and the output terminal of the calculation unit 48 are connected to the respective input terminals of the extended motion compensation prediction unit 62 that performs motion compensation prediction on the image data input via the calculation unit 48. The output terminals of the extended motion compensation prediction unit 62 are branched into three, and the other input terminal of the motion compensation prediction unit 60, the minus input terminal of the operation unit 34, and the other end of the operation unit 48 are connected to each other. Connected to input terminal.

On the other hand, the output terminal of the variable length coding unit 40 is connected to a buffer 52 for temporarily storing coded data, and one output terminal of the buffer 52 is a bit for controlling a transfer rate of the coded data. It is connected to the quantization unit 38 via a rate control unit 54.

Further, the other output terminal of the buffer 52 is connected to a terminal of the data converter 10 for outputting encoded data.

The motion compensation prediction section 60 corresponds to the preliminary motion detection means of the present invention, and the enlarged motion compensation prediction section 62 corresponds to the motion detection means of the present invention.

Next, the operation of the data converter 10 according to the first embodiment will be described with reference to FIGS.

First, when the coded data according to the motion JPEG to be converted is input to the variable length decoding unit 22, the variable length decoding unit 22 performs variable length decoding on the input coded data and performs variable length decoding. Decoding the encoded data into an inverse quantization unit 24
After inverse quantization, the inverse DCT unit 26 performs an inverse discrete cosine transform operation to obtain decoded image data.

The decoded image data is input to the rearranging section 32 and rearranged by the rearranging section 32.
DCT unit 3 for the decoded image data corresponding to the picture
6, a discrete cosine transform operation is performed, and then a quantization operation is performed by the quantization unit 38 to remove high-frequency components.

Thereafter, the data from which the high-frequency components have been removed by the quantization unit 38 are Huffman-encoded by the variable-length encoding unit 40, sequentially stored in the buffer 52, and output as I-picture encoded data.

Thereafter, in the data converter 10 according to the present embodiment, the inverse quantization unit 44, the inverse DCT unit 46, and the arithmetic unit 4
8 and the motion compensation prediction unit 60, the motion detection of the decoded image data corresponding to the P picture and the B picture is performed as follows.

The image data after the rearrangement by the rearrangement unit 32 is input to the motion compensation prediction unit 60. The motion compensation prediction unit 60 converts the input image data 72 into a predetermined size as shown in FIG. (16 × 16 pixels in this embodiment)
, A motion search (motion detection) of 3 × 3 macroblocks, that is, nine macroblocks is performed.

When a moving object is detected by this motion search, the motion compensation prediction unit 60 generates motion vector data 74 based on the detection result and outputs the motion vector data 74 to the enlarged motion compensation prediction unit 62.

When the motion vector data 74 is input from the motion compensation prediction unit 60 to the expanded motion compensation prediction unit 62,
The position of the moving object and its surroundings obtained by the motion vector data 74 are searched in more detail as shown in FIG. 2B to generate more accurate motion vector data.

On the other hand, when no motion is detected by the search in the motion compensation prediction unit 60, the motion compensation prediction unit 60 displays information 75 indicating that the motion is not detected.
To the expanded motion compensation prediction unit 62.

When the expanded motion compensation prediction unit 62 receives the information 75 indicating that no motion is detected, the motion search range is extended to the macroblocks outside the nine macroblocks as shown in FIG. A motion search is performed by enlargement, and motion vector data is generated based on the result of the motion search. In this case, the accuracy of the motion vector data is reduced and the SNR is degraded as compared with the case where a moving object is detected by the motion compensation prediction unit 60. However, in this case, the motion changes drastically. This deterioration of the SNR is not visually noticeable.

When the expanded motion compensation prediction unit 62 receives the information 75 indicating that the motion vector is not detected, the motion vector data 76 obtained as a result of the motion search is output to the motion compensation prediction unit 60 ( give feedback.

In the motion compensation prediction unit 60, the motion vector data 7 fed back from the expanded motion compensation prediction unit 62
6, the search is performed by expanding the range of the motion search. As a result, the hit rate (the probability that a moving object is detected) in the search by the motion compensation prediction unit 60 can be increased.

As described above, the MPEG encoding unit 30
Inverse quantization unit 44, inverse DCT unit 46, operation unit 4
8, and a moving system in the image is detected by the feedback system configured by the enlarged motion compensation prediction unit 62, and after the motion vector data is obtained, the image data corresponding to the motion vector data is a P picture and a B picture. The image data corresponding to the picture is subtracted from the image data by the operation unit 34, and only less image information is obtained by the DCT unit 36.
The data is sent to an encoding path including a quantization unit 38 and a variable-length encoding unit 40, and encoded data of P and B pictures is obtained. Therefore, the coded data of the P picture and the B picture is a combination of the coded data output from the buffer 52 and the motion vector data obtained by the extended motion compensation prediction unit 62.

