JP2000013608A - Image processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently compress/expand an image while maintaining image quality of a significant area designated by an operator in the case of compressing the image. SOLUTION: In this image compressing method that compress/expands image data, a significant area in an image is designated (S11), the significance of image data which are read using the distance between the position of the read image data and the significant area as a parameter is calculated (S12), the image data that are read in accordance with the calculated significance are subjected to encoding preprocessing (S13) and the compressibility or expansion rate of each pixel of the image data (S14) subjected to encoding preprocessing is changed (S15 and S16).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method in an image processing apparatus including an output device such as a facsimile and an image and video device such as a digital camera.

[0002]

2. Description of the Related Art Conventionally, various methods of image data compression technology have been considered for the purpose of shortening the transmission time and saving the capacity of a storage medium for storing data. The image data compression technology can be roughly divided into two types: a `` lossless compression method '', which compresses or decompresses image data as it is, and a `` lossy compression method, '' which truncates information during compression or decompression to reduce the size of the compressed data. There is. When the latter irreversible compression is performed, the decompressed image has a disadvantage that the image quality is deteriorated compared to the image before compression.
Further, the image compression method mainly comprises two processes of "pre-processing" and "encoding" of an image. This preprocessing is made up of “processing” and “sorting” of data. Here, data generated by "processing" of data is referred to as "processed data", data generated by pre-processing is referred to as "data before encoding", and encoded data is referred to as "compressed data stream".

[0003] In the current image compression method, the JPEG method, which is a standard image compression method, will be described as an example. The processing up to discrete cosine transform on the original image data is "processing" of the image data. The operation of scanning in the zigzag path order from information having a low spatial frequency component to information having a high spatial frequency component corresponds to “rearrangement”. The operation up to this point is “pre-processing”. Therefore, in the case of the above-described lossless compression method, there is no truncation of the image information, and thus the size of the “unencoded data” created in the preprocessing is the same as the size of the original image data. In the case of irreversible compression, the size of the “unencoded data” created in the preprocessing is smaller than the size of the original image data because information in the high frequency region is discarded.

Next, entropy coding is performed on the data before coding. The created “compressed data stream” is smaller than the data size of the original image even in the case of the lossless compression method. For example, when 100 pieces of data having a value of α are arranged in a row, two values of “α” and “100” are written in the compressed data stream instead of writing 100 pieces of α. On the receiving side, the received compressed data stream is composited in the reverse manner to the encoding method to obtain “data before encoding”, and the decompression is performed by performing the reverse processing of the preprocessing.
Get an image. That is, in the case of the lossless compression method, the size of the data before encoding is equal to that of the original image. That is, in order to perform "encoding" efficiently (to reduce the size of the compressed data stream as much as possible), it is important to handle data in preprocessing before encoding. The same can be said for the irreversible transformation. Specifically, ideally, the data before encoding is data in which the value does not change as much as possible when arranged one-dimensionally. Because, regardless of the encoding method used, if the value of the data changes, the information of the "change" must be added to the compressed data stream, and the data change amount is large and the change is large. This is because the information that is added frequently increases as the frequency of occurrence increases.

[0005] In the pre-processing, data "processing"
As an example of converting to spatial frequency information by a method, there is a method using a wavelet transform in addition to the discrete cosine transform described above. The information thus converted into the spatial frequency information is rearranged in the order of the spatial frequency mainly by zigzag scanning. The spatial frequency represents something like "fineness" of an image. Therefore, arranging the information of the spatial frequency components from “low” information to “high” information by zigzag scanning is equivalent to arranging the information in the image from a “rough” portion to a “fine” portion. In general, since a single image contains "rough" parts to "fine" parts evenly, and there is almost no case where the pixel value from there is drastically changed, the spatial frequency If the conversion and zigzag scanning methods are used, the conversion of the one-dimensionally arranged uncoded data is hardly severe.

Further, as a typical example of other pre-processing, there is a bit plane division method of dividing an 8-bit gradation image into bit planes and converting the image into binary images of 1 bit. This means that 8 bit data is 1b
It can be said that the amount of change in the value of the data before encoding is reduced by setting it to it. In many cases, raster scanning is mainly used for scanning data in the bit plane division method.

[0007] As described above, the most effective pre-encoding data for encoding is a small change in the value of the data when arranged one-dimensionally. That is, they have local correlation.

As an effective scanning method for creating one-dimensional data having local correlation, it is called "Peano scanning" as shown in FIGS. 20 (a), 20 (b) and 20 (c). Scanning methods have been proposed. This is a discretization of a “Peano curve” which is a kind of curve passing through all points in a certain area. Here, a curve, such as a Peano curve, that passes through all points in a certain area without overlap is referred to as a “fractal space filling curve”. Scanning methods obtained by discretizing such a fractal space filling curve include those shown in FIGS. 21 (a), (b) and (c) in addition to FIG. FIG.
FIG. 1 is a diagram illustrating a scanning method disclosed in Japanese Patent No. 70144.
In these scanning methods, unlike raster scanning, pixels in a continuous scanning order always exist at positions near each pixel, so that image processing and scanning that effectively use the correlation between pixels can be performed.

