JPH0998418A - Method and system for encoding and decoding picture - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image encoding/decoding method for preferentially handling an observation area while maintaining consistency with an international standard. SOLUTION: A position parameter for correcting a quantization scale code corresponding to observation density distribution in pictures is stored in a table 25. A contour extraction part divides an image supplied from a camera 11 into fine blocks and discriminates whether or not the contour of the image which is an observation part is present inside the respective blocks. A quantization part 22 corrects the quantization scale code based on the position parameter read from the table 25 and the presence/absence of the contour discriminated by the contour extraction part 24 corresponding to the position of the image to be compressed and quantizes the observation area by a small quantization scale code and a non-observation area by a large quantization scale code. Thereafter, a variable length encoding processing is executed and encoded data are generated. The quantization scale code used at the time of quantization is used for a decoding processing.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding / decoding technique, and more particularly to an image coding / decoding technique capable of decoding a high quality image at a low bit rate.

[0002]

2. Description of the Related Art Image coding (compression) / decoding techniques have received attention as key technologies in image communication systems. In the image coding (compression) / decoding technology, there is a problem that the coding rate must be increased while enabling the decoding of high quality images. As a method for solving this problem, a technique has been proposed in which a gaze area of a display image is preferentially treated, and non-linear sampling (decimation) is performed from the central portion of the image for transmission.

[0003]

However, in the method of nonlinearly sampling and transmitting from the center of the screen, it is necessary to interpolate the thinned region on the decoding side, which complicates control.
Further, since the sequence is transmitted from the center of the screen, there is a large gap from the sequence of the MPEG system, which is an international standard for moving picture encoding / decoding methods, and it is not possible to achieve consistency with existing products.

The present invention has been made in view of the above situation, and an object thereof is to provide an image encoding / decoding method and system which preferentially treats a gaze area while maintaining consistency with an international standard. And It is another object of the present invention to provide an image coding / decoding method and system capable of decoding higher quality images at the same bit rate.

[0005]

In order to achieve the above object, the image coding system according to the first aspect of the present invention includes an input unit for inputting an image and an image input by the input unit. A discriminator that discriminates a region to be gazed, and a first quantization scale code is used for the gaze region discriminated by the discriminator, and a second quantum that is larger than the first quantization scale code is used for a non-gaze region. It is characterized by comprising a quantizing means for quantizing the image using a quantization scale code.

Further, in the image coding system according to the second aspect of the present invention, the first correction amount of the quantization scale code based on the spatial distribution density of the gazing point is set for each minute region of the quantization target image. For setting the second correction amount of the quantization scale code based on the gaze time for each minute area of the quantization target image, and the quantization scale code of the quantization target image for the first area. It is characterized by comprising a quantizing means for performing a correction based on the correction amount and the second correction amount, and for quantizing the quantization target image using the corrected quantization scale code.

In the image coding method according to the third aspect of the present invention, the spatial distribution density of the gazing points on the image to be coded is discriminated, and the distribution of the gaze time on the image to be coded is discriminated. Based on the determined spatial distribution density and the distribution of the gaze time, finely quantize a region with a high degree of gaze in the target image for encoding, and roughly quantize a region with a low degree of gaze in the target image for encoding. To do.

The gazing point spatially has a distribution close to a normal distribution centered on the center of the screen. In addition, the gaze time of a complicated portion of an image or a portion that is difficult to recognize becomes long. From these spatial and temporal gaze degrees, the high gaze degree is finely quantized using a small quantization scale code,
By coarsely quantizing the low attention part using a large quantization scale code, it is possible to encode / decode an image of substantially the same quality as a conventional one at a low bit rate. It is possible to encode / decode an image of substantially higher quality than ever before.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, the present invention will be described with reference to FIG.
An embodiment of an encoding / decoding system applied to TM5 which is a test model of MPEG-2 which is an international standard of decoding technology will be described with reference to FIGS. In addition, TM
For details of 5, refer to Test Model Editing Commitee: "Test Mod
el 5 ", defined in ISO / IEC JTC1 / SC29 / WG11 N0400 (Apr.1993).

As shown in FIG. 1, this encoding / decoding system includes a camera 11, an encoding unit 12, and a decoding unit 1.
3 and a display 14.

The camera 11 outputs an image signal corresponding to the amount of input light. The coding unit 12 includes a conversion unit 21, a quantization unit 22, a variable length coding unit 23, a contour extraction unit 24, and a position parameter table 25. This configuration corresponds to a configuration in which the contour extraction unit 24 and the position parameter table 25 are added to the TM5 encoding unit.

