JP2000165664A - Resolution converter for image and resolution conversing method for image - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize magnification/reduction of an image being a multiple of an optional rational number with high image quality in a small processing time. SOLUTION: A multiple U up-sampler 1 applies up-sampling to a received image 100 and the processed image is fed to a digital filter 2. The digital filter 2 has a filter characteristic so as not to cause distortion for the up-sampling of a multiple U and down-sampling of a multiple of 1/D at the post-stage so as to conduct a filter arithmetic operation while excluding the redundancy. An output 102 from the digital filter 2 is fed to the down-sampler 3, where the image is down-sampled to a multiple of 1/D.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image resolution converting apparatus and method for enlarging / reducing an image at an arbitrary rational magnification.

[0002]

2. Description of the Related Art When a function of zooming a thumbnail image or a digital image, which is frequently used in an electronic still camera or a printer, is realized, resolution conversion of the image is required.

According to the conventionally used technology, simple enlargement of an integer multiple or reduction of a fraction of an integer can be realized relatively easily, but resolution conversion of an arbitrary rational number is complicated. Many require processing.

[0004]

That is, in the prior art, the simple enlargement of an integer multiple or the reduction of a fraction of an integral number as described above merely increases the resolution of the original image by an integer. 1 size (eg, 1/2
１ ／), and smooth zooming cannot be performed. The same applies to enlargement.

[0005] Further, even if the resolution conversion can be performed at an arbitrary rational multiple, there remains a problem that distortion is conspicuous depending on the magnification or that a very long processing time is required.

SUMMARY OF THE INVENTION The present invention has been made in view of the above-mentioned problems, and has been made in view of the above-mentioned problems, and is an image resolution conversion apparatus capable of easily realizing enlargement / reduction of an arbitrary rational number with high image quality, that is, with a short processing time. And a method.

[0007]

In order to solve the above-mentioned problems, an image resolution conversion method and apparatus according to the present invention are provided.
By up-sampling the input image by a factor of U (U is an integer), filtering the up-sampled image, and down-sampling the filtered image by 1 / D (D is an integer), It is characterized in that resolution conversion involving enlargement / reduction of U / D times is performed.

Here, digital filtering means for changing a transfer function according to a resolution conversion magnification so as not to cause distortion is used for the filtering. The transfer function of this digital filter means is preferably determined by the up-sample magnification U and the down-sample magnification D. More specifically, the transfer function H _U ( z) and the transfer function H _D (z) of the filter for preventing aliasing at the time of down-sampling, a combined transfer function H _UD (z) (= H _U (z) H
_D (z)).

When the value of the magnification U of the up sample is larger than the value of the magnification D of the down sample,
The filtering is preferably performed using a zero-order hold method and a down-sampling method. When the value of the magnification D of the down-sample is larger than the value of the magnification U of the up-sample, the filtering is performed. Preferably, the processing is performed using a zero-order hold method and an averaging method.

In this arbitrary rational resolution conversion,
Upsampling has the effect of increasing the resolution of the image, for example, by zero-order hold means. Downsampling has an effect of calculating the resolution of an image by, for example, a thinning process or an average value calculating process for a plurality of pixels. The digital filter means has a function of setting a transfer function in accordance with each value of the enlargement magnification U and the reduction magnification 1 / D, and performing a convolution operation by multiplying a filter coefficient for each pixel.

Further, the digital filter means has a function of limiting a band so that imaging or aliasing does not occur by upsampling or downsampling.

[0012]

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing an embodiment of an image resolution conversion apparatus and method according to the present invention.

A first aspect of the image resolution conversion apparatus according to the present invention.
1 is shown in FIG. The image resolution conversion apparatus according to the first embodiment of the present invention shown in FIG. 1 includes a U-fold upsampler 1 for upsampling an input image 100 U times (U is an integer) and an upsampler 1 for upsampling. A digital filter 2 for filtering the filtered image 101 and a 1 / D-fold downsampler 3 for downsampling the filtered image 102 to 1 / D (D is an integer). According to the configuration of FIG. 1, the input image 100 is subjected to resolution conversion to U / D times and output image 103
Is taken out as

It is preferable that the transfer function of the digital filter 2 is determined by the magnification U of the up-sample and the magnification D of the Dun-sample. Further, the transfer function of a filter for removing an imaging component which is a spectrum component newly generated by U-fold upsampling is represented by H
_When the transfer function of an anti-aliasing filter for removing aliasing distortion due to _U (z) and 1 / D times downsampling is denoted by H _D (z), the transfer function H _UD (z) of the digital filter 2 is: It is more preferable that H _UD (z) = H _U (z) H _D (z) to eliminate redundancy.

