JPH0435268A - Picture code quantity prediction system and picture coding system - Google Patents

Abstract

PURPOSE:To attain high speed processing by using part of picture elements of an original picture so as to predict the output code quantity of the entire original picture when a picture data image-picked up electronically is compression-coded into a variable length code data. CONSTITUTION:An input picture data is stored in a frame memory 1 for each pattern and an output code quantity of the entire original picture is predicted by using part of picture elements of the original picture in the case of compression coding. That is, a thinning processing circuit 9 is provided to apply thinning processing to the picture elements of a stored object picture and after the entire picture element number is much reduced and the picture is coded and the obtained code quantity is measured by using a code quantity measurement device 10, an S/N calculator 11 obtains an output S/N to control a scaling factor controller 6, a scaling factor (n) giving processing code quantity and a prescribed S/N is selected to predict the output code quantity when the picture is coded with the specific scaling factor (n). Thus, the code quantity is predicted at fast speed.

Description

[Detailed Description of the Invention] [Field of Industrial Application] This invention predicts the amount of data after compression when compressing still image data using variable length encoding and transmitting or recording it. The present invention relates to an image code amount prediction method and an image encoding method that control the amount of code so that the value becomes a predetermined value.

[Conventional technology]

A method that combines orthogonal transformation and variable length encoding is considered to be effective as a highly efficient compression encoding technology for natural images (still images), and this type of method has been adopted as an international standard for color natural image encoding. (Journal of the Institute of Image Electronics Engineers;
Vo1218, No. 6, P398-P40
(see 7).

In this type of encoding method, by controlling certain parameters, it is possible to control the image quality of a decoded image and the amount of code after encoding. The relationship between the amount of code and image quality is that the larger the amount of code, that is, the lower the degree of compression, the smaller the deterioration of the image quality from the original image. The deterioration of image quality will be significant. However, this type of encoding method performs adaptive processing using local correlations in the image during encoding, making it possible to perform highly efficient compression according to the image quality of the target image. .

[Problem to be solved by the invention]

By the way, images with detailed patterns generally have a large amount of code;
A flat image with many solid areas has a small amount of code.

Therefore, even if encoding is performed using the same code amount control parameter, the obtained code amount will differ depending on the target image. Therefore, in this type of encoding method, it is not possible to predict in advance the amount of code generated by encoding, and it is not always possible to perform encoding so that a constant amount of code is obtained for an arbitrary target image. It's not easy. In particular, the applications of natural image coding methods to date have mainly been considered for image transmission such as still image videophones and facsimiles, and storage of image data in image data frames. was not necessarily required to do so.

However, when applying this type of encoding method to an electronic still camera, it is desirable to guarantee the number of recordable images when recording compression-encoded image data on a recording medium such as an IC card. requires a technique to compress and encode the target image with a fixed amount of code. For this reason, the original image is prescanned and compression encoded,
A method has been proposed in which the code amount after compression is measured and the code amount control parameter is changed using this value to approach an asymptotically determined code amount. However, with this method, the amount of processing increases for images with a very large number of pixels, making real-time processing difficult.

When compressing and recording multiple images in a memory with a fixed storage capacity, this invention predicts the amount of code after compression using specific control parameters so that the number of images that can be recorded is guaranteed. An object of the present invention is to provide an image code amount prediction method and an image encoding method that control the amount of code to a predetermined amount of code.

[Means to solve the problem]

This invention predicts the output code amount of the entire original image using a part of the pixels of the original image when compressing and encoding electronically captured image data into variable length code data, and also predicts the output code amount of the entire original image. Based on the results, the output code amount of the entire original image is controlled to a predetermined code amount.

[For production]

In this invention, the image data of the original image to be compressed and encoded into variable length code data is compressed into 1/2 and 1/4 in the vertical and horizontal directions.

1/8. . . . The number of pixels is changed by thinning processing to 1/64, etc., and the relationship between the output code amount and the number of pixels is determined manually using the output SN ratio as a parameter. Then, from the relationship between the number of input pixels and the output code amount determined in advance for multiple representative images, a representative image with the image quality closest to the original image is determined, and the number of input pixels of the determined representative image is The output code amount for the number of pixels of the original image is predicted from the relationship with the output code amount.

