JPH01241270A - Picture encoding system - Google Patents

Abstract

PURPOSE:To adjust a required encoding parameter automatically when a request value with respect to an encoding index is given by extracting a characteristic quantity of an input picture, classifying the picture depending on the result of extraction so as to decide the encoded parameter. CONSTITUTION:A picture is inputted from a scanner and an input picture 57 and an encoded index 58 are inputted to an encoded parameter automatic adjustment device 41 via a picture buffer. A characteristic quantity extraction section 53 extracts the characteristic quantity of the input picture 57, sends it to a picture classifying section 54 to decide a classified class of the input picture 57 depending on the characteristic quantity and sends the result to an encoded parameter decision section 55. The parameter decision section 55 references a characteristic curve storages section 56 by class and uses the obtained characteristic table by class to select an encoded index of the classified class of the input picture 57 versus encoded parameter characteristic curve. Moreover, the value 59 of the encoded parameter is decided by the request of the compressed ratio represented by the encoded index 58 from the characteristic curve, the result is fed to the encoder 42 to encode the input picture 57 and fed to the encoded buffer as an encoded data 60.

Description

Detailed Description of the Invention [Industrial Application Field] The present invention relates to an image encoding method for compressing image data in an image storage and transmission system, and particularly relates to an image encoding method for compressing image data in an image storage and transmission system. The present invention relates to an image encoding method that allows, when a required value is set, to easily determine encoding parameters for realizing the required value.

[Prior art]

Conventionally, in order to utilize image data with a large amount of data at low cost, various encoding methods have been proposed for suppressing image redundancy. For example, methods such as predictive coding, transform coding, vector coding, and block coding have been proposed as coding methods for grayscale images.

Furthermore, we focused on the edge areas where the brightness changes are large and the flat areas where the brightness changes are gradual, and based on variable size block coding, we weighted the edge areas and flat areas to preserve the spatial resolution and gradation resolution. There are also proposals for ways to change the . In other words, the image is divided into blocks of non-equal length according to the size of local brightness changes, and the number of quantization bits of the encoded data is changed depending on the block size, and the total code amount is quantized under the constraint that it is fixed. Bit allocation is performed to minimize distortion. In this case, the block division structure can be expressed as a Quad-Tree.

In these image encoding methods, when the values of parameters (hereinafter referred to as encoding parameters) input to control the compression ratio and image quality are given during encoding, the compression ratio and image quality are determined as a result.

On the other hand, when required values are given for compression ratios and image quality, it is necessary to adjust parameters through trial and error in order to achieve these values.

Regarding this type of encoding method, see 1, for example, “Comparative Study of Coding Methods for Still Images,” Hosaka et al., IEICE Technical Report (198
3 years), IE83-106. pp. 45-52''.

[Problem to be solved by the invention]

In the above-mentioned conventional technology, no consideration is given to automatic adjustment of encoding parameters when a desired compression ratio or image quality of a decoded image is set.

For this reason, it has been difficult to meet the request to maintain a constant compression ratio or image quality regardless of the nature of the input image.6 However, in actual image storage and transmission systems, such requests cannot be met. Occurs often. For example, in fields where image quality is important, such as medical imaging, constant image quality is required.
Furthermore, in systems where upper limits are set for transmission time and recording time, a compression ratio (code amount) is required to be optimized.

The purpose of the present invention is to improve such problems, and when a required value for the compression ratio or the image quality of a decoded image (hereinafter referred to as an encoding index) is given, the code required to achieve this value is determined. An object of the present invention is to provide an image encoding method that automatically adjusts encoding parameters.

[Means to solve the problem]

In order to achieve the above object, the image encoding method of the present invention is as follows:
In the image encoding method of the image storage and transmission system, when encoding an input image, there are encoding parameters input to control the compression ratio and image quality of the decoded image, and encoding indicators such as the image quality and compression ratio of the decoded image. means for storing characteristic curves for each class (class characteristic curve storage unit), means for extracting feature quantities used to classify the input image (feature quantity extraction unit), , means for classifying the input image (image classification unit), and means for determining the value of the encoding parameter that realizes the value of the encoding index required for the input image based on the characteristic curve for each class (image classification unit); When an image and encoding index are input, the feature amount extraction section extracts the feature amount of the input image, and the image classification section classifies the input image based on the extraction result. The encoding parameter determination section determines the requested encoding index for the input image using the relationship between the encoding parameter and the encoding index indicated by the class characteristic curve stored in the class characteristic curve storage section. The feature is that the value of the encoding parameter that realizes the value of is determined.