The conversion unit 21 performs motion compensation prediction (if necessary) and DC on the image data supplied from the camera 11.
A conversion process such as T (Discrete Cosine Transform) is performed to convert the image data to be quantized. The quantization unit 22 corrects the quantization scale code using the data indicating the contour portion of the image supplied from the contour extraction unit 24 and the position parameter supplied from the position parameter table 25, and the corrected quantization is performed. An adaptive quantization process of MPEG-2 is executed using the scale code. A method of correcting the quantization scale code will be described later. The variable length coding unit 23
The image data quantized by the quantization unit 22 is subjected to variable length coding processing to obtain coded data. The encoded data thus obtained is transmitted via a transmission line,
Alternatively, it is stored in the storage device.

The contour extraction unit 24 has a frame memory, stores at least one input image, determines the position of the contour in the image, and supplies it to the quantization unit 22.
A method of discriminating the contour will be described later. The position parameter table 25 stores a correction value for correcting the quantization scale code for each position of the image, and supplies it to the quantization unit 22. The method of setting the correction value will be described later.

The decoding unit 13 is a circuit for decoding the encoded data received or read from the storage device, has the substantially same configuration as the conventional decoding unit, and has a variable length decoding. Conversion section 31, inverse quantization section 32, and inverse conversion section 33
With.

The variable length decoding unit 31 performs variable length coding processing on the coded data and decodes the quantized image data and the quantized scale code. The inverse quantizer 32
The decoded quantized scale code is discriminated, and the received image is dequantized using the quantized scale code. The inverse transforming unit 33 transforms the inversely quantized image data into the transforming unit 2
The inverse transform of the transform performed by 1 is performed, that is, the inverse DCT transform process is performed and the reverse prediction of the motion compensation prediction (when the motion compensation prediction is performed) is performed.

The display (display device) 14 is composed of a CRT which displays gradations corresponding to the supplied image data.

The correction of the quantization scale code is executed by a two-step process of correction based on the spatial distribution (distribution density) of the gazing points and correction based on the temporal distribution of the gazing points (gaze time distribution). To be done. The distribution density of spatial gazing points is
As shown by the shaded area in FIG. 2, the distribution is close to the normal distribution centered on the center of the screen. Therefore, set a position parameter that gradually decreases with distance from the center of the screen,
The position parameter is multiplied by the quantization scale code to correct the quantization scale code. As a result, the central portion of the screen where the distribution density is high is quantized more finely, and the edge of the screen where the distribution density is low is quantized more coarsely. The position parameter thus obtained is set in the position parameter table 25 and referred to at the time of quantization.

A method of setting the position parameter will be specifically described by taking an image of the NTSC system having an aspect ratio of 3: 4 as an example. As shown in FIG. 2, the screen is logically 16 × 1 vertically and horizontally.
Divide into 6-dot minute blocks (hereinafter referred to as microblocks). Then, from the center of the screen to each microblock, 0.6 to 1.3 in the horizontal direction and 0.8 to 1.3 in the vertical direction.
Values are evenly distributed in each range. Next, the horizontal setting value and the vertical setting value are averaged for each microblock, and this is set as a position parameter (correction value).

Therefore, for example, in the micro block at the upper right corner of the screen of FIG. 2, 1.3 (=
(1.3 + 1.3) / 2) is allocated, and 0.7 (= (0.6 + 0.8) / is assigned to the microblock in the center of the screen.
2) is assigned. The values of the positional parameters assigned to each μ block are shown in FIG.

The position parameter set in this way is set in the position parameter table 25, the corresponding position parameter is read from the position parameter table 25 at the time of quantization, multiplied by the quantization scale code, and the corrected quantization scale. Quantize using code.

The gazing point has both a spatial distribution and a temporal aspect. The gaze time for a portion having a large amount of information or a portion that is difficult to recognize becomes long. Here, the contour of the image (object) is taken up as a factor that prolongs the gaze time from such requirements, and the contour scale image is finely quantized by correcting the quantization scale code to a small value in the contour portion, and the non-contour portion is Then, the quantization scale code is corrected to a large value, and the outline image is roughly quantized. In general, the contour portion has more information than the peripheral portion. Therefore, determining the contour part and correcting the quantization scale code of the contour part to a small value determines the information amount of each part of the image to be coded, and the quantization scale of the region with more information amount than the peripheral part. It also means that the code is made smaller and finely quantized.

Next, a method of extracting the contour portion and correcting the quantized scale code will be concretely described. The contour in the image (object) is extracted based on the luminance signal Y. First, the luminance signal Y is binarized. The threshold value for binarization is a value in which a histogram with the pixel value on the horizontal axis is generated and the ratio of white pixels to black pixels is, for example, 9: 1. The value of this ratio is preferably set so that the contour is included in about half of the microblock.

