JPH03268679A - Picture interpolation coding device - Google Patents

Abstract

PURPOSE:To suppress production of a high frequency component in the result of orthogonal conversion and to prevent the deterioration in a picture compression efficiency by analyzing a density level of a picture in an area and interpolating a null part at the outside of a border through the use of the result. CONSTITUTION:A block decomposing section 19 of a picture interpolation section 18 decomposes a block of a picture element string in lateral and horizontal directions and sends the result to an analysis section 20 together with a data representing the border of the area of the picture element string. The analysis section 20 solves an orthogonal conversion equation analytically so that the high frequency component of the orthogonal conversion value is zero from a value of the picture element in the area in the picture element string, determines definitely the interpolation value to interpolate the picture, a block forming section 21 reconstitutes the block from the picture element string in the lateral or longitudinal direction in which the interpolation processing is finished. Through the constitution above, when an area not reaching a block is coded, the production of a high frequency component based on the orthogonal conversion independently of the substantial property of a picture is suppressed and efficient picture coding is implemented.

Description

[Detailed Description of the Invention] [Summary] Provided is an image complementary encoding device for compressing and encoding image data, which suppresses high frequency components in orthogonal transformation results to O and prevents a decrease in compression efficiency during encoding. For the purpose of this, the image encoding unit divides the image into blocks, which are partial images of a predetermined size, and encodes them using orthogonal transformation.The image encoding unit decomposes the blocks including the boundaries of the image area into pixel sequences. In addition, there is a block decomposition unit that generates data indicating the boundary of the region in the pixel row, and a block decomposition unit that generates data indicating the boundary of the region in the pixel row, and analytically calculates complementary values for pixels outside the region so that high frequency components become 0 from the values of pixels in the region in the pixel row. A block that does not include a boundary portion of an image area, and includes an image complementation unit that includes an analysis unit that determines a block, and a block formation unit that reconstructs a block from horizontal or vertical pixel rows obtained by performing the complementation processing. and the interpolated blocks from the image interpolation unit.

[Industrial application field]

The present invention relates to a device for compressing and encoding image data, and in particular, when encoding a partial image by orthogonal transform encoding, it is possible to efficiently perform interpolation of a portion that is less than a partial image by analytically performing interpolation of a portion that is less than a partial image. The present invention relates to an image complementary encoding device that can perform image encoding.

In recent years, it has become possible for computers to handle large amounts of data, and techniques for processing large amounts of data such as image data using computers have been developed. At the same time, the capacity of storage devices required to store data is also increasing.

However, the increase in the amount of data to be handled far exceeds the increase in the capacity of storage devices. Furthermore, the influence of the increase in the amount of data is even more significant not only in terms of data accumulation, but also in data transmission (communication).

In order to deal with such an increase in the amount of data, a technology that can compress data efficiently and with high quality is required. Many techniques have been proposed for data compression for image data, such as predictive coding, vector quantization, and transform coding, but among these, transform coding has the lowest compression rate and image quality. considered effective.

In the transform encoding method, the image is divided into rectangular partial images, orthogonal transformation is performed on each partial image to concentrate energy, and then the transformed data is quantized to compress the amount of data. ing.

In such a transform encoding method, when encoding is performed in units of partial images, it is desired to prevent compression efficiency from deteriorating based on the existence of areas that are smaller than the partial image.

(Prior Art) In general, the density levels of pixels constituting an image are not uniform.However, when an image is observed locally, the density levels between adjacent pixels change clearly, and the density There is a strong correlation between the levels. When encoding an image, pay attention to the fact that it changes smoothly and use this feature to perform encoding. As an example of encoding, an encoding method using discrete cosine transform (DCT) will be described.

FIG. 7 explains the block division of an image, and the image is a partial image (
blocks). This block is the minimum unit of processing when performing compression encoding. The reason why the image is divided into such small regions is that, as described above, when looking at each small region, there is a strong correlation between the density levels of the image.

Next, each block is subjected to two-dimensional discrete cosine transformation. By discrete cosine transformation, each block is converted from density value distribution data to frequency component intensity distribution data as a result of the transformation. A strong correlation between the density levels of pixels means that in the conversion result, the intensity of low frequency components is large and the intensity of high frequency components is small. Therefore, in each block, energy is concentrated in low frequency components by discrete cosine transformation.

FIG. 8 explains the concentration of energy due to the discrete cosine transform, and shows how energy is concentrated in low frequency components due to the discrete cosine transform being performed on a certain block.