On the other hand, in MPEG, the amount of data changes depending on the type of image data to be encoded and the like. Therefore, it is preferable to control the transfer rate of the encoded data to a desired rate.

Therefore, the data conversion device 1 according to the present embodiment
In the case of 0, the amount of encoded data stored in the buffer 52 within a predetermined time is detected by the bit rate control unit 54 and fed back to the quantization unit 38. That is,
When the data transfer rate is high, the degree of quantization by the quantization unit 38 is increased to reduce the amount of data. Conversely, when the transfer rate is low, the degree of quantization by the quantization unit 38 is reduced to reduce the data amount. Increase.

The above processing is repeatedly performed on all the input encoded data of motion JPEG, thereby converting all the encoded data into the encoded data of MPEG.

As described in detail above, in the data conversion apparatus according to the first embodiment, the motion compensation prediction which performs preliminary motion detection in a predetermined area prior to motion detection in motion prediction performed at the time of MPEG compression. And the target area of the motion detection performed at the time of the MPEG compression is set based on the result of the motion detection by the motion compensation prediction unit. Even in this case, it is possible to follow the fast movement, so that it is possible to improve the SNR, and as a result, it is possible to prevent a decrease in image quality in MPEG compression.

In the data converter 10 according to the first embodiment, the motion vector data generated based on the result of the motion detection by the enlarged motion compensation prediction unit is fed back to the motion compensation prediction unit, Since the motion detection target area is set based on the fed back motion vector data, the probability of detecting a moving object in a moving image can be improved. As a result, MPEG A decrease in image quality in compression can be further suppressed.

[Second Embodiment] Generally, it is known that a motion search in the frequency domain and a moving image reproduced by the motion vector data obtained by the motion search have an excellent visual SNR. Have been. So this book 2
In the embodiment, a description will be given of a mode in which the JPEG decoding unit 20 performs motion prediction in the frequency domain in order to utilize the above properties.

First, the configuration of the data converter 10 'according to the second embodiment will be described with reference to FIG. Note that the same parts as those in FIG. 1 of FIG. 3 are denoted by the same reference numerals, and description thereof will be omitted.

As shown in FIG. 3, the data conversion device 10 'according to the second embodiment differs from the data conversion device 10 according to the first embodiment in that the MPEG encoding unit 30 has the motion compensation prediction unit 60, Quantization unit 44 and inverse DCT unit 4
MPEG encoding unit 3 in which the rearrangement unit 64 and the quantizing block motion compensation prediction unit 66 are added.
The only difference is that it is 0 '.

The input terminal of the rearranging unit 64 is connected to the output terminal of the variable length decoding unit 22 in the JPEG decoding unit 20, and the output terminal of the rearranging unit 64 is connected to one input terminal of the quantization block motion compensation prediction unit 66. The output terminal of the quantized block motion compensation prediction unit 66 is connected to the input terminal of the enlarged motion compensation prediction unit 62. The other input terminal of the quantized block motion compensation prediction unit 66 is connected to the output terminal of the enlarged motion compensation prediction unit 62.

Data converter 1 according to the second embodiment
At 0 ′, the data in the frequency domain obtained by the variable length decoding unit 22 is rearranged by the rearrangement unit 64 that rearranges the images in the same manner as the rearrangement unit 32. By this rearrangement, the frequency domain data output from the variable length decoding unit 22 is input to the quantized block motion compensation prediction unit 66 in the same order as the decoded image data output from the rearrangement unit 32.

The quantized block motion compensation prediction unit 66 performs a motion search on the input frequency domain data in the same manner as the motion compensation prediction unit 60 of the first embodiment, and detects a moving object by this motion search. If so, the motion vector data 74 is generated based on the detection result and output to the extended motion compensation prediction unit 62.

In the extended motion compensation prediction unit 62, the motion vector data 74
Is input, the position of the moving object obtained by the motion vector data 74 and its surroundings are shown in FIG.
More detailed search is performed as shown in (B) to generate more accurate motion vector data.

On the other hand, the quantization block motion compensation prediction unit 66
In the case where no moving object is detected by the search in, the quantized block motion compensation prediction unit 66 sends information 75 indicating that no motion is detected to the expanded motion compensation prediction unit 6.
Output to 2.

When the expanded motion compensation prediction unit 62 inputs the information 75 indicating that no motion detection has been performed, the motion search range is extended to the macroblocks outside the nine macroblocks as shown in FIG. A motion search is performed by enlargement, and motion vector data is generated based on the result of the motion search. In this case, the accuracy of the motion vector data is reduced and the SNR is degraded as compared with the case where the moving block is detected by the quantization block motion compensation prediction unit 66. Because there is
This deterioration of the SNR is not visually noticeable.