Regarding the binarized image data, the length and width are 3
× 3 pixel image data is sequentially extracted, and 3 ×
3 Create a table. (A) If the central pixel of 3 × 3 pixels is black, the output is black regardless of the values of the remaining 8 pixels. (B) If the central pixel of the 3 × 3 pixels is white and the remaining 8 pixels are all white, the output is black. (C) In all cases except (a) and (b), the output is white. By performing such processing, it is determined that the contour portion of the display image is located in the microblock in which the pixel determined to be black exists.

By such processing, the standard image C
The processing result of processing the image of IF "Miss Ameriaca" (352 × 288 pixels) is shown in FIG. In FIG. 3, a microblock with a halftone line is a block determined to include a contour. It can be confirmed from FIG. 3 that the contour can be properly discriminated.

The quantization scale code for quantizing the image in the contour-extracted microblock is multiplied by 0.8, and the quantization scale code for quantizing the image of the microblock in which the contour is not extracted is set to 1. .2 times correction, and correction based on the gaze time is realized.

Finally, the position parameter and the correction value 0.8
Alternatively, the quantization scale code is multiplied by 1.2 to obtain the corrected quantization parameter, and the image is quantized using the corrected quantization scale code.

Next, the process of acquiring an image, encoding it, decoding it, and displaying it will be described. First, the image acquired by the image pickup unit of the camera 11 is sequentially converted into digital data, supplied to the conversion unit 21 of the encoding unit 12, and subjected to motion compensation prediction (for necessary frames), DCT conversion processing, and the like. . On the other hand, the image data supplied from the camera 11 is also supplied to the contour extraction unit 24 and stored in the frame buffer. The contour extraction unit 24 uses the 3 × 3 image data stored in the frame buffer as described above.
A table is created, black pixels are detected, and the microblock where the contour of the image is located is determined. Contour extraction unit 24
Supplies the position information of the microblock determined to have the contour to the quantizer 22.

The quantizing unit 22 is supplied with the image data converted by the converting unit 21, and the quantizing unit 22 carries out quantization for quantizing the image data according to a processing routine of a normal adaptive quantizing process. Calculate the code scale code. T
In the M5 mode, the quantizer scale code is set so that the flat part where visual deterioration is easily noticed is finely quantized, and the complicated part of the pattern where deterioration is relatively less noticeable is roughly quantized (Nakajima, Otaka, Tahara). : "MPEG coding method-video compression", TV magazine, 49, 4, pp43
5-466 (1991)).

Next, the quantizing unit 22 discriminates the microblock in which the image data to be quantized is located, and the value (0.7) of the positional parameter assigned to the microblock.
~ 1.3) is read from the position parameter table 25 and is multiplied by the quantization scale code.

Further, from the discrimination result of the contour extracting section 24,
It is determined whether or not the contour is located in the microblock where the image data to be quantized is located. If it is located, the quantization scale code is multiplied by 0.8. If it is not located, the quantization scale code is 1. Double. That is, the quantizer 2
In step 2, the quantized scale code obtained in accordance with the processing routine of the normal adaptive quantization processing is multiplied by the correction value based on the position parameter and the presence or absence of the contour, and the quantized scale code is corrected based on the gaze degree.

The quantizing unit 22 quantizes the image converted by the converting unit 21 using the quantization scale code thus obtained.

The variable length coding unit 23 performs variable length coding processing on the quantized image data supplied from the quantization unit 22 and the quantized scale code used for quantization. The variable-length coded image data thus generated is transmitted via the transmission path or stored in the data storage device.

When decoding the variable length coded image data received or read from the storage device, first, the variable length decoding unit 31 decodes the quantized image data and the quantized scale code, and dequantizes them. Supply to the section 32. The inverse quantizer 32 inversely quantizes the quantized image data using the decoded quantization scale code, and the inverse transformer 33 further performs inverse DCT transform on the inversely quantized image data to perform motion compensation. The prediction opposite to the prediction is performed and the image data is decoded. The decoded image data is supplied to the display 14 and displayed.

As described above, according to this embodiment, if the other conditions are the same, a small quantization scale code is assigned to quantize the image data of the area watched by the viewer. The image is finely quantized, and a large quantization scale code is assigned to quantize the image data in the unfocused area, and the image is coarsely quantized. Therefore, the quality of the decoded image obtained by the subjective evaluation can be improved by preferentially assigning the code amount to the gaze area. Further, if the code amount assigned to the non-gaze region is reduced without changing the code amount assigned to the gaze region, it is possible to realize the low bit rate encoding while maintaining the image quality obtained by the subjective evaluation. .

