JPS6386963A - Image information encoding system - Google Patents

Abstract

PURPOSE:To reproduce an image which is scarcely deteriorated, at the time of decoding, by taking notice of a feature in a block, executing category classification, and performing quantization. CONSTITUTION:Data which has segmented picture element data of one picture element and 8 bits (256 gradations) to blocks of 4 picture elements X 4 picture elements is brought to Hadamard conversion by a Hadamard converting part 101. A coefficient Y11 is quantized to data of 8 bits by a scalar quantizer 104, and it becomes a lightness average in the block. Y11-Y44 are classified into several kinds of categories corresponding to a feature of original image data in accordance with whether values of thier coefficients are high or low, by a segmentation part 102. Subsequently, a quantizer corresponding to the discriminated category is selected, and a result of quantization is outputted as a pattern code of 9 bits. In this way, by setting the coefficient utilized for rounding processing of scalar quantization, in accordance with the category, the scalar quantization suitable for the category is executed, and a good image can be reproduced.

Description

DETAILED DESCRIPTION OF THE INVENTION [Technical Field] The present invention relates to an image information encoding method suitable for encoding halftone image information.

[Prior art]

Conventionally, when transmitting and storing images, efficiency has been considered and redundancy has been suppressed through encoding (generally called compression).

In recent years, with the development of digital image processing technology and device technology, the images to be transmitted and stored have shifted from binary to multivalued images, and the resolution has also increased. As a result, the amount of data has become enormous, and highly efficient encoding technology has become necessary.

But what about the existing encoding technologies? Many of them are aimed at value images, and representative ones are MH, MR, MMR, etc., which are defined for facsimile, and are inherently not suitable for encoding multi-value images.

In addition, several encoding techniques for multivalued images have been reported, including block encoding, predictive encoding, orthogonal transform encoding, and the like. Most of these methods are intended for television images, and are not suitable for encoding general document images or halftone dot images.

Especially in orthogonal transform coding, the small q of the conventional sequence
It takes advantage of the fact that electricity is concentrated in a certain place. This method is effective for continuous tone images such as photographs, but when applied to images with unique edge structures such as document images or halftone dot images, significant deterioration occurs.

〔the purpose〕

It is an object of the present invention to eliminate the drawbacks of the above-mentioned conventional examples and to efficiently encode multivalued data of document images, halftone dot images, and photographic images without significant deterioration.

〔Example〕

The encoding method according to the present invention cuts out multilevel image information to be encoded into mXn blocks, and divides the blocks into 101
This is then subjected to orthogonal transformation (Hadamard transformation), features within the block are extracted, and the blocks are classified into several categories. After that, optimal quantization is performed for each category.

FIG. 1 shows an example of the configuration of an encoding circuit to which the present invention is applied.

101 is a Hadamard transform unit, 102 is a segmentation unit that performs category classification, and 103 is a selector that selects a quantizer corresponding to the category based on the output from the segmentation unit 102. 104 is a scalar quantizer for the coefficient Y11 representing the brightness average value, and 105,1
06 and 107 are scalar quantizers that perform scalar quantization suitable for each category, and 108, 109, and 110 are vector quantizers that are suitable for each category. Note that the number of scalar quantizers and vector quantizers is increased or decreased depending on the number of categories. 111 is a selector that selects one according to the block to be processed based on the output from the segmentation unit 102 from among the outputs of the vector quantizer of each category, and has the same function as the selector 103 in the @ stage. It is something to do.

The operation will be explained here. One pixel is 8 bits (2
When data obtained by cutting out pixel data (56 gradations) into blocks of 4 pixels x 4 pixels is subjected to Hadamard transform by the 7 Damart transform unit 101, the coefficient Y 11 takes a value of 0 to 1020, that is, 1 Obit, and Y12 to Y44 each take a value of -510 to +510, that is, 10 bits. Therefore, the coefficient Y11 is quantized into 8-bit data by the scalar quantizer 104, and this is taken as the intra-block brightness average.

Y11 to Y44 are determined by the segmentation unit 102,
Depending on the magnitude of the values of these coefficients, the original image data is classified into several categories according to its characteristics (16 in the tree example).
selectors 103 and 111 each select a quantizer according to the category determined by the segmentation unit 102, and output the quantization result as a 9-bit (7) code (pattern number code).

Below, Hadamard transform, segmentation, scalar quantization, and vector quantization will be explained individually.

FIG. 2 shows a 4×4 Hadamard transform in the Hadamard transform unit 101, or before the transform, Y is the signal after the transform. Note that this conversion is expressed as () (16: 16th order Hadamard conversion!D).

Here = [X11, X12・X13, X14, X21
,X22・---,X43・X44] ”= [Yl
l-Y12・Y13. Y14・Y21, Y22・-
=, Y43- Y44], then it becomes.

rfI, Figure 3 shows an example of the Hadamard transform, and when there is a vertical edge in the block as shown in (1) in the figure, a large value appears in the coefficient Y12. The sign is the slope of the brightness of
plus).

However, since the coefficient Y11 represents the brightness average, it will not be particularly described here.