However, in the case of these scanning methods,
In some cases, the shape of the region where scanning can be performed is fixed. For example, in the case of a Peano scan, the base Peano
Since the curve is obtained using a recursive process based on the horizontal and vertical division into two, there is a drawback that only a square area having a length of two exponents on one side can be scanned.

As a technique for solving this problem, "Wyv
"ull". This is a technique of connecting relatively small scans to obtain a continuous scan as shown in FIG.

There is also a method of dividing an image proposed by "Cole" into an odd number of regions. FIG. 24 shows a scan using this technique. Also, FIGS. 25 (a) and (b)
Is proposed by "Skarbek" or the like, and is a method of preparing a square area larger than image data and scanning using Peano scanning. In addition to these methods, a method in which Nagae et al. Scan an arbitrary rectangular area has been proposed (Nagae, Yasuiin, Nagahashi: "Pean"
o Generalization of Scanning and Application to Halftone Processing ”, TV Gakuho, Vol. 16, No. 9, pp. 25-30 (1
992)). FIG. 26 shows scanning using this method. Here, this scanning method is referred to as “generalized Peano scanning”.
I will call it. Generally, these scanning methods can be classified into a cis type as shown in FIG. 27A and a trans type as shown in FIG. 27B.

On the other hand, in the irreversible image compression method, image information is truncated, but most of the information is currently automatically performed by a compression program. Taking the above-mentioned JPEG as an example, the truncation of image information is performed in order from high frequency components of spatial frequency components toward low frequency components. In the bit plane division method, bit planes are cut off in order from a high-bit plane, that is, a plane including high-frequency components as in JPEG. However, depending on the type of image such as a medical image, information in a high frequency region may be more important than information in a low frequency region. Further, a specific image portion such as a face portion in a portrait image may be important.

In other words, when the compression program automatically sets a portion of an image to be truncated when performing irreversible change, information in a portion different from that imagined by the operator may be truncated. is there.

[0014]

As one of the proposed image processing and compression methods using Peano scanning, there is an image processing method described in Japanese Patent Application Laid-Open No. 10-70664. In this method, an image is subjected to error diffusion processing using Peano scanning, and compressed image data is generated using a predictive coding method or a run-length coding method. Is not considered at all.

SUMMARY OF THE INVENTION It is an object of the present invention to efficiently compress / decompress an image while maintaining the image quality of the important area specified by the operator when compressing the image.

[0016]

According to the present invention, there is provided an image processing method for compressing / decompressing image data, in which an important area in an image is designated and the read image data is read. The importance of the read image data is calculated using the distance between the position and the important area as a parameter,
The feature is that the compression ratio or the expansion ratio of each pixel of the image data is changed according to the calculated importance. Therefore, the image can be efficiently compressed / decompressed.

According to another aspect of the present invention, in an image processing method for compressing / expanding image data, an important area in an image is designated, and a distance between the position of the read image data and the important area is used as a parameter. It is characterized in that the importance of each pixel of the read image data is calculated, and the number of quantization bits is assigned to each pixel of the read image data according to the calculated importance. Therefore, images can be efficiently compressed /
Can be extended.

Still another aspect of the present invention is an image processing method for compressing / expanding image data, in which an important area in an image is designated, and a distance between the position of the read image data and the important area is used as a parameter. The importance of the read image data is calculated, the calculated importance is quantized by a predetermined threshold, and a distribution map of the importance is created and held for the image data using the quantized importance. It is characterized in that the sampling pitch is determined based on the created and stored importance distribution map, and the image data is read out by the determined sampling. Therefore, the image can be efficiently compressed / decompressed.

The image data is divided into a predetermined area size and subjected to image processing. Therefore, the image can be efficiently compressed and decompressed at high speed.

Further, another aspect of the present invention is an image processing method for compressing / expanding image data, in which an important area in an image is designated, and the read image data is recursively divided into rectangular areas. The importance of each rectangular area is calculated using at least the distance between the rectangular area and the important area as a parameter, and it is determined whether the rectangular area is recursively divided again based on the calculated importance and the area size of the rectangular area. Then, when not dividing again, the sampling pitch in the rectangular area is determined according to the calculated importance for the rectangular area determined not to be divided, or the number of quantization bits is assigned according to the importance. It has special features. Therefore, the image can be efficiently compressed and decompressed at high speed.

The process of recursively dividing image data into rectangular regions is a recursive division process performed when creating a fractal space filling curve. Therefore, the image can be efficiently compressed and decompressed at high speed.

