JP2557965B2 - Image coding method - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to encoding for storing and transmitting an image in which each pixel is represented by N color components, and in particular, N with correlation
The present invention relates to an image coding method capable of highly efficient coding even if each component is independently coded by converting an image made up of components into N unrelated components.

(Prior Art) Conventionally, as an encoding method of an image composed of highly correlated components, for example, as an encoding of a printing image, "compression of a printing image using adaptive vector quantization", Aizu, Takagi, 1986 Using the correlation between cyan (C), magenta (M), yellow (Y), and black (Bk), as in the Image Coding Symposium, 4x4 blocks are expanded in the CMYBk direction as well.
This is a vector quantization technology as a 6-dimensional block.

Here, with vector quantization, a plurality of pixels are grouped together,
It is implemented by constructing a vector and approximating possible patterns of the vector to a vector having a smaller number of patterns. In this vector quantization, the image can be approximated by the number of bits required to show the index indicating a smaller number of patterns, and the number of patterns can be reduced by using the feature that similar pixels are likely to occur in the vicinity in images. I am letting you. In this conventional technique, a vector is set across CMYBk to utilize not only one color component but also correlation between color components. Further, since the print image is required to have high quality, the approximate pattern cannot be made too small. When there are many approximate patterns, a huge amount of calculation is required to select a pattern for approximation.

Also, "Study on compression coding of printing plate making data"
Nakajima, Yasuiin, et al., 1987 Image Coding Symposium,
As described above, there is a technique in which, after converting CMY to YIQ in CMYBk, each component is independently used by a conventional encoding method such as GBTC, PCS, DCT-VQ.

However, the former method described above has a problem that it takes a very long time to create a codebook due to a technique called vector quantization. This problem can be solved by reducing the amount of codebook depending on the required image quality, but it is not suitable for an image that requires particularly high quality such as a printed image.

Also, in the latter method, since the correlation with the Bk component remains, there was a problem in efficiency in independently encoding the YIQ component and Bk. Furthermore, regarding the YIQ component, if the three components are uncorrelated with each other, the YIQ axis and the YIQ axis that are uncorrelated should coincide with each other, and there should be no gap between both axes, as shown in Fig. 4. In fact, depending on the image, there is a large gap between both axes. That is, the correlation between the three components remains, and there is a problem in encoding each component.

(Object of the Invention) The present invention has been made to solve the above-mentioned problems. N components, which have a correlation between components and have a problem in efficiency in independent encoding, are generated for each image. In addition, it is an object of the present invention to be able to efficiently code an image signal by independently coding all N components by converting them into uncorrelated components.

(Structure of the Invention) (Differences Between Features of the Invention and Prior Art) In order to achieve the above object, the present invention obtains a covariance matrix between components of each image and projects each component in the direction of its eigenvector. The most important feature is that it is converted into a non-correlated component and used as additional information for decoding the conversion matrix.

According to the composite method of the present invention, it is possible to code each component completely independently by converting the N components of the image into uncorrelated components, as compared with the conventional technique.
Since this operation is performed for each image, it differs in that the most effective processing can be performed for each image.

(Example) Hereinafter, one example of the present invention will be specifically described with reference to the drawings.

In all the drawings for explaining the embodiments, those having the same functions are designated by the same reference numerals, and the repeated description thereof will be omitted.

FIG. 1 shows cyan (C), magenta (M), and
The block diagram which shows the system configuration of the encoding system when applied to the image for printing which consists of yellow (Y) and black (Bk) is shown.

In FIG. 1, 1 is a color scanner, 2 is a covariance sequence calculator between CMYBk, and 3 is an eigenvector calculator of the covariance sequence, which outputs a conversion matrix composed of eigenvectors. 4 is a conversion unit that makes the CMYBk components uncorrelated with each other,
5 is an inverse conversion matrix calculation unit for the conversion matrix, and 6
Is a DCT conversion unit, 7 is a quantization unit, and 8 is a recording processing unit.

In the image coding method of the present embodiment, the CMYBk component of the image signal input from the color scanner 1 is the covariance matrix calculation unit 2
Sent to Here, first, the average value of each component is obtained, and the difference between the average value and the pixel value is integrated among C, M, Y, and Bk, and divided by the total number of pixels to obtain a 4 × 4 covariance. The matrix is required. The eigenvectors of the covariance matrix are calculated by the eigenvector calculator 3 by diagonalizing the covariance matrix.

