JP2919990B2 - Image data compression method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data compression system for controlling the amount of code for compression coding of a still image or a moving image, and more particularly to improvement of coding efficiency.

[0002]

2. Description of the Related Art There is a demand that the encoding bit rate be kept at a constant level when compressing and encoding a still image or a moving image, and various methods have been proposed.

For example, “Sc by WHChen, WKPratt
ene Adaptive Coder ”, IEEE Trans. COM-32, No.3 198
4 assigns a uniform amount of code to each block of the image,
By controlling the quantization width of the block using the integrated value of the difference between the actually generated code amount of the block and the allocated code amount as a parameter, the code amount of the block approaches the allocated code amount, and the code amount in one image is reduced. There is disclosed a method of setting a set value.

In Japanese Patent Application Laid-Open No. 2-124690, the number of coded bits is assigned to one part of the screen from the generated information amount (activity) of one part or the whole of the screen.
By encoding within the range of the number of encoded bits,
A technique for keeping the total code amount in a screen constant is disclosed.

Further, Japanese Patent Laid-Open Publication No. Hei 3-16491 discloses a technique in which the quantization step width is changed depending on the amount of information generated in the past block and the occupation amount of the encoded data stored in the buffer. ing.

[0006]

However, the conventional code amount control method as described above, in particular, "Scene Adaptive C
oder "and the method described in JP-A-3-16491, the quantization step width is changed using the past generated code amount as a parameter, and the nature of the part to be coded from now on is ignored. Therefore, in such a method, the quantization step width becomes coarse in a portion where the quantization step width is desired to be reduced, and conversely, the quantization step width becomes small in a portion where the quantization step width may be reduced. There is a problem that the coding efficiency is low due to a decrease.

Further, as disclosed in Japanese Patent Application Laid-Open No. 2-124690, when a code amount is assigned to one area (block), the local property of an image in which each pixel is uniformly quantized is used. This method is effective when one quantization step width is used for one screen. However, when the quantization step width is changed using the past generated code amount as a parameter as in the above-described two methods, the quantization step width similarly changes regardless of the properties of the image, so that the coding efficiency is high. It can not be said.

[0008] Instead of uniformly assigning the code amount to each block as described in the above "Scene Adaptive Coder",
As disclosed in Japanese Patent Application Laid-Open No. 2-124690, it is considered that the coding efficiency is improved by changing the allocated code amount according to the property of each block. However, this is the case where the assigned code amount and the generated code amount match relatively well, and if they do not match, the fluctuation of the quantization step width becomes large conversely, and the code amount is uniformly assigned. Efficiency may be lower than that.

After all, when the code amount is controlled by the buffer occupation amount, the generated code amount is increased in a portion requiring a large code amount in one screen, and the generated code amount is reduced in a portion which may be reduced. It is difficult to control.

The present invention has been made in view of the above problems in controlling the code amount by the buffer occupancy, and
It is an object of the present invention to provide an image data compression system with improved encoding efficiency.

[0011]

In order to achieve the above object, an image data compression system according to the present invention employs a dividing means for dividing an input image into a plurality of areas (a plurality of areas).
Corresponding to the chair coding order determiner 18).
Means for setting importance for multiple areas (actual
(Corresponds to the slice importance setting unit 182 in the embodiment.)
And a buffer for storing encoded data (actual
(Corresponding to the buffer "I" 22 in the embodiment) and the quantum
Occupancy of data occupying the buffer
Means for determining based on the amount (the buffer in the embodiment)
"I" 22), the importance and the quantization
Corresponding to the plurality of divided areas
Encoding order determining means for determining the order of image encoding (implementation
(Corresponding to the slice coding order determiner 18 in the example)
And the order determined by the encoding order determining means.
Encoding an image corresponding to the plurality of divided areas,
Encoding means for generating encoded data (separation in the embodiment)
(Corresponding to the scattered cosine transformer 24, the quantizer 26, etc.)
It is characterized by having .