In the figure, the energy is concentrated in the dark colored parts and shows a large value, and as the color becomes lighter, the energy becomes lower. Since the conversion value is approximately 0 in the white part, the amount of data can be compressed by removing this part.

As for the size when dividing an image into blocks.

A square block having a side size of 2 pixels, such as 8 (pixels) or 16 (pixels), is used.

This is because a high-speed discrete cosine transformation algorithm exists for a rectangular area with sides of 2X pixels, and it is easy to implement in hardware.

However, when dividing a real image into blocks,
There are cases where the size of the image is not as thick as an integral multiple of the block, resulting in a block that includes the edge of the image.

FIG. 9 shows a block including the edge of the image, where A indicates the edge of the image, and the shaded area is a block including the edge of the image.

In a block including such an edge of an image, there is a difference in density level between the image and the blank area other than the image, so a high frequency component is generated in the discrete cosine transform result, reducing the effectiveness of data compression.

Furthermore, there is a method in which an image is divided into regions in which parts of the image have similar properties into one region, and then each region is divided into blocks and encoded.

FIGS. 10(a) and 10(b) explain the area division of an image, and by dividing the original image shown in (a) into areas, the areas shown in (b) 1 to 5 are obtained. Moon, sky.

It has been shown that five regions, each with similar properties, can be obtained, such as mountains, tree leaves, and tree trunks.

It is known that in encoding using orthogonal transformation, the stronger the correlation between pixels, the higher the efficiency and the better the quality. Therefore, by dividing an image into regions with similar properties and compressing each region, more efficient and high-quality encoding can be performed.

However, in the case of encoding using such image region division, a large number of blocks including region boundaries are generated.

FIGS. 11(a) and 11(b) show the block division of the area, and each area 1 shown in FIG. 10(b)
An example of dividing ゜4 into rectangular blocks,
As shown in the shaded area in both figures,
It is shown that blocks containing region boundaries occur.

Conventionally, when encoding a block that includes the edge of an image, that is, the boundary of a region, blank areas outside the region are
Encoding was performed by filling in an appropriate constant value, such as a value representing white or black, or a value at a boundary.

[Problem to be solved by the invention]

In an image encoding method, the image is divided into blocks and the image is encoded using orthogonal transformation using each block as a unit.
The edge of the image, that is, the boundary of the region segmentation result, exists within the block. In such a block, values in blank areas outside the area are interpolated, but a density difference occurs at the boundary of the area, and a strong high frequency component is generated when orthogonal transformation is performed.

Generally, images do not contain high-frequency components, so the high-frequency components generated in this way not only distort the properties of the image, although they are unrelated to the properties of the image itself, but also cause high-frequency components to be removed during compression encoding. Since it is necessary to encode images up to the maximum length, there is a problem that the efficiency of image compression decreases.

The present invention aims to solve the problems of the prior art, by analyzing the density level of an image within an image or within a region, and using the results to fill in blank areas outside the boundary. It is an object of the present invention to provide an image complementary encoding device that suppresses high frequency components in orthogonal transformation results to O and prevents a decrease in compression efficiency in encoding.

[Means to solve the problem]

As shown in FIG. 1, the image complementary encoding device of the present invention is an image coder that divides an image into blocks, which are partial images of a predetermined size, and encodes them using orthogonal transformation. The conversion unit 11 includes an image complementation unit 18 consisting of a block decomposition unit 19, an analysis unit 20, and a block formation unit 21, and includes a block that does not include the boundary of the image area and a block that is complemented from the image complementation unit 18. Encoding is performed using blocks.

Here, the block decomposition unit 19 decomposes a block including a boundary portion of an image area into pixel columns, and generates data indicating the boundary of the area in this pixel column,
The analysis unit 20 analytically determines complementary values for pixels outside the area so that high frequency components become 0 from the values of pixels within the area in this pixel column. Block forming part 2
1 reconstructs a block from the horizontal or vertical pixel rows obtained by performing the interpolation process in this way.

[Effect]

In the present invention, an image is decomposed into blocks, a block consisting of 2" x 2 pixels that requires interpolation is decomposed into a pixel string of 2x pixels, which is the unit of interpolation, and interpolation is performed based on information on the boundary of the area. The required range is determined, a complementary value is analytically determined from the density difference within the region, and a block is reconstructed from the complemented pixel rows.