Image data is read out by an image scanning process performed by discretizing the fractal space filling curve. Therefore,
An image can be compressed / decompressed efficiently and at high speed.

The processing for calculating the importance is performed by using a distribution function in which the distance from the important area, the size of the spread of the important area, and the importance of the most important point, which is the most important point, are at least parameters. Calculate importance. Therefore, the importance can be easily set.

The process of designating an important region is performed on a plurality of regions, and the process of calculating the importance is individually performed on each important region, and new importance is calculated using the individually calculated importance. Calculate importance. Therefore, the importance can be set accurately.

The process of designating an important region is performed individually for the importance for the gradation of the image and the important region for the resolution. Therefore, it is possible to easily set the importance according to the quality characteristics of the image.

The distribution function is set for each of the horizontal direction and the vertical direction of the image, and the importance at an arbitrary position is calculated by synthesizing the importance individually calculated by the distribution function for the horizontal and vertical directions. . Therefore, optimal compression / decompression can be performed on important regions having various distributions.

The processing for calculating the importance is performed by using parameters such as the distance from the important area, the size of the spread of the important area, the magnitude of the importance at the most important point, and the inclination angle of the image with respect to the horizontal or vertical direction. Function. Therefore,
Optimum compression / decompression can be performed on an image having a more complicated distribution.

The pre-encoding process includes a first process of reading image data at a predetermined sampling pitch or a sampling pitch determined by the sampling pitch determination process, and a one-dimensional image data obtained by the first process. A second process of calculating a difference between adjacent pixels with respect to
The number of quantization bits is assigned to the image data obtained by the second processing according to a predetermined number of quantization bits or the number of quantization bits given by the quantization bit number assignment processing. Therefore, the change in the difference value is small because the local correlation is higher than the raster-read image data, so that the compression ratio / expansion ratio can be improved.

When performing image processing on a color image, the compression ratio or expansion ratio is changed for each color space in the color space of the color image data. Therefore, deterioration is not conspicuous, and the compression ratio / decompression ratio can be increased, and more effective compression can be performed.

[0030]

DESCRIPTION OF THE PREFERRED EMBODIMENTS In an image processing method for compressing / decompressing image data, an important area in an image is designated, and the distance between the position of the read image data and the important area is used as a parameter. Calculate the importance of the encoded image data, apply the pre-coding process to the read image data in accordance with the calculated importance, and compress or decompress each pixel of the pre-coded image data. To change.

[0031]

Embodiments of the present invention will be described below with reference to the drawings. The present invention is configured by a program based on the following, which is included in a processing device connected to a transmission path, generally a workstation or a personal computer.

FIG. 1 is a flowchart showing the principle of the image compression method of the present invention. First, when the operator designates an “important area” of image data (S11), importance = a · X + b ···, which is an importance calculation equation using a distance between an arbitrary pixel position of the image data and the important area as a parameter.・ (1)
(a, b: real number, X: distance from important area), "importance" at each pixel position is calculated (S12). Although various distribution functions such as a linear form and a normal distribution function as in the above equation (1) can be considered as the importance calculation equation, the present invention does not specify this. Also, the method of specifying the important area is not particularly specified.

Next, one or a plurality of pre-coding processes are performed in accordance with a method determined in advance according to the calculated "importance" at each pixel position (S13), and pre-coding data is generated (S13). S14). For example, a sampling (thinning-out) process of the read image is performed as a first pre-encoding process, a difference calculation process of determining the difference between adjacent sampled image data is performed in the second pre-encoding process, and As a third pre-encoding process, a process of adjusting the number of bits allocated to the difference value can be considered. But,
The present invention does not specify these encoding preprocessing methods.

Finally, the data before encoding is encoded (S1).
5) Then, an image compression data stream in which the image quality of the important area is maintained is obtained (S16). When the compressed data stream is decompressed, the opposite operation to that for compressing the image may be performed. At this time, for example, pixel interpolation is performed by a predetermined number from adjacent pixels. Various methods are conceivable for designating the important region, but the present invention does not specify such a method. Various methods are conceivable for the pre-coding process, and there are other pre-coding processes other than those described in the present invention, but the combination with them is not specified. Various methods such as run-length coding and vector quantization are conceivable as encoding methods, and any of them may be used.