[0012]

[0013]

According to the image data compression method of the present invention, the encoding order is not the raster scan order from the upper left to the lower right of the image regardless of the size of the quantization scale. By changing the coding order of the divided regions, visually important regions can be coded when the quantization scale is small, and conversely, unimportant regions can be coded when the quantization scale is large. Therefore, coding efficiency is improved.

[0014]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a first example of the image data compression method of the present invention.
FIG. 4 is a block diagram of an embodiment of FIG. Now, it is assumed that the input image is provided as a macroblock as shown in FIG. Such an input image is input to a subtractor 10 and is subtracted from a prediction macroblock output from a motion compensation predictor (MC) 12 as described later. The result of this subtraction is input to the switch 14 and the intra-frame / inter-frame encoding determiner 16. Further, the input image further includes:
The signal is directly input to the switch 14 and the intra-frame / inter-frame coding determiner 16.

The intra-frame / inter-frame encoding determiner 16 determines which of the two types of macroblock data (that is, the input image given via the subtractor 10 and the input image given directly) by one of the statistics. One of the macro blocks is selected, and the selection signal is supplied to the switch 14. Here, the statistic is, for example, a variance value of a pixel value, a sum of absolute values of output values of a bandpass filter, a sum of squares of pixel values, and a value quantized by provisional quantification by performing an orthogonal transform. It is an amount that can be used to determine which macroblock to use, such as the amount of code at the time. The intra-frame / inter-frame coding determiner 16 further determines the statistics of the selected macroblock by a slice coding order determiner 18.
Supply to

On the other hand, the macroblocks that have passed through the switch 14 according to the result of selection by the intra-frame and inter-frame coding determiner 16 are stored in a frame memory (FM).
20 are stored in order, and one image is stored.

In the slice coding order determiner 18, as shown in FIG. 2, several slices (an area where several macro blocks are gathered in one image), and macro blocks in one slice are rasterized. (Connected in the order of scanning), and the divided areas are assigned an order of importance.

The slice encoding order determiner 18 stores the occupation amount of the buffer “I” 22 at the time when the data of one image is accumulated in the frame memory 20.
From the buffer “I” 22, it is determined which slice is to be selected first among the slices having importance levels based on the occupation amount. Thereafter, the slice coding order determiner 18 selects slices to be coded according to the occupation amount in the buffer “I” 22.

FIG. 3 shows the slice coding order determiner 18.
It is a figure showing the content of. That is, the activity per macroblock (sum of absolute differences between adjacent pixels, etc.) output from the intra-frame or inter-frame coding determiner 16 or a quantum weighted on a temporary visual characteristic after orthogonal transformation. The actual code amount using the conversion step width (in either case, hereinafter referred to as the macroblock information amount) is input to the slice boundary setting, information amount calculation, and assigned code amount calculation unit 181. Is done. Here, the slice boundary may be one macroblock line (a state in which the macroblocks are arranged in a row in the horizontal direction of the image) as one slice, or may be divided at a point where the information amount of the input macroblock suddenly changes. May be. Then, the information amount of each divided slice is calculated, and the average information amount per macroblock of each slice is calculated. Further, the average information amount of the entire image is calculated.

These values are set in the slice importance setting section 18.
2 and the importance is set to the slices in the order in which the slice average information amount is large. The quantization scale is predicted from the average information amount of the entire image and the code amount assigned to one image.

The slice importance and the predicted quantization scale from the slice importance setting unit 182 and the buffer occupancy from the buffer “I” 22 are input to a slice encoding order determination unit 183. The slice coding order determination unit 183 determines the order of importance or the order of importance according to the magnitude relationship between the quantization scale determined by the buffer occupancy and the predicted quantization scale as shown in FIG. An address of the frame memory 20 at which the slice starts and a serial number of the slice counted from the upper left of the image in raster scan order are output so as to be sequentially encoded.