Then, a control parameter for obtaining the predicted output code amount is determined, and the original image is encoded using the determined control parameter, thereby compressing the original image to a predetermined code amount.

〔Example〕

FIG. 1 is a block diagram showing an embodiment of an image code amount prediction method and an image encoding method according to the present invention, and shows an example in which the present invention is applied to an electronic still camera.

In FIG. 1, input image data captured from an imaging unit (not shown) is stored in a frame memory 1 for each screen, and is cut out into one block of 8×8 pixel size.
CT (Discrete Co51ne Trans
A two-dimensional discrete cosine transform is performed using the former) 2. The DCT coefficient F ij obtained by DCT2 represents a component obtained by decomposing the input image data for one process into spatial frequencies, and is the coefficient F of the coefficient F ij. 0 represents a value (DC component) proportional to the average value of input image data, and variable 1
．． As j becomes larger, it represents a higher frequency component (AC component).

The DCT coefficients F ij obtained in this way are sent to the quantization circuit 3
Each coefficient is linearly quantized with a different quantization step width. A matrix representing each quantization step and amplifier width at this time is supplied from the quantization matrix circuit 4 as a quantization matrix. An example is shown in FIG. Each dark value of this quantization matrix is processed by a multiplier 5 with a coefficient of 2'' (n=o
t±1. ±2. ) is multiplied and the step width is changed to control the resulting code amount and code image quality. The width n of the coefficient 2'' is called a scaling factor and is controlled by a scaling factor control circuit 6.

Each coefficient of the DCT coefficients F and J quantized by the quantization circuit 3 is scanned from low-order coefficients to high-order coefficients by the zigzag scan circuit 7 and converted into one-dimensional data, and is converted into one-dimensional data. The long value and valid value (value other than "0") are converted into two-dimensional data. FIG. 3 shows a zigzag scan table. The output of the zigzag scan circuit 7 is a Huffman encoder 8
is Huffman encoded and stored on recording media such as IC cards (
(not shown) is stored as variable length encoded data.

The amount of code finally obtained as variable length encoded data is
As described above, it can be controlled by changing each threshold value of the quantization matrix. In this example,
Each threshold value of the quantization matrix is multiplied by a coefficient 2'' to change each darkness value to control the code amount and image quality after compression.

The relationship between the scaling factor n, the quantization step width, and the output code amount is as shown in the table of FIG. 4. That is, when the scaling factor n becomes larger, the quantization step width becomes larger and the amount of codes decreases, and conversely, when the scaling factor n becomes smaller, the quantization step width becomes smaller and the amount of codes increases. Although the above relationship generally holds true for any image, the quantitative relationship between the scaling factor n and the amount of code changes for each target image and is not uniquely determined. Therefore, the same scaling factor n
Even if the target image is encoded using , the amount of code obtained will differ depending on the target image.

Therefore, in this embodiment, a thinning processing circuit 9 is provided, and the pixels of the target image stored in the frame memory 1 are thinned out to make the total number of pixels very small, and then encoded.
The obtained code amount is measured by the code amount measuring device 1.0, and the output SN ratio is determined by the SN ratio calculator 11, and the scaling factor controller 6 is controlled from these values to obtain a predetermined code amount and a predetermined SN ratio. Select a scaling factor n that is the ratio. Then, from the result, the output code amount when the target image is encoded with a specific scaling factor n is predicted.

In other words, as shown in the fifth graph, the target image is 1/2, 1/4, 1/8, etc. vertically and horizontally, respectively.
The number of pixels is varied by thinning out to 1/64, and the relationship between the number of input pixels and the amount of output code is determined using the output SN ratio as a parameter. In addition, the original 8-bit straight line in the figure shows the output code amount when the thinned out original image is linearly quantized with 8 bits, and the ITE (TV Society) test chart "Skin Color Chart" is used as the original image. There is. FIG. 6 is a graph comparing with other images. In the figure, Pl is the aforementioned "skin color chart" and P2 is the ITE test chart "Switzerland Mountain Village". The reason why P2 has a larger code amount at the same SN ratio is that P2 is an image with more changes overall, and in such a case, the code amount increases.