FIG. 2 shows an importance distribution by the importance calculation in the present invention, and FIG. 3 is a flowchart showing the operation of the image processing method according to the first embodiment of the present invention. As shown in FIG. 3, in the first embodiment, the above-described importance designation processing (S31)
And one or more pre-encoding processes are performed according to the importance calculated by the importance calculation process (S32) (S33), and appropriate quantization is performed on each pixel of the image data for which the pre-encoding data (S34) can be obtained. Give the number of bits (S3
5) Encode (S36) and obtain an image compressed data stream that retains the image quality of the important area (S37).
As shown in FIG. 2, the distance between the center point or the most important point of the important area specified by the important area specifying process and an arbitrary pixel position of the image data is not specified for the image to be subjected to the compression encoding. Is a distribution of the importance calculated and quantized by the importance calculation formula using the parameter as a parameter, and the importance decreases from the area 1 to the area 5.
Although it does not prescribe the image reading method and the reading order, image data is read out by, for example, raster scanning, and importance calculation and quantization are sequentially performed by an importance calculation formula. If you have a very high importance, give a large number of bits. For example, if the data of the original image is represented by 8 bits, 8 bits of each color are assigned to each data of the area 1. Conversely, 4 bits are allocated to a low importance part such as the area 5, and an intermediate bit number of 8 to 4 bits is allocated to an intermediate important area between the areas 2 to 4, and so on. Performs a process of assigning a smaller or the same number of bits. As a result, although the compression efficiency is lower in the important area portion, the quality of the compressed / decompressed image is less deteriorated.

It should be noted that the series of processes of this embodiment can be executed without the need to create a distribution map of importance as shown in FIG. 2 in advance. The number of bits may be assigned while referring to the distribution map.

FIG. 4 is a flowchart showing the operation of the second embodiment of the present invention. First, the importance of each pixel is determined by the above-described importance designation processing (S41) and importance calculation processing (S42). Next, the importance of each pixel is quantized by an importance quantization process (S43), and an importance distribution is created (S44). This is held by importance distribution holding means such as a memory of a computer (S45). Further, a sampling pitch is set for each importance level (S46). FIG. 5 is a diagram showing an example in which importance levels are quantized into five levels. The original image data in this embodiment is read out while being sequentially thinned out according to a predetermined sampling pitch (S
47, S48), and the process of storing it in the memory is repeated. Therefore, unlike the first embodiment, it is necessary to hold a distribution map of importance. Various methods are conceivable for holding the quantized importance distribution map and the thinned image data, and any of them may be used. In any case, the image may be sampled according to the level of importance. Thereafter, each pixel of the image data for which the pre-encoding data (S49) is generated is encoded (S49).
An image compressed data stream in which the image quality of the important area is held together with 50) is obtained (S51).

Next, with respect to the image data to be subjected to the compression encoding, an area in the image determined by the operator to be important is designated by a radius r as a parameter indicating the most important point S (x, y) and its spread. An example in which this is done will be described with reference to FIG.
The procedure for specifying S (x, y) and r is not limited here. On the other hand, the image data is divided into rectangular areas having a preset size. At this time, a fraction may occur depending on the size of the image. In this case, the adjustment of the division size and the adjustment of the region size on either the left or right or the top and bottom are handled. In any case, the image data is divided into rectangular areas having substantially the same size.
Next, the importance for each area is calculated. An expression for calculating the importance uses, for example, a normal distribution function shown in Expression (2) described later. The present invention does not prescribe a method of calculating the importance of each rectangular area. For example, if coordinates of four vertices of the rectangular area are A, B, C, and D, vertices A,
The importance in B, C, and D can be easily obtained using the distance between each vertex and the most important point S. Vertices A, B,
G1, G2, G3, G
If it is set to 4, the importance G of the area may be obtained as an average value G = (G1 + G2 + G3 + G4) / 4 of G1 to G4. When the importance of the area is determined, the sampling pitch or the number of quantization bits is determined according to the importance. For example, when the importance is standardized from 0 to 1 and 1 is the value at the most important point, the correspondence as shown in FIG. 7 may be determined in advance.

Next, in the present embodiment, an area in the image determined to be important by the operator is used as a parameter indicating the most important point S (x, y) and the spread of the image data to be subjected to compression encoding. It is assumed that the radius r is specified. The procedure for specifying S (x, y) and r is not limited here. On the other hand, image data is divided by recursive division means. For example, consider a case where a predetermined area is divided into two in the horizontal and vertical directions as shown in FIG. The meaning of recursive division means that each of the four regions obtained by dividing into two parts in the horizontal and vertical directions is again divided into two parts in the horizontal and vertical directions.
That is, the process of dividing is repeated. At this time,
If the vertical or horizontal size of the image is an even number, equal division is possible, but if it is an odd number, equal division is not possible. However, it is only necessary to perform division so that the size of the area after division becomes even and odd. Also, the number of times the division processing has been performed is called a rank, and for example, a small area obtained by one division processing is called a rank 1 area. Further, the four regions obtained by the n-th division are respectively denoted by n / 1, n / 2, n / 3, n
/ 4.