Similarly, in (2), when there is a horizontal edge in the block, a large value appears in the coefficient Y21. Others (3) to (16) show patterns of vertical, horizontal lines, joint shapes, diagonal edges, and diagonal lines as shown in the figure.

In this way, the image pattern and brightness of the block to be encoded can be known from the Hadamard transform result.

FIG. 4 shows the meaning of parameters when performing segmentation for each block in the segmentation unit 102, and each parameter is obtained from the Hadamard transform result Y using the following equation.

VEE= IY12++ 1Y131VLE=
l Y14 1 + I Y241HEE=
I Y21 l'+ l Y31 1HLE
= l y411 + l Y4210TH= 1
Y221+IY331+1Y441EF =lVE
E-HEEI LF =lVLE-HLEI The parameters are each determined by paying attention to the characteristics of the pattern shown in FIG. VEE is the vertical edge component parameter, VLE is the vertical line component parameter, HEE is the horizontal edge component parameter, HLE is the horizontal line component parameter, OTH
is a component parameter for other diagonal edges, etc., and EF is a parameter representing the strength of an edge, and when this is large, it is understood that the edge is strong. Similarly, LF is a parameter indicating the strength of a line segment.

FIG. 5 shows how the segmentation unit 102 classifies each block into 16 categories using the five parameters described above, and each branch and each category has the following meaning. This is called segmentation.

20: Diagonal edge, line 21: Diagonal edge, line + complex pattern 22: Vertical edge 10 complex pattern 23: Horizontal edge 10 complex pattern 24: Vertical line + complex pattern 25: Horizontal line medium complex pattern 30: Monotonous flat part 31: Flat edge 32: Vertical edge with small gradation difference 33: Horizontal edge with small gradation difference 40: Vertical edge with large gradation difference 41: Horizontal edge with large gradation difference 50: Vertical edge pattern 51 with large gradation difference : Horizontal edge pattern with large gradation difference 52 : Vertical line pattern with large gradation difference 53 : Horizontal line pattern with large gradation difference Also, each branch Ij) to 1'GI in FIG. 5 is shown below. It has various meanings.

Branch Φ> Separation of strong diagonal line (large OTH) <branch ■> Branch that separates edge pattern (large VEE, HEE) and linear pattern (large VLE, HL) +5・> Strong horizontal edge pattern (EF 4) Branching to extract vertical edges (large VEE) and horizontal edges (large HEE) Branching to extract complex patterns including diagonal edges (large OTH)
ζ・〉 Branch/foot that separates a weak vertical and horizontal pattern (EF is slightly large) and a flat part〉 Vertical edge (VEE is large)
Separation of horizontal edge (large BEE) and low branching
Branch monk that extracts those with large OTH from complex patterns) Branch 4 that extracts those with large HLE from complex patterns Branch 0 that extracts those with large VLE from complex patterns from inside) (E
Branch 4j that separates E and VER > Branch 6 that extracts a strong vertical and horizontal line pattern (large LF) > Vertical line pattern (
A branch that separates a horizontal line pattern (with large Vl and E) and a horizontal line pattern (with a large ILE) Branch that extracts a pattern that includes an edge component (with a slightly large EF) from the line pattern separated by branch 2.]5 〉 Separation of vertical edge system (large VEE) and horizontal edge system (large HEE) - 9〉 Figure 6 shows the separation of complex patterns including diagonal edges (large OTH) and flat parts. This is an explanation. In the diagram,
One pattern of category 40 is explained as an example. first,
+/- of each coefficient except Y11 for the pattern classified into categories by the segmentation unit 102
The sign and its absolute value are each separated as a “phase component”.
, called the "oscillation@component", and each coefficient of the amplitude component is weighted, and nonlinear scalar quantization is performed by scalar quantizers 105 to 107.The coefficients in the shaded part in the figure are There is a predetermined ignore for category 40, and the numbers are the result of rounding by scalar quantization.

FIG. 7 shows the results of nonlinear scalar quantization for all 16 categories, and the coefficients indicated by dots in the figure are coefficients to be ignored depending on the categories mentioned above. Also, the number following the category number is the bitl assigned to each coefficient.
It represents.

That is, when reproducing an image from encoded data, important coefficients are extracted for each category in order to reproduce an image close to the original image, and scalar quantization is performed based on the coefficients. Therefore, by setting the coefficients used for the rounding process of scalar quantization according to the category, scalar quantization suitable for the category can be performed, and a good image can be reproduced.

FIG. 8 is a hardware block diagram for realizing the Hadamard transform in the Hadamard transformer 101, in which all coefficients are computed in parallel to increase speed.

A data input line 801 receives 16 pieces of 8-bit data for each pixel. Reference numeral 802 is a block buffer for this, which is composed of 4×4=16 latches. 8
03 is an address generator that specifies pixels in the X block buffer 802 and the H element generator 804.
In other words, the intersecting elements are specified in the calculation of the Adabul transformation Y=genus W 1sX (4X4 blockization) shown above.