When the expanded motion compensation prediction unit 62 inputs the information 75 indicating that no detection has been made,
The motion vector data 76 obtained as a result of the motion search is output (feedback) to the quantized block motion compensation prediction unit 66.

The quantized block motion compensation prediction section 66 performs a search by expanding the range of motion search based on the motion vector data 76 fed back from the enlarged motion compensation prediction section 62. As a result, the hit rate in the search performed by the quantized block motion compensation prediction unit 66 can be increased.

The quantized block motion compensation prediction section 66 corresponds to the preliminary motion detection means of the present invention, and the enlarged motion compensation prediction section 62 corresponds to the motion detection means of the present invention.

As described in detail above, in the data converter according to the second embodiment, the quantization block for performing preliminary motion detection in a predetermined area prior to motion detection in motion prediction performed in MPEG compression. Since the motion compensation prediction unit is provided and the target area of the motion detection performed at the time of the MPEG compression is set based on the result of the motion detection by the quantization block motion compensation prediction unit, the same as in the first embodiment described above. In addition, even when a moving image to be compressed has a fast movement, the moving image can follow the fast movement, so that the SNR can be improved, and as a result, a decrease in image quality in MPEG compression can be prevented. be able to.

In the data converter according to the second embodiment, preliminary motion detection in the quantized block motion compensation prediction unit is performed in the frequency domain. As a result, it is possible to further suppress the deterioration of the image quality, and therefore it is possible to reduce the number of the inverse quantization unit and the inverse DCT unit in the MPEG encoding unit as compared with the first embodiment. Can be realized.

In the second embodiment, the case where the inverse quantization unit 44 and the inverse DCT unit 46 shown in FIG. 1 are reduced by performing motion detection in the frequency domain by the quantization block motion compensation prediction unit 66 will be described. But
The present invention is not limited to this, and in order to obtain coded data with better SNR visually,
As shown in FIG. 4, it is needless to say that the inverse quantization unit 44 and the inverse DCT unit 46 can be applied.

[0070]

As has been described in detail above, according to the present invention, there is provided preliminary motion detection means for performing preliminary motion detection in a predetermined area prior to motion detection by the motion detection means. Since the target area of the motion detection is set based on the result of the motion detection by the preliminary motion detection means, even if the motion of the moving image to be compressed is fast, it can follow the fast motion. SN
R can be improved, and as a result, it is possible to prevent the image quality from deteriorating in image compression.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a schematic configuration of a data conversion device according to a first embodiment.

FIG. 2 is a schematic diagram for explaining the operation of the data conversion device according to the first embodiment.

FIG. 3 is a block diagram illustrating a schematic configuration of a data conversion device according to a second embodiment.

FIG. 4 is a block diagram showing a schematic configuration of another embodiment of the data conversion device according to the second embodiment.

FIG. 5 is a schematic diagram for explaining MPEG.

FIG. 6 is a block diagram illustrating an example of a schematic configuration of a conventional data conversion device.

[Explanation of symbols]

10, 10 ', 10''data conversion device 20 JPEG decoding unit 22 variable length decoding unit 24 inverse quantization unit 26 inverse DCT unit 30, 30' MPEG encoding unit 32 rearrangement unit 36 DCT unit 38 quantization unit 44 inverse quantum Transformation unit 46 Inverse DCT unit 60 Motion compensation prediction unit (preliminary motion detection unit) 62 Enlarged motion compensation prediction unit (motion detection unit) 64 Rearrangement unit 66 Quantized block motion compensation prediction unit (preliminary motion detection unit) 70 Macroblock 76 Motion vector data (information indicating the state of motion)

Claims (4)

[Claims] 1. A motion detection unit performs motion detection on moving image data to be compressed, generates information indicating a motion state based on a result of the motion detection, and generates information based on the information indicating the motion state. An image compression apparatus for compressing the moving image data, comprising: preliminary motion detection means for performing preliminary motion detection in a predetermined area prior to motion detection by the motion detection means; An image compression apparatus for setting a target area for motion detection based on a result of motion detection by a detection unit. 2. When the motion detection means detects a moving object as a result of the motion detection by the preliminary motion detection means, the moving object and its surroundings are set as the motion detection target area. 2. The image compression apparatus according to claim 1, wherein when no moving object is detected, an area outside the predetermined area is set as the motion detection target area. 3. The motion indicating the state of the motion generated based on the result of the motion detection by the motion detecting means is fed back to the preliminary motion detecting means. The image compression apparatus according to claim 1 or 2, wherein a motion detection area is set based on information indicating the state of (1). 4. The apparatus according to claim 1, wherein said preliminary motion detection means performs said preliminary motion detection in a frequency domain.
The image compression apparatus according to claim 3.