FIG. 9 is a flowchart showing the procedure of recursive division according to the third embodiment of the present invention. First, the entire given image data is divided into two parts in the horizontal and vertical directions (S91, S92, S93). At this time, the rank is 1, and the divided areas are 1/1, 1/2, 1/3, 1 /
It is represented as four areas (S94, S95, S96). First,
The importance is calculated for the 1/1 area (S97). The method of calculating the importance of the region may be the same as the method described in the second embodiment, for example. Also, it is determined whether the 1/1 area is further divided based on the calculated importance G and the area size (S98, S99). The judgment is made in the following manner. That is, the calculated importance is classified into a plurality of levels by a correspondence table as shown in FIG. Further, the size of an area for determining whether to perform further division is set for each level (S100). Therefore, when the importance calculated for an area whose area size is N pixels is level L and the number of pixels for determining whether or not to perform recursive division at level L is M, if N ≦ M , No division is performed. Conversely, if N> M, division is performed.

If it is determined that the n / 1 area is not to be further divided, the importance is calculated for the n / 2 area and the determination is made (S101). If the determination result of the n / 2 area indicates that division is to be performed, further division is performed,
The above-described importance calculation and determination are repeated for the divided region n + 1/1. In addition, when all the regions of n ranks 1 to 4 do not perform further division, the region returns to the region one rank before, and the calculation and determination of the importance of the n-1 / 2 region are repeated. Further, the upper limit of the rank is, of course, when the number of vertical or horizontal pixels is reduced to 2 pixels or less. At this time, no further division is performed. The above processing is repeated until it is determined that no further division is performed in all the areas.
The importance of each divided area is assigned a level as shown in FIG. 10, and the sampling pitch or the number of quantization bits is determined for each level as in the description of the second embodiment. The reading method and the reading order for reading the data in the divided picture area are not limited at all.

Next, a feature of the fourth embodiment according to the present invention resides in that a recursive division process performed when a fractal space filling curve is obtained is used in the division process performed in the third embodiment. As described above, the use of the generalized Peano scan by Nagae et al. Enables Peano scan for an arbitrary rectangular area. Also, recursive 2
Scanning obtained by division is called binary scanning, and scanning obtained by recursive division is called ternary scanning. Therefore, the process of dividing the space into two in the horizontal and vertical directions, that is, the process of dividing the space into four, used in the third embodiment is equivalent to the quaternary division. However. The quaternary division cannot be applied to an area composed of odd-numbered pixels both vertically and horizontally. In such a case, recursive division may be performed on an area excluding the last one row, compression processing may be performed, and different compression may be performed on the last row. Note that the present invention does not limit which scanning method is used. The method of calculating the importance of the divided region, the method of determining whether to continue the recursive division, and the allocation of the sampling pitch or the number of quantization bits are the same as those in the third embodiment. Further, in this embodiment, the final scanning of the image data is performed according to a scanning method (Peano scanning) obtained by quantizing the fractal space filling curve. Therefore, it is necessary to perform a division process at least once for obtaining all the scanning orders. Peano
Image data is read out by scanning, and at this time, the reading is performed according to the sampling pitch determined for each area to be read out. If a predetermined number of quantization bits is assigned to each pixel, the processing is performed after the reading.

Next, FIG. 11 is a view for explaining a fifth embodiment of the present invention. As can be seen from the figure, first, the operator specifies an important area (shaded area in FIG. 4) for the target image. Then, for the designated important region, a circle and a radius approximating the region and a center point of the circle are determined. As a method of designating the important region, a method in which the operator designates the most important point and the radius of the image can be considered, but any of them may be used.

The importance of each pixel can be represented by using, for example, a normal distribution function f (xij) shown in the following equation (2). In the equation, σ is a standard deviation, μ is an average value, and a distance from the average value is xij. K is a normalized constant.

[0045]

(Equation 1)

Here, the state of f (x) is shown in FIG.

If the central point of the important region is regarded as μ and the distance from the central point is regarded as x, the importance of an arbitrary pixel can be obtained using the above equation (2). Further, since the spread of the distribution of the distribution function f (x) can be adjusted by making σ proportional to the radius of the important region, it is possible to obtain an appropriate degree of importance according to the size of the spread of the important region. it can. Also, the magnitude of importance at the most important point (x = μ) is
This is performed by giving an appropriate value to the normalized constant K. As the distribution function, various functions other than the normal distribution function are conceivable, and any of them may be used. In any case, according to the present embodiment, an important distribution area is defined as a center at one point in the area, and an arbitrary distribution function having a parameter indicating the distance from the center and the extent of the spread and a parameter indicating the intensity at the center is defined as the important area Is calculated by calculating the degree of importance for an arbitrary point of the image data.

Next, a sixth embodiment according to the present invention will be described with reference to FIGS. 13 and 14. FIG. More specifically, FIG. 13 illustrates a state in which a plurality of importance regions (region 1 and region 2) are specified in the image. FIG. 14 shows the importance at an arbitrary position (x, y) in the image, and shows the distance from the position (x, y) to the center position of the importance region 1 and the importance region 2 on the horizontal axis. is there. As shown in FIG. 14, the importance at the position (x, y) in the importance area 1 is α (x,
y), and the position (x,
y) is calculated as the importance β (x, y). The final importance P (x, y) for the position (x, y) is, for example, the sum P (x, y) of the importance calculated for each important area.
y) = (α (x, y) + β (x, y)). The same applies to the case where there are two or more important regions, but the same applies. In the present embodiment, the sum of the respective degrees of importance is used for the calculation of the new degree of importance. However, other various operations are conceivable, but any of them may be used. In any case, a new importance distribution is calculated from the respective importance distributions of the plurality of important regions.