Moreover, since the quantization scale code is set based on the spatial distribution and the temporal distribution of the human gazing point, the moving image can be appropriately encoded / decoded.

The present invention is not limited to the above embodiment, and various modifications and applications are possible. For example, in the above description, the present invention has been described by taking TM5, which is a test model of MPEG-2, which is an international standard for encoding / decoding of moving images, as an example, but encoding any other image data. -Can be applied to decryption processing. Further, the present invention is not limited to moving images, and can be applied to still image encoding / decoding processing. However, in the case of a still image, the influence of the spatial gaze is increased and the influence of the gaze depending on the gaze time is reduced. Further, although the contour portion of the image is selected as the area having a long gaze time, another portion may be selected. Also, other methods of extracting the contour may be used. Further, the size of the microblock and the like are also arbitrary.

Further, not only the normal image of the NTSC system,
The present invention is similarly applicable to the wide mode of the NTSC system and the image of the PAL system. Further, the position parameter is an example in the case of a normal image of the NTSC system, and any other set value can be used. Further, the correction value based on the presence or absence of contour detection is not limited to 0.8 or 1.2, and other values can be used.

For example, in the above embodiment, the transform unit 21 and the inverse transform unit 33 perform the DCT transform and the inverse DCT transform. However, subband coding and subband decoding,
Various transform methods that generate image data suitable for quantization, such as wave red transform and inverse wave red transform, can be used.

Even though the device from the acquisition of the image to the variable length coding and the device from the data reception and display of the variable length coded image are physically separated, Good. Similarly, the image obtained by the camera 11 may be temporarily stored and then supplied to the encoding unit 12. Alternatively, the image data may be generated by a computer or the like and encoded without using the camera 11. You may supply to the conversion part 12.

[0040]

According to the present invention, a small quantization scale code is assigned to the image of the gaze area and a large quantization scale code is assigned to the image of the gaze area. Therefore, at a low bit rate. Images of substantially the same quality as in the past can be encoded / decoded, and if the bit rate is the same, images of substantially higher quality than in the past can be encoded / decoded.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an image encoding / decoding system according to an embodiment of the present invention.

FIG. 2 is a diagram showing an example of a spatial distribution of a gaze area and positional parameters assigned to microblocks.

FIG. 3 is a diagram showing a position of a contour in a display image.

[Explanation of symbols]

 11 camera 12 coding unit 13 decoding unit 14 display 21 conversion unit 22 quantization unit 23 variable length coding unit 24 contour extraction unit 25 position parameter table 31 variable length coding unit 32 dequantization unit 33 inverse transformation unit

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Taichi Nakamura 3-3-3 Toyosu, Koto-ku, Tokyo NTT Data Communications Corporation

Claims (8)

[Claims] 1. An input unit for inputting an image, a discriminating unit for discriminating a gaze region in the image input by the input unit, and a first quantization for the gaze region discriminated by the discriminating unit. And a quantizing means for quantizing the image using a second quantized scale code that is larger than the first quantized scale code for the non-gaze region and uses a scale code. Image coding system. 2. The discriminating means comprises means for discriminating a spatially focused area on the image and means for discriminating a temporally focused area according to a predetermined standard. The image coding system according to claim 1. 3. The means for discriminating the spatially gazed area includes means for discriminating a relative position in the image, and the means for judging the temporal gaze area is an outline position of the image. The image coding system according to claim 2, further comprising means for discriminating. 4. A means for setting a first correction amount of a quantization scale code based on a relative position in the quantization target image for each minute region of the quantization target image, and the quantization target image. Means for discriminating the amount of information in the micro region for each micro region, and setting a second correction amount of the quantization scale code according to the discrimination result; An image code characterized by comprising a quantizing means for performing correction based on the first correction amount and the second correction amount, and quantizing the image to be quantized using the corrected quantization scale code. System. 5. An image coding / decoding system, further comprising means for decoding an image coded by the image coding system according to any one of claims 1 to 4. 6. A spatial distribution density of gaze points on an encoding target image is discriminated, a distribution of gaze times on an encoding target image is discriminated, and based on the discriminated spatial distribution density and gaze time distribution,
Finely quantize the high-gaze area of the image to be encoded,
An image coding method characterized in that a region of a target image to be coded having a low degree of gaze is roughly quantized. 7. The image coding method according to claim 6, wherein the determination of the gaze time distribution is performed by extracting a contour in the image to be coded. 8. An image coding / decoding method, which decodes an image coded by the image coding method according to claim 6.