In addition, the above feature is characterized in that statistical quantities (average of standard deviations and standard deviation of standard deviations) calculated from local standard deviations of an image are used as the target of feature quantity extraction.

[Effect]

In the present invention, an index that can appropriately classify the image is selected as the feature amount of the image. Appropriate lath classification means that images that have a similar relationship between the encoding parameters and the encoding index (for example, compression ratio) when encoding is performed based on these are considered to be in the same class. It is a classification.

Furthermore, the class to which the input image belongs is determined according to the feature amount. Note that each class is associated with a typical characteristic curve representing the relationship between encoding parameters and encoding instructions for images in that class.

Furthermore, when the required value of the encoding index is input, the value of the encoding parameter is determined from this value and the characteristic curve.

The value of this encoding parameter is determined from the typical characteristic curve of the class, so it is not necessarily the optimal value, but by classifying with necessary and sufficient detail, an appropriate approximate value can be obtained. be able to.

〔Example〕

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is a flowchart showing an image encoding method in an embodiment of the present invention, and FIG. 2 shows the relationship between allowable distortion and compression ratio R in an embodiment of the present invention for images with different properties. FIG. 3 is an explanatory diagram of the classification obtained by calculating μ and σ for the image input to the image encoding mechanism in one embodiment of the present invention and plotting them two-dimensionally. Figure 5 is a block diagram of an image transmission system according to an embodiment of the present invention, Figure 5 is a functional diagram of an automatic encoding parameter adjustment device according to an embodiment of the present invention, and Figure 6 is a code diagram according to an embodiment of the present invention. FIG. 2 is an explanatory diagram of a parameter determination unit.

Note that in this example, the compression ratio is used as the encoding index,
The input image is encoded by the method described in Japanese Patent Application No. 61-306416.

First, the encoding method described in Japanese Patent Application No. 61-306416 will be described.

In this encoding method, as a first process, preprocessing is performed on the input image.

That is, the DC component is suppressed by dividing the input image into 16×16 size sub-images and subtracting the average luminance of the sub-images from each pixel.

Next, as a second process, block division and bilinear approximation are performed.

In other words, size the sub-image to 2 x 2 (4≧ to ≧1)
into blocks. This division is performed by recursively integrating four blocks (pixels at the first stage) into one block.

However, the blocks are integrated only when the distortion expressed by the following equation (1) is less than or equal to the allowable value.

(1) d=Σ(f (i, j) −f (it
j))2 Note that f(x+j) is a bilinear function that approximates the luminance change within the block, and is defined by four coefficients A-D.

Next, as the third process, quantization and Quad-Tree
Perform encoding.

That is, the coefficients A-D are quantized. The quantized coefficients are Q quad −T re representing the block partitioning structure.
e data (depth-firstW structure), and the quad-tree data and coefficient data become encoded data.

As mentioned above, in the encoding method described in Japanese Patent Application No. 61-306416, the encoding parameter is the allowable distortion, and decreasing the value of D provides a low compression ratio, and increasing the value of D provides a low compression ratio. High compression ratios can be achieved.

Next, the encoding parameter (allowable distortion D) used in this example
and the encoding S (compression ratio R), and the feature quantity (
The average μ of the local standard deviation of the image, and the standard deviation σf
).

When this allowable distortion and compression ratio R are determined for images having different properties, the relationship between them is shown, for example, as shown in FIG. 2. In addition, these images ~ [phase] are medical images,
These images have different characteristics (eight degree change), such as human images and landscape images.

In this case, the compression ratio R is expressed by the following equation (2).

(2) R=bit amount of original image/bit amount of encoded data This shows that the RD characteristics differ depending on the type of image. Qualitatively, it can be said that an image with many flat parts can be easily compressed, and the value of the compression ratio R increases for the same value of allowable distortion.

Furthermore, these R-D characteristic curves are divided into five classes (■
- m, m', (2).

In this embodiment, the average and standard deviation of local standard deviations of an image are used as image feature quantities that enable such lath division.

In this case, in a small block Bn of 4×4 pixel size,
Standard deviation σ. is calculated using equation (3).

However, fiJ is the brightness of pixel (i, j), and n is a subscript indicating a block.

Next, the local standard deviation σ obtained in this way. the average over the entire image, and the local standard deviation σ. The standard deviation of is calculated using the following equations (4) and (5).

(4)μ, = 1/N−deδ.