FIG. 15 is a diagram relating to a seventh embodiment of the present invention. As shown in the figure, in the image, there is an area 1 that emphasizes resolving power that emphasizes the reproduction of fine structures such as buildings and leaves of trees, and subtle changes in color tone such as paint colors of people's faces and cars. If the image characteristics to be important are different, such as the region 2 where the importance of reproduction is important, such as gradation and color reproducibility, and if they are mixed, an arbitrary position P (x,
The importance of y) is individually set according to each image characteristic. For example, the importance at an arbitrary position P (x, y) is calculated by an importance α (x, y) calculated by the importance calculating means using the distance between the position P (x, y) and the area 1 as a parameter.
y), importance β (x, y) calculated using the distance to region 2 as a parameter is set. And encoding means 1
, The sampling pitch is determined according to the importance α (x, y), and the encoding means 2 determines the number of quantization bits as importance β
(X, y). The same applies to a case where there are two or more important regions having different image characteristics. In the present embodiment, the resolution and the number of gradations are used as the image characteristics. However, various other image characteristics are conceivable, but any of these may be used. In any case, it is an object of the present invention to perform pre-encoding processing according to the importance of a plurality of important regions having different image characteristics according to the characteristics.

FIG. 16 is a diagram related to the eighth embodiment of the present invention. Although the distribution function of the eighth embodiment shown in the figure is a distribution function on a concentric circle from the center of the important region, the specified important region is not always circular.
Therefore, for example, two distribution functions (fx (x, y), fy) having different standard deviations of the distribution functions and being orthogonal to each other.
(X, y)) is used to create a new distribution function F (x, y). The contour distribution of the importance of F (x, y) is as shown in FIG. In this embodiment, two distribution functions (fx (x,
Although y) and fy (x, y)) have different standard deviations of distribution, various other image characteristics and differences in equations can be considered. Either one of them may be used, but in any case, a new distribution function is created using two orthogonal importance distribution functions having different function characteristics.

Next, FIG. 17 is a diagram showing a ninth embodiment according to the present invention. In the figure, a new distribution function F
In the contour distribution of the importance of (x, y), the short axis of the ellipse is oriented in the horizontal direction of the image, and the long axis is oriented in the vertical direction. Here, if θ is set as shown in FIG. 17 using the angle parameter θ, F (x, y) can have an axis in an arbitrary direction. There are various methods for setting the angle parameter θ, and any of them may be used. In any case, the distribution function of importance in the present embodiment has a degree of freedom in the direction of the rotation axis.

FIG. 18 is a flowchart showing the procedure of the tenth embodiment according to the present invention. First, certain importance is determined for an image by the importance calculation processing (S180).
1, S1802, S1803), according to the second embodiment of the present invention, the image data is sampled by the first pre-coding process (first pre-coding data) (S18).
04a). At the same time, the size, position information, and importance of each pixel or divided area are stored in the memory. Next, the first pre-encoding data is left unchanged from the first data by the second pre-encoding process, and a difference from adjacent data is obtained from the second data, thereby obtaining sampled difference data. (Second pre-encoding data) (S1804b). Here, when the image data is called out using the fractal space filling curve of the third embodiment according to the present invention, the local correlation is high and the respective difference values are small. Next, by a third pre-encoding process,
The size, position information, and importance of each pixel data or each divided area stored in the memory are called, and the number of quantization bits is assigned to the second pre-coding data (the third pre-coding data). (Processing data) (S1804
c). The third encoded data is encoded by a predetermined encoding method (S1805, S1806), and becomes a compressed data stream (S1807). In the present embodiment, the difference between adjacent pixels is obtained in the second pre-encoding process, but any other efficient method for data having local correlation may be used.

Next, FIG. 19 is a flowchart showing the procedure of the eleventh embodiment according to the present invention. When performing compression on a color image, for example, before compressing the image,
If the color space is converted into a color space taking into account human visual characteristics such as the IELAB color space, the image quality is improved even with the same compression ratio. Generally, human vision is sensitive to a luminance plane such as Y, but has a characteristic that a color difference plane such as UV is not so sensitive. Utilizing this property, a color image is converted into a color space, and the image is truncated aggressively with respect to information on a color difference plane. Such an operation is called sampling. In general, full sampling of Y information is performed, and about half of the Y information is sampled for UV information. Here, especially when the image data is read out using the fractal space filling curve of the third embodiment, since the local correlation is high, the UV information is about half of the Y information as described above. Even if sampling is performed, an image with less color degradation can be obtained by expanding the UV information using the adjacent pixel data for the compressed data stream on the decoding side. In this embodiment, a YUV color space is used (S1901 to S1908, S19
11 to S1918), there are various other color spaces. Either of them may be used. Either way,
Efficient compression can be performed by changing the degree of compression for each color space of color image data using human visual characteristics.