When the value of U is larger than the value of D, it is preferable to use a zero-order hold circuit and a down-sampling circuit as the digital filter 2, and the value of D is larger than the value of U. In this case, it is preferable to use a zero-order hold circuit and an averaging circuit as the digital filter 2.

Here, four techniques which are basic techniques used in each embodiment of the present invention will be described. Specifically, a zero-order hold method for image enlargement, a linear interpolation method,
There are four methods of downsampling and averaging for image reduction.

First, as a first method, the zero-order hold method is known as the simplest method, and when considering a one-dimensional pixel row, an N-fold enlargement means that N-1 pixels are located in the vicinity of the pixel. This corresponds to a process of interpolating pixels of the same value. Although the zero-order hold method is simple in processing, it has a problem that when the enlargement ratio is large, the image quality becomes discontinuous in a block shape.

FIG. 2 shows a specific example of a three-fold enlargement of an image by the zero-order hold method. FIG. 2A shows a part of the original image, for example, a portion of 3 × 3 pixels in the vertical and horizontal directions. In the case of 3 times enlargement (9 times in area ratio), one pixel is used. By arranging 3 × 3 pixels of the same value in the vertical and horizontal directions, the 3 × 3 portion of the original image has 9 × 9 corresponding portions of the enlarged image as shown in FIG. Pixels. FIG. 2C shows a portion L of one horizontal row of the 9 × 9 pixel portion of the enlarged image, and takes the pixel value on the vertical axis. In FIG. 2 (C), the dashed circles indicate the pixels interpolated by the zero-order hold, and two pixels of the same value next to each pixel (solid circle in the figure) of the original image. Pixel P is interpolated.

The second linear interpolation method is also called a bilinear interpolation method, and can generate a relatively smooth enlarged image. In this method, pixels to be interpolated with enlargement are generated by linear approximation between pixels of adjacent original images.

FIG. 3 shows a specific example of triple-magnification of an image by the linear interpolation method. FIG. 3A shows a portion of the original image, for example, a 3 × 3 pixel portion in the vertical and horizontal directions.
At the time of triple magnification, as shown in FIG. 3B, an interpolated pixel P is arranged between these pixels to form an enlarged image portion of 9 × 9 pixels vertically and horizontally. As shown in FIG. 3C, the interpolated pixel P is obtained by linearly approximating adjacent pixels of the original image. Therefore, in order to obtain an interpolated pixel following the rightmost or lowermost pixel of the original image, the interpolation pixel P is further obtained. An outer pixel Q is required. Therefore, when the number of pixels is strictly enlarged to an integral multiple, a linear approximation is performed by adding a temporary pixel Q in order to generate an interpolation pixel at the end. The added pixels are usually selected in consideration of the smoothness of the image so as to have a symmetrical relationship with respect to the pixel at the end of the image (see FIG. 3C), and are removed after linear interpolation. You. With the linear interpolation method, a smooth enlarged image can be obtained, but on the other hand, the edge portion tends to lack sharpness.

Next, the third down sampling method is known as the simplest method. Considering a one-dimensional pixel array, a 1 / N reduction by the down-sampling method is N
This corresponds to a process of selecting a pixel for each pixel and reconstructing an image. Further, while this method is simple in processing, if it is performed on an image containing many high-frequency components, aliasing may occur.

FIG. 4 shows an image 1 obtained by the down-sampling method.
It shows a specific example of / 3 reduction. In the 9 × 9 pixel portion of the original image shown in FIG. 4A, pixels are selected for every three pixels, and the remaining pixels R are thinned out to obtain a pixel as shown in FIG. A reduced image portion of 3 × 3 pixels vertically and horizontally is formed.

Next, the fourth averaging method will be described. In the case of a one-dimensional pixel array, the reduction of 1 / N by the averaging method calculates an average value for every N pixels, and calculates the value. Is a process of reconstructing an image using pixel values of the reduced image as pixel values. In addition, since the averaging process corresponds to a low-pass filter, occurrence of aliasing (aliasing) can be suppressed.

FIG. 5 shows a specific example of 画像 reduction of an image by the averaging method. 5 × 9 pixels of the original image shown in FIG.
For each pixel area, take the average of the nine pixel values in that area,
By using each pixel value of the reduced image, a portion of the reduced image of 3 × 3 pixels vertically and horizontally as shown in FIG. 5B is formed.