804 is written in ■- or generates,
It is composed of ROM. In addition, the outputs are h, h
, h ・Fist...h and 16 pieces L9.21
3i 18i parallel, 4 bits output from the address cinerator are input to the ROM address, and the statement = 1.・Conflict...In the 0 diagram that will be specified up to 16, i, j = L・number ・-life 4゜k, L; L
=1. - φ...16.

805 is an arithmetic unit for the element Y11 of the Y matrix, and as can be seen from the calculation formula shown above, it is composed of an adder. 806 and 807 are arithmetic units for coefficients other than Y11, and as shown in the calculation formula, they are composed of adders and subtracters. Reference numerals 808, 809 and 810 are percentage dividers, which round down the lower 2 bits. 811,8
12,813 is the result of these calculations, and Yll is 1o
As described above, 16 coefficients are operated in parallel in the operation of the 0 circuit which outputs a positive number of bits, a flag bit indicating a sign of +/- for other bits, and a 9-bit complement. The calculation details are as shown in the formula.

That is, data is read out from the block buffer 802 for one pixel i, and addition or subtraction is performed according to the +/- sign output from the H1ll element generator 804.

FIG. 9 is a hardware block diagram of the segmentation unit 102 that performs segmentation. 901.
902, 903, 904, and 905 are parameter calculation units for segmentation, which include a circuit (inverter and adder) that converts the complement into an absolute value, and an adder that adds two or three absolute values. configured. 9
Similarly, 06°907 is a computing unit for each parameter for performing segmentation, and is composed of a subtracter.

908.909,910,911,912゜913 is
In order to determine the category, the comparison shown in the figure is performed, and it consists of a comparator. When this output is input to the segmentation lookup table ROM 91B, first, HEE is determined by the lower 4 bits of the output of the ROM.

The 0 determination unit 917, which outputs the result of which of VEE, HLE, and VLE has the largest value, compares the parameter it specifies with OTH according to this output, and the result is used again. Input to segment cheese lookup table ROM918 h6.9
14°915.916 also performs the comparison as shown in the figure, and is composed of a comparator and a selector.

The above results are input to the segmentation lookup table ROM 918, segmentation is performed, and the resulting categories are stored in the ROM 9 as a 4-bit code.
18 outputs are output from the upper 4 bits. Note that the determination method is as shown in Table 1.

Incidentally, since the scalar quantization and the hectoral quantization shown in this embodiment are not particularly limited, a detailed explanation will not be provided.

In this example, the Hadamard transform was used for ease of calculation and /\- code conversion, but a similar orthogonal transform (
The same idea may be applied using discrete CoS transforms, slant transforms, etc.).

In addition, in this embodiment, segmentation includes VEE and VEE.
We used five parameters: LE, HEE, HLE, and OTH, but basically any method that performs categorization focusing on vertical, horizontal edges, vertical, horizontal lines, and other diagonal edges can be used instead. For example, UEE=l
Y121.

VLE=lY141, HEE=IY211, HLE
=I Y41 +, 0TH=I Y221, etc.

In addition, although scalar quantization and vector quantization are used for quantization in this embodiment, the quantization is not limited to this, and scalar quantization or vector quantization alone may also be used.
A method is used to separate and quantize the "amplitude components", but
The size of the matrix serving as the unit of encoding is not limited to 4×4, but can be changed depending on the content, resolution, and circuit elements of the image to be encoded.

[Effect] 1Jl-stop tapping 1. By focusing on e-groups within a block, classifying them into categories, and applying quantization, it is possible to perform encoding appropriate for the image to be encoded, and it is possible to reproduce images with less deterioration during decoding. . It is also possible to determine the area of an image by looking at the classified categories and the categories of adjacent blocks.

In addition, the shape of the edge mainly depends on the sign, and its gradation level depends on the absolute value, so when quantizing, by separating the sign (phase component) and absolute value (amplitude component), the block It is possible to prevent significant deterioration of the shape of the inner edge.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing an example of the configuration of an encoding circuit to which the present invention is applied, Fig. 2 is an explanatory diagram of Hadamard transform, Fig. 3 is a diagram showing an example of Hadamard transform, and Fig. 4 is an explanatory diagram of segmentation. , Figure 5 is a diagram showing a specific example of segmentation, Figure 6 is a diagram showing an example of quantization, Figure 7 is a diagram showing scalar quantization results, and Figure 8 is hardware for realizing Hadamard transform. Block diagram, Figure 9 is a hardware block diagram for realizing segmentation.
In the figure, 101 is a Hadamard transformer, 102 is a segmentation unit, 103 is a selector, and 105 to 107
is a scalar quantizer.

Claims (1)

[Claims] In an image information encoding method that cuts out image information into blocks of m×n size and encodes each block, the block is regarded as an m×n matrix, and means for performing orthogonal transformation, and coefficients of the transformed matrix are provided. It is equipped with means for calculating several kinds of parameters predetermined from the above, and means for determining the magnitude relationship of the parameters, and for classifying each block of input image information into several predetermined categories according to the magnitude relationship of the parameters. , an image information encoding method characterized by performing quantization for each category.