FIGS. 2(a) and 2(b) explain two-dimensional orthogonal transformation. First, as shown in (a), -dimensional orthogonal transformation is performed in the horizontal direction, and then (b) A one-dimensional orthogonal transformation is performed in the vertical direction as shown in . Note that this order may be reversed.

Since the image is complemented in one dimension accordingly, the block decomposition section first decomposes it into pixel columns in the direction in which orthogonal transformation is to be performed.

For example, if a block consisting of 8 x 8 pixels is
, (b>, when orthogonal transformation is performed in the horizontal and vertical order, it is decomposed into eight 1×8 pixel columns in the horizontal direction.

FIG. 3 shows a 1×8 horizontal pixel column.

The horizontal one-dimensional pixel row obtained through ζ is sent to the analysis section along with information indicating which part of the pixel row is within the region and which part is the region to be complemented.

The analysis section determines the complementary value as shown below.

One-dimensional orthogonal transformation can be expressed by a first-order polynomial as shown in the following equation (1).

Is =aN+X+ +aszxz +aN:+
x3+ a N4 X 4 +--+a NN X N
Here, Xi (i=1.2.-, N) is the density level of the pixel. f, (i=1.2.-1N) is the intensity of the frequency after conversion (conversion value), and the larger i is, the higher the frequency is. Also, a==(i, j= 1, 2, -,
N) is the coefficient of the variable. N is the size of the block, with N=2.

The orthogonal transformation can be obtained as fi by substituting the density value Xi of a given pixel into equation (1).

If there is no boundary within the block, the converted value can be obtained by using equation (1) as is. If a boundary exists within a block as shown in FIG. 8, a converted value is determined as follows.

FIG. 4 shows a one-dimensional pixel array made up of N pixels.

Now, in the pixel column of FIG. 4, Xl is the boundary of the area, and X, to X1 are inside the area, XL. Assuming that 1 to xN is the blank space to be filled in, in this case, instead of finding f to fN, fL. Substituting 0 for 1 to f, f l - fL
+xt-+~χ8 can be obtained. This is because this is an N-element simultaneous -dimensional equation with an unknown number N, so it can be arbitrarily determined by analytically solving equation (1).

This reduces the high frequency components of the converted value f.

Complementary values X, 41 to XN, such that ~f8 is set to 0, are determined.

In the example of FIG. 4, the pixels to be complemented are the continuous part in the latter half of the pixel column, but they may also be scattered within the pixel column.
Even in that case, for the number K of pixels to be complemented, the converted value f
, (for i=N-, -, K), the complementary value is determined by setting the number to O.

Although the processing for pixel rows in the horizontal direction has been described above, the complementation processing can be performed in the same way when complementing pixel rows in the contraction direction.

In the block forming section, by arranging the pixel columns complemented in this way in two horizontal or vertical directions,
Reconstruct the interpolated 2×2 block of pixels.

In this way, in the present invention, the orthogonal transform equation is analytically solved so that the high frequency component of the transform value of the orthogonal transform becomes O, and the complementary value is uniquely determined. Perform image completion.

This makes it possible to prevent the generation of high frequency components that are originally unrelated to the nature of the image and are caused by interpolation.

In the following, interpolation in the case of N=4 will be exemplified. The linear equation representing the orthogonal change at this time is: C4=cos
(π/4)/2. C3-c o S (π/8)
/2. 58=s i n (z/8)/2.

FIG. 5 is a diagram illustrating density levels of pixel columns.

Now, with the third pixel as the boundary, the density value of the pixel within the head is xl = 64. Xz=32. When x3=64, it is assumed that the fourth pixel is complemented.

At this time, if we calculate the concentration value X4 with f,=O, we get (2
), it becomes: The converted value f at this time is f,=106.50 f2--41. 81 f3=38.63 f4=0. 0, and it is sufficient to have data for three pixels.

On the other hand, when the converted value f is obtained by substituting the density value X4=64 using the conventional method, f, =79. 20 fz--6, 12 f:+=11.31 fn=14.78, and data for four pixels is required. At this time, if f4 = 0 and the data is reduced to 3 and the data is restored, the high frequency components of the data are removed and x, -58, 34 xz = 45. 66 X3=50.34 x4=69.66, resulting in a large error.

〔Example〕

FIG. 6 shows an embodiment of the present invention, and shows an image complementary encoding device that performs image encoding using area division.

This image complementary encoding device divides an input image into regions such as -, divides the resulting regions into blocks, and complements pixels in blank areas outside the regions within the blocks. , which encodes images.