Here, a specific slice coding determination order in slice coding order determination section 183 is determined as follows. That is, first, the quantization scale α _B obtained from the buffer occupancy is compared with the prediction scale α _p obtained from the average information amount of one image. If α _B ≧ α _p , the importance is the lowest. Encoding is performed from the slice,
When α _B <α _p , encoding proceeds in order from the slice having the highest importance. If the above conditions are alternately α _B ≧ α _p and then α _B <α _p , α _B ≧ α _p ,.
The encoding is performed in such a manner that the slice having the lowest importance, the slice having the next highest importance, and the non-encoded slice having the lowest importance (that is, the slice having the second lowest importance). Here, the setting of the slice boundary, the calculation of the information amount, and the slice allocation code amount in the allocation code amount calculation unit 201 are performed by keeping the average allocation code amount per macroblock constant or by the slice information amount. Is weighted.

The frame memory 2 indicated by the slice coding order determiner 18 determined as described above
The slice data read from the frame memory 20 is sent to the discrete cosine transformer (DCT) 24 in accordance with the slice start address position of 0, for example, where the transform is performed in units of 8 × 8 blocks, and each transform coefficient is quantized. Quantizer (Q) 26 performs quantization. The result is supplied to a variable length encoder (VLC) 28 for encoding, and is also supplied to an inverse quantizer (Q ^-1 ) 30 for inverse quantization. Note that the present invention is not limited to the discrete cosine transform, and other orthogonal transforms can be used.

The data encoded by the variable length encoder 28 is stored in the buffer "I" 22. Here, the buffer “I” 22 integrates the difference amount between the code amount of the coded slice and the slice allocation code amount sent from the slice coding order determiner 18, The sum is returned to the slice coding order determiner 18 as the buffer occupancy so as to be used for selecting a slice to be sliced. Further, the integrated amount is also sent to the quantizer 26 and the inverse quantizer 30, and the quantization step width in them is changed according to the integrated amount.

Here, the change of the quantization step width is such that an initial quantization step width q _ij is provided for each of the transform coefficients in the block as shown in FIG. _This is performed by uniformly multiplying _ij by the quantization scale α. That is, the quantization step width of each transform coefficient is q _ij × α. The quantization scale is
As shown in FIG. 4A, the quantization scale is changed according to the buffer occupancy of the buffer “I” 22.

As shown in FIG. 5, the buffer "I" 22 stores the coded data of the slice, and stores the start address and code amount of the buffer "I" 22 with respect to the slice number (No.). , Store in another area. When the encoding of one screen is completed, the data of the buffer “I” 22 is transferred to the buffer “II” 32 by referring to the start address and the code amount in ascending order of the slice number. Then, the data is rearranged, and the data in the buffer "II" 32 is output to a transmission path and a storage medium (not shown).

The data inversely quantized by the inverse quantizer 30 is inversely transformed by an inverse discrete cosine transform (IDCT) 34. Here, the inverse transformer is not limited to the inverse discrete cosine transformer, but may be any inverse transformer that forms a pair with the orthogonal transform at the time of conversion. The inversely converted value is calculated by the adder 3
6 and is added to the output value of the switch 38 switched according to the determination result of the intra-frame / inter-frame coding determination unit 16 and stored in a frame memory (not shown) in the motion compensation prediction unit 12. Written. in this case,
The write address of the frame memory is indicated by the output of the slice coding order determiner 18.

Then, by the motion compensation predictor 12,
Next, the image to be encoded and the image stored in the frame memory in the motion compensation predictor 12 are motion-compensated in macroblock units. The motion-compensated prediction macroblock is input to the switch 38 and the subtractor 10 to obtain a difference from the macroblock of the image to be encoded. Thereafter, the above contents are repeatedly processed.

As described above, in the encoding system in which the quantization step width is changed according to the buffer occupancy and the amount of generated code is controlled, the input image is divided into several regions, and the importance is assigned to these regions. Is performed, and the coding order of the regions ordered by the importance is determined according to the magnitude of the quantization scale.When the quantization scale is large, the region of low importance is small, and the quantization scale is small. In some cases, a region having a high degree of importance can be encoded, so that encoding efficiency is good and image quality can be improved.

FIG. 6 is a view showing a second embodiment of the present invention. The same reference numerals as in FIG. 1 denote the same parts, and a description thereof will be omitted. That is, in the second embodiment, the slice number from the slice encoding order determiner 18 at the head (slice header information) of each slice, and thereafter, the encoded data are directly stored in the buffer 40. I try to write. At this time, a decoder (not shown) can generate a decoded image by rearranging the decoded slices based on the slice number.