(5) σ, = I N · Σ σ. −μ.

Here, μ is the average local standard deviation of the image, and σ.

is the standard deviation of the local standard deviations of both images, and N is the total number of blocks.

Qualitatively, the average local aS deviation of an image (hereinafter referred to as μ) represents the average intensity of local brightness changes on the image, and the standard deviation of the local standard deviation of one image (hereinafter referred to as μ) represents the average intensity of local brightness changes on the image. σ
1) is the non-−
It can be seen as an indicator of the degree of In general, if μ is small, the image mainly has gradual changes and can be easily compressed.

Furthermore, if σ is small, the situation of local brightness changes will be uniform throughout the image, and when the allowable distortion value is changed in the above encoding method, block integration will rapidly progress at a certain point, and R - The slope of the D characteristic curve will change. This is illustrated by the class ■' curve in FIG.

Furthermore, for each image ■~[phase], μ, and σ.

When calculated and plotted in two dimensions, it can be divided as shown in Figure 3.

In this way, the R-D characteristic curve ■~[phase] shown in FIG.
By calculating μ and σ for each image and performing threshold processing, each image is divided into classes (classes I to III, III'
, IV).

Next, the configuration of the image transmission system of this embodiment will be described.

The image transmission system of this embodiment includes image buffers 40, 49, a coding parameter automatic adjustment device 4, as shown in FIG.
1, an encoding device 42, code buffers 43, 47, modems 44, 46, a communication path 45, and a decoding device 48.

This image buffer 40 stores images read by the scanner 50.

Further, the encoding parameter automatic adjustment device 41 receives the image data as input and determines the value of the encoding parameter (allowable distortion (hereinafter referred to as “total distortion”)).

Furthermore, the encoding device 42 encodes the image data according to the value of the encoding parameter D, and outputs it to the encoding buffer 43.

This encoded data is transmitted through Mochi 4441 communication channel 45 and modem 4.
6, and is stored in the code buffer 47 on the receiving side.

Further, the decoding device 48 decodes the encoded data to form image data and outputs it to the image buffer 47. This decoded image is finally output to the display 51.

Furthermore, the encoding parameter automatic adjustment device 41 of this embodiment
As shown in FIG. It includes an encoding parameter determination section 55 and a class-by-class characteristic curve storage section 56.

The class-specific characteristic curve storage section 56 stores a characteristic curve that provides a relationship between the encoding parameter D and the encoding index (compression ratio (hereinafter referred to as R)), and this characteristic curve is set for each image class.

Further, upon receiving the input image 57 from the scanner 50 via the image buffer 40, the feature amount extraction unit 53 classifies the input image 57 using equations (4) and (5). The feature quantity (μ1.σ,) used for is calculated.

Further, the image classification unit 54 performs the following on the input image 57:
When μ and σ1 are extracted by the feature extraction unit 53,
Based on this result, the class to which the input image 57 belongs is determined using the threshold shown by the broken line in FIG.

Furthermore, the encoding parameter determining unit 55 selects one R-
Select the D characteristic curve and input the input coding index (R)5
8 and its characteristic curve, an appropriate encoding parameter (D) 59 is determined and output.

The details of the encoding parameter determination unit 55 are shown in FIG. 6, and include a selector 64 and a class characteristic table 61. In addition, class characteristics table 61
For example, class ■～■.

It has an RD characteristic curve divided into III' and IV.

With this configuration, one of the RD characteristic curves for each class (for example, the RD characteristic curve of class ■) is selected based on the belonging class information 62 sent from the one-image classification unit 54.
is selected. Furthermore, the required value 63 of the compression ratio indicated by the encoding index (R) 58 is set to R of the R-D characteristic curve for each class.
An appropriate encoding parameter (D) 59 can be obtained by calculating the D-axis value as the axis value.

This encoding parameter (D) 59 is sent to the encoding device 42, and in the encoding device 42, encoding is performed on the input image 57 using the value of this encoding parameter (D) 59, and the encoded data is 60 is output.

Next, the procedure of the image encoding method in this embodiment will be described.

As shown in FIG. 1, when performing image encoding processing in the image transmission system of this embodiment, a class-specific R-D characteristic curve is set and stored in the class-specific characteristic curve storage section 56 in advance (101
).

Next, an image is input from the scanner 50, and the input image 57 and encoding index (R) 58 are passed through the image buffer 40 to the encoding parameter intransitive! ! When the input image 5 is input to the l device 41 (102), the feature extraction unit 53 extracts the input image 5.
7 feature quantities (μ, σg) are extracted and sent to the image classification unit 54 (103).