Next, before describing the resolution conversion method of a U / D-fold image, a U-fold enlargement process and a 1 / D-fold reduction process will be described below. First, the U
The double magnification method is described.

FIG. 6A shows a basic configuration for this U-fold image enlargement. The U-fold up sampler 6 and the digital filter 7 whose transfer function is H _U (z) are shown in FIG. Be composed. The transfer function H _U (z) of the digital filter 7 is z
The impulse response of the filter, which is expressed on the axis (z = exp (jωT)), is expressed on the time axis by h _U (n). n is the number of samples (sample number, sample position) on the time axis. Similarly, the input image 104 is x (n), the output 107 of the upsampler 6 is c (n), and the output 108 of the digital filter 7 is y (n). Next, the operation will be described.

The input image 104 is a U-fold upsampler 6
And the image is upsampled by a factor of U. FIG. 7 specifically shows the upsampling operation at this time. As shown in FIG. 7, the U-fold upsampling means an operation of inserting (U-1) zero values.

That is, the original image portion of 3 × 3 pixels in the vertical and horizontal directions in FIG. 7A is triple-upsampled by the triple upsampler 6a shown in FIG. As shown in C), two 0 values are inserted between pixels in the one-dimensional direction, and eight 0 values are inserted per original pixel in the two-dimensional direction. Therefore, a 3 × 3 pixel portion of the original image is converted into a 9 × 9 pixel portion in the enlarged image.

The upsampled image 107 thus generated is input to the digital filter 7 of the transfer function H _U (z) and filtered, and as a result, a U-fold enlarged image 1 is obtained.
08 is output.

Next, characteristics of the transfer function H _U (z) to be satisfied by the digital filter 7 will be described. In order to perform an ideal U-fold enlargement, a low-pass filter must be designed in consideration of a change in the frequency band of a signal due to upsampling. FIGS. 8A and 8B show a frequency band X (z) of the original signal and a frequency band X _U (z) after upsampling by a factor of U, respectively. Here, ω is a normalized angular frequency. The broken line in FIG. 8B indicates an imaging component (spectral component newly generated by upsampling), and it is necessary to remove this component in order to obtain an enlarged image. Therefore, by using the ideal low-pass filter H _U (z) shown in FIG. 9A, an enlarged image having the frequency band Y _U (z) shown in FIG. 9B can be obtained. As described above, as the digital filter 7 of the transfer function H _U (z), a filter having the characteristics of the ideal low-pass filter H _U (z) shown in FIG. 9A is selected.

Next, a 1 / D-fold reduction processing method will be described. FIG. 6B shows a basic configuration for reducing the image by a factor of 1 / D. The digital filter 4 has a transfer function of H _D (z), the downsampler 5 has a factor of 1 / D. Consists of As an expression on the time axis, digital filter 4
Let h _D (n) be the impulse response of
(n), the output 105 of the digital filter 4 is p (n), and the output 106 of the downsampler 5 is y (n). Next, the operation will be described.

The input image 104 is first input to the digital filter 4 and a filter output 105 is output. The filter output 105 is down-sampled by the down-sampler 5 to 1 / D times. The downsampling of 1 / D times means an operation of extracting (D-1) pixels. 1 / D-fold reduced image 106 generated by this
Is output.

FIG. 10 specifically shows such a downsampling operation with D = 4 as an example. In FIG. 10, a 1/4 down sampler 5a shown in FIG. 10B corresponds to a 4 × 12 pixel portion of the original image shown in FIG.
Three pixels are extracted per pixel, and in two dimensions, one pixel is left out of 4 × 4 pixels and the remaining pixels R are removed, and a reduced image as shown in FIG. The image is generated as a 3 × 3 pixel portion in the vertical and horizontal directions of the image.

Next, characteristics of the transfer function H _D (z) to be satisfied by the digital filter 4 will be described. To perform an ideal 1 / D times reduction in order to avoid overlap of the band in the frequency domain by downsampling (aliasing), the band limitation by a low-pass filter H _D (z) with respect to the original signal Must be done. 11 (A),
(B) shows the frequency characteristics of the frequency band X (z) of the original signal and the low pass filter H _D (z), respectively. H _D (z) by the frequency band of the after receiving the band limited X _L (z) is as shown in FIG. 12 (A). Further, 1 / D
Frequency band Y _D (z) after double downsampling
Is as shown in FIG. 12B, a reduced image without aliasing distortion can be obtained.

The above is the condition to be satisfied by the means and the filter in the case of U-fold enlargement and 1 / D-fold reduction.