The CMYBk signals from the color scanner 1 are matrix-calculated by a conversion matrix composed of the eigenvectors obtained by the eigenvector calculation unit 3, so that components that are uncorrelated with each other are obtained.
It is converted into K1, K2, K3, K4 by the decorrelation conversion unit 4. The four images thus obtained are independently DCT-transformed by the DCT transform unit 6, and the transform coefficients are quantized by the quantizer of the quantizer 7. On the other hand, the inverse matrix of the transformation matrix obtained by the eigenvector calculating unit 3 is calculated by the inverse matrix calculating unit 5, and the quantized transform coefficient obtained by the quantizing unit 7 is added as additional information at the time of decoding. At the same time, it is passed to the recording processing unit 8, where it is encoded and recorded.

FIG. 2 is a block diagram showing a system configuration of another embodiment in which the present invention is applied to encoding a print image.

In FIG. 2, 9 is an eigenvalue / eigenvector calculation unit,
Reference numeral 10 is a normalization / sampling control unit, 11 is a normalization unit, and 12 is a sub-sampling unit.

In this embodiment, a normalization unit 11 and a sub-sampling unit 12 are added to the embodiment shown in FIG. 1, and the four components that are uncorrelated by the decorrelation conversion unit 4 are normalized and sub-sampled. Normalize by comparing the eigenvalue input to the control unit 10 and the eigenvalue obtained by the eigenvector and the eigenvector calculation unit 9 at the same time with a predetermined value.
Alternatively, it is determined whether to subsample.

Here, the normalization is to make the data fall within a predetermined range for convenience of operation.
The component for which normalization is determined is input to the normalization unit 11 and normalized by the square root of the eigenvalue. Further, the component for which sub-sampling has been determined is input to the sub-sampling unit 12, is subjected to smoothing filtering to suppress aliasing noise after sub-sampling, and then, in the horizontal direction.
Subsampled to 1/2. In the case of FIG. 2, K1 is determined to be normalized, and K2, K3, and K4 are determined to be subsampling.

Further, the normalized or sub-sampled component is transformed by the DCT transform unit 6, and the transform coefficient is transformed by the quantizer unit 7.
Is quantized by. The quantized transform coefficient is stored in the recording processing unit 8 together with the inverse matrix output from the inverse matrix calculating unit 5 and the normalization factor additional information output from the normalizing unit 11.
Then, it is encoded and recorded.

(Effects of the Invention) As described above, according to the present invention, N color components are converted into mutually uncorrelated components, so that each component can be encoded independently. Further, since this conversion is performed for each image, it is an optimum process for each image and is suitable for encoding a printing image in which high quality is important.

As a result of performing a simulation based on the embodiment shown in FIG. 2 of the present invention, as shown in FIG. 3, the S / N was improved by about 1 dB as compared with the method by CMY-YIQ conversion.

In the embodiment described above, the encoding method of the printing image composed of the four color components of CMYBk has been described.
In the case where the image is composed of two color components, the image is composed of three color components such as RGB, and the color saturation is reduced when C, M, Y, and Bk are mixed, a special color is newly added to make it five or more colors. It is also applicable to.

In addition to the DCT transform method,
It is also possible to use truncation coding (BTC), vector quantization method (VQ), or Karhunen-Loeve (KL) transformation.

[Brief description of drawings]

1 and 2 are block diagrams showing respective system configurations when the present invention is applied to a printing image encoding system,
FIG. 3 is a diagram showing the result of the simulation based on the embodiment shown in FIG. 2, and FIG. 4 is a YIQ space in the prior art, in which the axis having a non-correlated distribution and the Y, I, Q axes are formed. It is a figure which shows a corner. 1 ... Color scanner, 2 ... Covariance matrix calculation unit, 3 ...
… Eigenvector calculation unit, 4 …… Decorrelation conversion unit, 5 ……
Inverse matrix calculation unit, 6 ... DCT conversion unit, 7 ... Quantization unit, 8
...... Recording processing unit, 9 ... Eigenvalue, eigenvector calculation unit,
10 …… Normalization / subsampling control unit, 11 …… Normalization unit, 12 …… Subsampling unit.

Claims (1)

(57) [Claims] 1. When encoding an image in which each image is represented by N color components, a means for obtaining a covariance matrix between the respective components and an eigenvector of the covariance matrix in the N-dimensional space By using the direction as a new coordinate axis and re-expressing each component, a means for obtaining N components that are uncorrelated with each other, and a matrix for coordinate conversion in an N-dimensional space as additional information,
An image coding system characterized by comprising means for independently coding the obtained N uncorrelated N components by a predetermined coding method.