As is clear from both graphs, even if the original image is thinned out to a very small number of pixels, the characteristics of the original image, such as whether it is an image with many changes or an image with many smooth parts, are preserved as is. I can see that it is being done.

Using the above results, for example, if the original image is 16x16
By determining the output code amount and output SN ratio of an image thinned out to pixels, it is possible to predict the output code amount of the original image at an arbitrary SN ratio, and to obtain a predetermined output code amount and predetermined SN ratio. It becomes possible to obtain a scaling factor n.

Therefore, for representative images, as shown in Figure 6,
A curve or table showing the relationship between the number of input pixels and the output code amount using the output SN ratio as a parameter is obtained in advance, and the target image is thinned out to 16 x 16 pixels and encoded.
By determining the output code amount and output SN ratio and plotting them on the graph of FIG. 6, a representative image having properties similar to the target image can be determined.

For example, if the representative image is P2, the output code amount and output SN ratio for a given number of input pixels of P2 can be found from the graph, so these values can be used as the output code amount and output SNR of the target image.
It can be considered as a predicted value of the N ratio.

Then, since the point on the graph of P2 closest to the predetermined output code amount of the target image is found, the output SN corresponding to this point
The ratio and scaling factor n are determined. By encoding the target image using the obtained scaling factor n, it becomes possible to compress the target image to a predetermined code amount.

Although the above explanation has been made using graphs, it goes without saying that the graphs may be stored as tables. Also, if the number of input pixels is SxS and the output code amount is Q, then Q=a(IogS)" (1-exp(-bSC))
However, a, b: constants determined by the properties of the image, and the following relationship holds: Cζ1.2. Since this equation expresses the curves of FIGS. 5 and 6, this equation may be used in place of the curve or table described above.

In the above-mentioned embodiments, DCT was used as the encoding method, but other encoding methods such as Hector quantization method, predictive encoding method such as DPCM, etc.
Alternatively, it can be applied to block coding methods, etc., and can be applied to all image coding methods that use variable length codes.

〔Effect of the invention〕

According to this invention, when predicting the code amount after compression of the original image, the original image is thinned out to, for example, 16x16 pixels and compressed and encoded, so that it is possible to speed up the code amount prediction. This is particularly effective when applied to target images with a very large number of pixels, such as printed images.

Furthermore, when this invention is applied to a conventional method in which pre-scanning is performed to obtain an asymptotically determined output code amount, high-speed processing is possible because the number of pixels processed by pre-scanning is extremely small, 16 x 16 pixels. Encoding processing can be performed multiple times at high speed, and real-time processing is also possible for high-resolution images.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an example of an image code amount prediction method and an image encoding method according to the present invention, FIG. 2 is a diagram showing an example of a quantization matrix, and FIG. 3 is a diagram showing the order of zigzag scan. , Figure 4 is a table showing the qualitative relationship between the scale factor, quantization step width, and code amount, and Figures 5 and 6 are graphs showing the relationship between the number of input pixels and the output code amount of a typical image. be. Patent applicant Rico Co., Ltd.

Claims (3)

[Claims] (1) When compressing and encoding electronically captured image data into variable-length code data, a part of the pixels of the original image is used to predict the output code amount of the entire original image. Image code amount prediction method. (2) When compressing and encoding electronically captured image data into variable-length code data, a portion of the pixels of the original image is used to control the output code amount of the entire original image to a predetermined code amount. An image encoding method characterized by: (3) The method of controlling the output code amount of the entire original image to a predetermined code amount using a part of the pixels of the original image is to thin out and encode the pixels of a plurality of representative images in advance, and then adjust the thinning rate. Find the relationship between the number of input pixels and the output code amount using the output SN ratio as a parameter by changing the number of pixels, find the output code amount and output SN ratio obtained by thinning out the target image, and calculate the output code amount and output SN ratio obtained by thinning out the target image. is compared with the relationship between the number of input pixels and the output code amount of the plurality of representative images obtained in advance, and the representative image closest to the image quality of the target image is selected, and the above image is selected from the selected representative images. 3. The image encoding method according to claim 2, wherein a quantization parameter that gives a predetermined code amount to the number of pixels of the target image is determined, and the target image is encoded using the determined quantization parameter.