Here, each signal x (n), c shown in FIG.
The mutual relationship between (n), y (n) and p (n) is expressed by a mathematical expression. The relational expression of each signal x (n), p (n) and y (n) in the case of 1 / D-times reduction shown in FIG. 6B is expressed as follows: p (n) =＝ x (k) h _D ( nk) (1) y (n) = p (Dn) = Σx (k) h _D (Dn−k) (2) The relational expression of x (n), c (n), and y (n) in the case of U-fold magnification shown in FIG. 6A is as follows: y (n) = Σc (k) h _U (nk) (3) y (n) = Σc (Uk) h _U (n-Uk) = Σx (k) h _U (n-Uk) (4) H _D (n) and h _U (n) in these equations indicate the impulse responses of the filters 4 and 7, respectively.

Next, a description will be given of how the U / D-fold resolution conversion of the embodiment of the present invention is realized by using the above-described basic technology.

To perform this U / D-fold resolution conversion, as shown in FIG. 13, a configuration for enlarging a U-fold image in FIG. 6A and a 1/1 in FIG. 6B. What is necessary is just to connect with the structure for image reduction of D times. It can be easily understood that this is the configuration shown in FIG. Further, the lower part of the figure is a configuration in which the redundancy of this configuration is eliminated. Here, obviously, H _UD (n) = H _U (n) H _D (n) holds.

When d (n) and y (n) in FIG. 13B are expressed by mathematical expressions, the following expression (5) is obtained. ). That is, d (n) = Σx (k) h _UD (n-Uk) (5) y (n) = Σx (k) h _UD (Dn-Uk) (6) This is an expression excluding the redundancy.

Here, the elimination of the redundancy in the resolution conversion processing will be described. Eliminating this redundancy is important for efficient execution of the conversion process.

In the configuration shown in FIG. 13A, the components for reducing the image by a factor of 1 / D correspond to those shown in FIG. 6B. It is represented by the convolution of x (n) and p (n) as shown in 1). In this case, for example, in the example of D = 2, as shown in FIG. 14A, an operation of cutting out data and performing convolution processing by shifting one pixel at a time is performed, thereby obtaining a filter output p (0). , p (1), p (2), p (3),..., and then downsamples one out of every two data by the downsampler 5 to obtain every other data p (0) , p
(2),... May be extracted as outputs y (0), y (1),. On the other hand, it is conceivable to directly calculate the above equation (2) to obtain the output y (n). This is shown in FIG.
As shown in FIG. 4 (B), redundancy is eliminated by shifting the convolution of the input data x (n) by two pixels. That is, when D = 2, FIG.
The processing amount of (B) of FIG. 4 is almost half of the processing amount of (A) of FIG. 14, and the processing can be simplified or the processing time can be shortened.

Further, in the configuration shown in FIG.
The component for double image magnification corresponds to FIG. 6A, and the relationship between c (n) and y (n) is a convolution as shown in the above equation (3). expressed. However,
The non-zero value of c (n) is every U number, and it is useless to calculate directly. Therefore, if we pay attention to x (n) = c (Un),
The relationship between x (n) and Y (n) is given as in the above equation (4), and redundancy is eliminated by directly calculating this.
FIG. 15 shows, for example, the case where U = 2. For input data x (n) shown in FIG. 15A, 1 (= 2- 1)
The data c shown in FIG.
(n), and when this is linearly interpolated by a filter, redundancy is eliminated by shifting two pixels at a time as shown in FIG.

Further, as shown in FIG. 13B, the filters of the transfer functions H _U (n) and H _D (n) are put together and the transfer function H _UD (n) (= H _U (n) H _D (n)), and this filter operation is directly performed by the above equation (5).
Further, considering downsampling, the above equation (6)
By directly executing the calculation of, the elimination of redundancy can be largely achieved.

It is clear that the configuration shown in FIG. 13B is the same as FIG. 1 shown as the basic configuration of the first embodiment of the present invention. Since the image is enlarged or reduced depending on the value of U and D, specific examples in each case will be described in the following embodiments.

Next, a second embodiment of the present invention will be described. The second embodiment relates to a resolution converter and method for multiplying an image by an arbitrary rational number. In the second embodiment, the configuration and operation in the case of U> D, that is, the case of enlargement by an arbitrary rational number will be described.