[0054]

As described above, according to the present invention,
Since the compression ratio varies depending on the importance in the image,
Optimum compression that minimizes the loss of required image information can be performed.

Since the assignment of the number of quantization bits is changed in accordance with the degree of importance in an image, optimal compression can be performed on image data in which resolution is important without changing the resolution.

Further, since the image data is thinned out at a sampling pitch according to the degree of importance in the image, optimal compression is performed on the image data in which the gradation is important without changing the gradation. be able to.

Further, since the pre-encoding processing such as the allocation of the sampling pitch or the number of quantization bits of the importance is performed on the area divided into the predetermined size, it is not necessary to prepare the importance distribution map in advance. . Since the same processing is performed on the divided image data stored in the memory,
The processing is fast and the system configuration is simple.

Further, since the division processing for recursively dividing the image data is used, the number of times of performing the division, the calculation of importance, and the selection of the processing according to the importance can be reduced.

Further, since the more important the area, the finer the setting, the more optimal compression / expansion can be performed.

Since the reading of the image data is performed in accordance with the scanning method in which the fractal space filling curve is discretized, the local correlation of the read and one-dimensionalized image data is high, and a compression method using this is possible. Also, the recursive division is efficient because the recursive division performed at the time of calculating the fractal space filling curve is used.

In calculating the importance, the importance distribution can be represented using a distance from the important region, the spread of the important region, and a distribution function using at least the importance of the important region as a parameter. Since an optimal importance distribution can be set, more optimal image compression / decompression can be performed.

By permitting the setting of a plurality of important regions, and setting the importance at an arbitrary point in the image as a new importance calculated from the plurality of importance that each important region makes at a point of interest,
Appropriate compression / decompression processing can be performed on an image having a more complex distribution of importance.

Since the designation of importance in an image is set for each of image characteristics such as resolution and gradation, image compression / encoding more suitable for sensitivity can be performed.

Since the importance is calculated using different distribution functions in the horizontal direction and the vertical direction of the image data, for example, a person arranged in a central portion in a portrait image in a portrait image or a mountain range distributed in a horizontal direction is used. For example, optimal compression / decompression processing can be performed on important regions having various distributions.

Since the distribution function has a parameter of the inclination from the horizontal direction or the vertical direction, optimal image compression / decompression can be performed on an image having a more complicated distribution.

Further, since the difference between adjacent image data is obtained with respect to image data read according to a reading method obtained by discretizing a fractal space filling curve, the difference is more local than image data read by raster. Is high, the change in the difference value is small,
The compression ratio can be improved.

The compression ratio of each point in the image set in accordance with the importance calculated by the importance calculation is different in each color plane of the color space of the image data. Can be reduced, the compression ratio can be increased, and more effective compression can be performed.

[Brief description of the drawings]

FIG. 1 is a flowchart showing the procedure of the principle of the image processing method of the present invention.

FIG. 2 is a diagram showing an importance distribution by importance calculation according to the present invention.

FIG. 3 is a flowchart illustrating a procedure of an image processing method according to the first embodiment of the present invention.

FIG. 4 is a flowchart illustrating a procedure of an image processing method according to a second embodiment of the present invention.

FIG. 5 is a diagram illustrating an example in which importance levels are quantized into five levels.

FIG. 6 is a diagram illustrating an example in which a radius is designated as a parameter indicating the most important point and its spread.

FIG. 7 is a diagram showing a correspondence relationship among importance, sampling pitch, and number of quantization bits.

FIG. 8 is a diagram showing a case where a predetermined area is divided in the horizontal and vertical directions.

FIG. 9 is a flowchart showing a procedure of recursive division according to the third embodiment of the present invention.

FIG. 10 is a diagram illustrating a correspondence relationship among importance, area size, sampling pitch, and number of quantization bits of each divided area.

FIG. 11 is a diagram illustrating a fifth embodiment of the present invention.

FIG. 12 is a diagram showing a state of f (x).

FIG. 13 is a diagram illustrating a state where a plurality of importance areas are designated in an image according to a sixth embodiment of the present invention.

FIG. 14 is a diagram showing the importance at an arbitrary position in an image in the sixth embodiment according to the present invention, with the horizontal axis representing the distance from the position to the center position of importance region 1 and importance region 2; is there.

FIG. 15 is a diagram related to a seventh embodiment according to the present invention.

FIG. 16 is a diagram showing an eighth embodiment according to the present invention.

FIG. 17 is a diagram showing a ninth embodiment according to the present invention.