In this case, in the buffer 40, the difference between the allocated code amount of each slice and the actual code amount is calculated and the amount of the difference is integrated. Decision unit 18, Quantizer (Q) 2
6 and an inverse quantizer (Q ⁻¹ ) 30.

That is, the second embodiment is different from the first embodiment shown in FIG. 1 in that the number of buffers is one and the slice number is transmitted together with the encoded data instead of rearranging the encoded data. This is different from the embodiment of FIG.

According to the second embodiment, similarly to the first embodiment, the input image is divided into several regions, and these regions are ranked in order of importance. The coding order of the regions ordered by the importance is determined according to the size of the quantization scale, so the region of low importance is used when the quantization scale is large, and the region of high importance is used when the quantization scale is small. Can be encoded, and the encoding efficiency can be improved and the image quality can be improved.

That is, according to the present invention, in the coding system in which the quantization step width is changed according to the buffer occupation amount and the generated code amount is controlled, the input image is divided into several regions, and important regions are assigned to these regions. The order of the degrees is determined, and the coding order of the regions ordered according to the importance is determined according to the magnitude of the quantization scale.When the quantization scale is large, the region of low importance is determined, and the quantization scale is determined. When the size is small, a region with high importance can be encoded, the encoding efficiency can be improved, and the image quality can be improved.

[0035]

As described in detail above, according to the present invention,
An image data compression system with improved encoding efficiency can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram of a first embodiment of an image data compression system according to the present invention.

FIG. 2 is a diagram illustrating a configuration of an input image.

FIG. 3 is a block diagram of a slice coding order determiner in FIG. 1;

4A is a graph showing a relationship between a buffer occupancy and a quantization scale, and FIG. 4B is a diagram showing an initial quantization step width for a transform coefficient in a block.

FIG. 5 is an operation block diagram for explaining an operation of a buffer “I” in FIG. 1;

FIG. 6 is a block diagram of a second embodiment of the image data compression system of the present invention.

[Explanation of symbols]

10: subtractor, 12: motion compensation predictor (MC), 14,
38 switch, 16 intra-frame / inter-frame coding determiner, 18 slice coding order determiner, 20 frame memory (FM), 22 buffer "I",
24 discrete cosine transformer (DCT), 26 quantizer (Q), 28 variable length coder (VLC), 30 inverse quantizer (Q ^-1 ), 32 buffer "3
4 ... Inverse discrete cosine transformer (IDCT), 36 ... Adder, 40 ... Buffer.

Claims (6)

(57) [Claims] 1. A dividing method for dividing an input image into a plurality of regions.
And setting importance levels for the plurality of divided areas.
Means for storing encoded data, and a quantization scale for the data occupying the buffer.
Means for determining based on the occupancy, and the importance and the quantization scale
The encoding order of the images corresponding to the
Encoding order determining means, and the order determined by the encoding order determining means.
Encode an image corresponding to a plurality of divided areas and code
Image data compression method is characterized in that and a coding means for generating data. 2. Encoding data for each image corresponding to the divided area, from the upper left of the image in raster scan order.
2. The apparatus according to claim 1, further comprising means for rearranging the divided areas so as to have the same data order as when the divided areas are encoded.
Image data compression method described in 1. 3. The divided area is composed of a number of blocks, which are consecutively arranged in raster scan order from the upper left of the image and serially numbered in raster scan order from the upper left of the image. 2. The image data compression system according to claim 1, further comprising means for adding the serial number of the area to be encoded as header information before the encoded data of the area. 4. The image data compression method according to claim 1, wherein the importance is determined by an information amount or a feature amount of an image in each of the divided areas. Wherein said coding order determining means, the quantization scale is when the more values predetermined threshold performs coding from a less the significance region, the quantization scale value less than the threshold value 2. The image data compression method according to claim 1, wherein an encoding order is determined such that encoding is performed from the region of higher importance at the time. 6. The image data compression method according to claim 5 , wherein said threshold value is determined by an information amount of one screen of an input image.