The image classification unit 54 uses this feature amount (μ, .

σ, ), the class to which the input image 57 belongs is determined (104) and sent to the encoding parameter determination unit 55.

The encoding parameter determination unit 55 uses the class characteristic table 61 obtained by referring to the class characteristic curve storage unit 56 to determine the R of the class to which the input image 57 belongs.
- Select a D characteristic curve (104). Furthermore, that R-
From the D characteristic curve, the value of the encoding parameter (D) is determined according to the compression ratio request value 63 indicated by the encoding finger m(R) 58 (105), and is sent to the encoding device 22.

In the encoding device 22, the encoding parameter (D) 59
The input image 57 is encoded by the method and sent to the encoding buffer 43 as encoded data 60 (107).

By applying this example to images other than the training images (■ to @l in Figure 2), the requested value and actual value of the compression ratio (compression ratio actually obtained as a result of encoding) can be calculated. By comparison, it was confirmed that even with the classification shown in Figure 3, the error between the required value and the actual value was within about ±25%.

In addition, in this embodiment, the compression ratio is used as the encoding index,
The method of Japanese Patent Application No. 61-306416 was used as the encoding method.1 For example, if image quality (SN ratio), etc. is used as the encoding index, and other encoding methods are used, the present invention also applies. Effects similar to those of the embodiment can be obtained.

〔Effect of the invention〕

According to the present invention, when a required value is given regarding the compression ratio, the image quality of a decoded image, etc. in image encoding, it is possible to automatically determine the value of the encoding parameter for realizing the required value.

Therefore, even in image storage and transmission systems that are required to maintain a constant compression ratio and image quality regardless of the nature of the input image,
Encoding processing can be performed automatically without human intervention.

[Brief explanation of the drawing]

FIG. 1 is a flowchart showing an image encoding method in an embodiment of the present invention, and FIG. 2 shows the relationship between allowable distortion and compression ratio R in an embodiment of the present invention for images with different properties. FIG. 3 is an explanatory diagram of μ for an image input to an image encoding mechanism in an embodiment of the present invention. and σ are calculated and plotted in two dimensions to explain the classification. FIG. 4 is a configuration diagram of an image transmission system according to an embodiment of the present invention. FIG. 5 is a diagram showing encoding parameters according to an embodiment of the present invention. FIG. 6, a functional configuration diagram of the automatic adjustment device, is an explanatory diagram of the encoding parameter determining section in an embodiment of the present invention. 40.49: Image buffer, 41: Encoding parameter automatic adjustment device, 42: Encoding device, 43° 47: Code buffer, 44, 46: Modem, 45: Communication path, 48: Decoding device, 50: Scanner, 51 ：Display, 53: Feature amount extraction section, 54: Image rasterization section, 55: Encoding parameter determination section. 56 Two-class characteristic curve storage unit, 57: Input image. 58: encoding index (compression ratio R), 59: encoding parameter (allowable distortion D), 60: encoded data, 61: class characteristic table, 62: belonging class information. 63: Required value of compression ratio, 64: Selector 2 μm: Average local standard deviation of image, σ1 Standard deviation of local standard deviation of bivalent image. Fig. 1 Fig. 2 Fig. 5 10 15 D Fig. 3 Fig. 5 10 15 20μσ ・〃■Me〃 [phase] Neko 22F) Mini 21 To Naki

Claims (1)

[Claims] 1. In an image encoding method for an image storage and transmission system,
When encoding an input image, characteristic curves showing the relationship between the encoding parameters input to control the compression ratio and image quality of the decoded image and encoding indicators such as the compression ratio and the image quality of the decoded image are stored for each class. means for extracting feature amounts for classifying the input image, means for classifying the input image based on the feature amount, and means for classifying the input image according to the characteristic curve for each class. The feature extraction means extracts the feature amount of the input image when the image and the coding index are input, and extracts the feature amount of the input image and extracts the feature amount from the input image. The dividing means classifies the input image based on the extraction result, and the encoding parameter determining means determines the encoding parameters and encoding index indicated by the class characteristic curve stored in the class characteristic curve storage means. Using the relationship with
An image encoding method characterized by determining a value of an encoding parameter that realizes a value of an encoding index required for the input image. 2. Claim 1, characterized in that the feature amount is a statistic calculated from the local standard deviation of the image.
Image encoding method described in section.