FIG. 18 is a flowchart showing a procedure of a tenth embodiment according to the present invention.

FIG. 19 is a flowchart showing a procedure of an eleventh embodiment according to the present invention.

FIG. 20 is a diagram showing Peano scanning as a conventional example.

FIG. 21 is a diagram showing another scan described as a conventional example.

FIG. 22 is a diagram showing a scan disclosed in Japanese Patent Application Laid-Open No. 6-70144.

FIG. 23 is a diagram showing scanning of Wyvil and the like, which is mentioned as a conventional example.

FIG. 24 is a diagram showing Cole scanning as a conventional example.

FIG. 25 is a diagram illustrating scanning of Skarbek and the like cited as a conventional example.

FIG. 26 illustrates a generalized Peano scan.

FIG. 27 is a diagram illustrating cis-type scanning and trans-type scanning.

[Explanation of symbols]

 S11 Important area designation processing S12 Importance calculation processing S13 Pre-encoding processing S14 Pre-encoding data S15 Encoding processing S16 Compressed data stream

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5C059 LC09 MA43 PP16 TA06 TA53 TC02 TC06 TC24 TC31 TC43 TD05 UA02 UA05 UA38 5C078 AA09 CA02 CA03 CA31 DA00 DA01 DA02 DA21 DB04 DB07

Claims (14)

[Claims] In an image processing method for compressing / expanding image data, an important area in an image is specified, and the distance between the position of the read image data and the important area is used as a parameter. An image processing method comprising: calculating importance of image data; and changing a compression ratio or an expansion ratio of each pixel of the image data according to the calculated importance. 2. An image processing method for compressing / decompressing image data, wherein an important area in an image is designated, and the distance between the position of the read image data and the important area is used as a parameter. An image processing method comprising: calculating the importance of each pixel of image data; and assigning a quantization bit number to each pixel of the image data read according to the calculated importance. 3. An image processing method for compressing / decompressing image data, wherein an important area in an image is designated, and the distance between the position of the read image data and the important area is used as a parameter. Calculates the importance of image data, quantizes the calculated importance with a predetermined threshold, creates and holds a distribution map of importance for image data using the quantized importance, creates and holds An image processing method comprising: determining a sampling pitch based on a determined importance distribution map; and reading out image data by the determined sampling. 4. The image processing method according to claim 1, wherein image processing is performed by dividing the image data into a predetermined area size. 5. An image processing method for compressing / decompressing image data, wherein an important area in an image is designated, and the read image data is recursively divided into rectangular areas. Calculate the importance of each rectangular area using at least the distance to the area as a parameter, determine whether to recursively divide the rectangular area based on the calculated importance and the area size of the rectangular area, and do not divide again In some cases, for a rectangular area determined not to be divided, a sampling pitch in the rectangular area is determined according to the calculated importance, or a quantization bit number is assigned according to the importance. Image processing method. 6. The image processing method according to claim 5, wherein the process of recursively dividing the image data into rectangular regions is a recursive division process performed when creating a fractal space filling curve. 7. The image processing method according to claim 5, wherein image data is read out by an image scanning process performed by discretizing the fractal space filling curve. 8. The processing for calculating the importance level includes a distribution in which a distance from the important area, the size of the spread of the important area, and the importance degree of the most important point, which is the most important point, are at least parameters. The image processing method according to claim 1, wherein the importance is calculated using a function. 9. The process of designating an important region is performed for a plurality of regions, the process of calculating the importance is performed individually for each important region, and the individually calculated importance is calculated. The image processing method according to claim 1, wherein the new importance is calculated using the new importance. 10. The image processing method according to claim 1, wherein the process of designating the important region is performed separately for the importance for the gradation of the image and the important region for the resolution. 11. The distribution function is set for each of a horizontal direction and a vertical direction of an image, and the importance at an arbitrary position is obtained by synthesizing the importance calculated individually by the distribution functions for the horizontal and vertical directions. Claim 1 to be calculated
11. The image processing method according to any one of items 10 to 10. 12. The processing for calculating the importance includes parameters such as a distance from the important area, a size of the spread of the important area, a degree of importance at the most important point, and an inclination angle of the image with respect to the horizontal or vertical direction. The image processing method according to claim 1, wherein the distribution function is: 13. The pre-encoding process includes a first process of reading image data at a predetermined sampling pitch or a sampling pitch determined by the sampling pitch determination process, and a one-dimensional process performed by the first process. Processing for obtaining a difference between adjacent pixels with respect to the obtained image data, and a quantization bit number given by a predetermined number of quantization bits or a number of quantization bits to the image data obtained by the second processing. The image processing method according to claim 1, wherein the number of quantization bits is assigned according to the number. 14. The image processing method according to claim 1, wherein when performing image processing on a color image, a compression rate or an expansion rate is changed for each color space of a color space of color image data.