In FIG. 6, reference numeral 11 indicates an image encoding unit that performs compression encoding of an image, 12 an image input unit, 13 an area division unit,
14 is a region extracting section, 15 is a block dividing section, 16 is a block classifying section, and 17 is an encoding section. Reference numeral 18 denotes an image complementation section, and the image complementation section 18 includes a block decomposition section 19, an analysis section 20, and a block formation section 21. 2
Reference numeral 2 indicates an image database that receives and transfers encoded data, and data is also received and transferred through a line 23. Reference numeral 24 indicates an image restoration unit that restores encoded data to an image. The image restoration unit 24 includes a decoding unit 25
, an area generation section 26 , and an image reconstruction section 27 .

In the image encoding section 11, an image input section 12 converts an analog image into a digital image and inputs the digital image into the apparatus.

As shown in FIG. 10, the region dividing unit 13 divides the input image into regions having similar properties. The region extracting section 14 cuts out each region from the image divided into regions by the region dividing section 13, and extracts a boundary line for each region. The region data and boundary line data extracted by the region extraction section 14 are sent to the block division section 15.

The block dividing unit 15 divides the area into rectangular partial images (blocks) having a minimum size that is an integral multiple of pixels. The block classification unit 16 refers to the boundary line of the area and classifies the blocks into blocks that include the boundary part of the area and blocks that do not include the boundary part of the area. In FIG. 11, blocks including boundary portions of regions are shown with diagonal lines.

Blocks that do not include area boundaries are sent to the encoding unit 17 as is. On the other hand, the block including the boundary portion of the area is sent to the image interpolation unit 18 together with the boundary line data. The image complementation unit 18 performs complementation of blank areas outside the area using the above-mentioned analytical means for the block including the boundary part of the area.

That is, the block decomposition unit 19 decomposes the block into pixel columns in the horizontal direction or contraction direction, and sends it to the analysis unit 20 together with data indicating the portion that needs to be complemented.

The analysis unit 20 solves the equation shown in (1) above to find a complementary value for the part that needs to be complemented, and substitutes it into that part. Block forming part 2
1, the horizontal or vertical pixel rows for which the interpolation process has been completed are collectively returned to the form of a block.

The encoding unit 17 encodes the blocks sent from the block classification unit 16 and the blocks sent from the image complementation unit 18. The encoding unit 17 encodes the block without distinguishing whether it is a block sent from the block classification unit 16 or a block sent from the image complementation unit 18.

The data encoded in this way is stored in the image database 22 together with the boundary line data.

In the image restoration unit 24, the decoding unit 25 decodes the encoded data extracted from the image database 22 in units of blocks, thereby reproducing an image including the area. The region generation section 26 applies boundary line data to the image decoded by the decoding section 25, and the image encoding section 11 extracts only the region and performs region division. Generates the same area as . The regions generated in this manner are collected in the image reconstruction unit 27 and combined at the original position of the image, thereby restoring the original image.

Note that instead of passing through the image database 22, the encoded data may be efficiently transmitted to a remote location via the line 23, and the image may be restored in the image restoration section 24.

〔Effect of the invention〕

As explained above, according to the present invention, in an apparatus that divides an image into blocks and encodes and decodes the image in units of blocks, high-frequency By suppressing the generation of components,
Efficient image encoding can be performed.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the principle configuration of the present invention, Section 2 (a),
(b) is a diagram explaining two-dimensional orthogonal transformation, and Figure 3 is 1×
FIG. 4 is a diagram showing a one-dimensional pixel row of N pixels, FIG. 5 is a diagram illustrating the density level of the pixel row, and FIG. A diagram showing an example,
FIG. 7 is a diagram explaining image block division, FIG. 8 is a diagram explaining energy concentration by discrete cosine transformation,
FIG. 9 is a diagram showing a block including the edge of the image, and FIG. 10(a). (b) is a diagram explaining the area division of an image, and FIG. 11 (a)
. (b) is a diagram showing block division of an area. 11 is an image encoding section, 18 is an image complementing section, 19 is a block decomposition section, 20 is an analysis section, and 21 is a block forming section.

Claims (1)

[Claims] An image encoding unit (1
1), a block decomposition unit (19) that decomposes a block including a boundary portion of an image region into pixel columns and generates data indicating the boundary of the region in the pixel column; The high frequency component is 0 from the value
an analysis unit (20) that analytically determines complementary values for pixels outside the area so that (21), and performs encoding using blocks that do not include the boundary portion of the image area and the complemented blocks from the image complementation unit (18). Image complementary coding device.