[0046] Specific examples of expansion, consider the case of 3/2 times is U = 3, D = 2, the filter characteristics of the H _U (z) and H _D (z) is, respectively, of FIG. 16 ( A) and (B) are shown. At this time, these two filtering processes are:
Obviously, this is equivalent to the processing by one filter shown in FIG. 16C, and this filter is the same as that shown in FIG. From the above, the filter H _UD for use in enlargement processing U / D times (U> D) (z) can be used filter H _U during expansion of U times the (z) as it is.

As described above, the zero-order hold method by up-sampling (see FIG. 2) can be used for U-fold enlargement, but the linear interpolation method shown in FIG. 3 can be used to keep the image quality smooth. . The relationship at this time can be expressed as follows.

Zero-order hold method: H _U (z) = (1 + z− ¹ + z− ² ++ z− ^(U−1) ) = _Hzero (z) (7) Linear interpolation method: H _U (z) = H _zero (z) H _zero (z ^-1 ) / U = (1 + z ^-1 + z ^-2 + + z- ^(U-1) ) × (1 + z ¹ + z ² + + z ^U-1 ) / U (8) That is, the transfer function of the linear interpolation method is expressed by multiplying the transfer function of the zero-order hold method by delaying the transfer function by time (H _zero (z) and H _zero (z ^-1 )). Also, 1 of U / D times
/ D may be simple downsampling.

The above is the processing in the case of U> D enlargement.

Next, a third embodiment of the present invention will be described. As the third embodiment, the configuration and operation in the case of resolution conversion of an arbitrary rational number and U <D, that is, reduction of an arbitrary rational number will be described.

As a specific example of the reduction, for example, U / D = 2 /
Consider a three-fold reduction process. From the above, the filter characteristics of H _U (z) and H _D (z) are shown as (A) and (B) in FIG. 17, respectively. At this time, the processing by the two filters is performed by one filter H _{UD in} FIG.
It is equivalent to the processing by (z). Also, H _UD (z) = 2H
_Because of the relationship of _D (z), reduction of U / D (U <D
The filter H _UD (z) used in the case of D) is obtained by applying a gain adjustment (U times) to the filter H _D (z) of the 1 / D reduction processing.
H _D a (z) can be used.

Therefore, in the case of U / D-fold reduction, the transfer function H _U (z) of the zero-order hold method of the above equation (7) is used as U-fold expansion, and 1 / D-fold reduction. It can be realized by a transfer function represented by the product using the averaging means (see FIG. 5) of the following equation (9). _{That, H D (z) = (} 1 + z -1 + z -2 + + z - (U-1)) / D ··· (9) H UD (z) = H U (z) H D ( z).

Next, a fourth embodiment of the present invention will be described. In the fourth embodiment, a high-frequency gain emphasizing filter is provided in the preceding stage of the above-described resolution conversion device, and the portion is operated when an image is enlarged.
That is, in the fourth embodiment, blurring of an image and loss of sharpness which occur in the case of enlargement by an arbitrary rational number of U> D are suppressed.

FIG. 18 shows an example of the configuration of the fourth embodiment, which comprises a U-fold upsampler 1, a digital filter 2 and a 1 / D-fold downsampler 3 described with reference to FIG. A high-frequency gain emphasis filter 9 is connected to the front stage of the image resolution conversion device. Next, the operation will be described.

As the high-frequency gain emphasis filter 9, for example, a two-dimensional (3 × 3) two-dimensional filter as shown in FIG. 19 can be used. The high-frequency emphasis filter coefficients f ₁ ~f ₉ example of the filter shown in FIG. 19, for _{example, f 5 = 8, f 1} = f 2 = f 3 = f 4 = f 6 = f 7 = f 8 =
It may be determined as f ₉ = -1. This high-frequency gain emphasis filter 9 has the effect of enhancing the high-frequency components of the input image. In addition,
Generally, as the high-frequency gain emphasis filter 9, a filter having an NxN filter coefficient, a coefficient value in the center portion being a positive number, and other coefficients having a value of -1 can be used. Not done.

A filtered image 111 obtained by multiplying the pixels by these coefficients is input to a U-fold upsampler 1 having the same configuration as in the first embodiment. The processing after the U-fold upsampler 1 may be the same as that described above, and the output 11 of the upsampler 1 may be used.
2, the output 113 of the digital filter 2 and the output 114 of the 1 / D downsampler 3 are the outputs 10 of FIG.
Descriptions corresponding to 1, 102 and 103 are omitted.

As in the fourth embodiment, by providing the high-frequency gain emphasis filter 9 at the front stage, blurring and lack of sharpness of an image generated at the time of enlargement are suppressed, and a high-quality enlarged image is obtained. Has the advantage of being able to

As another embodiment, a value obtained by multiplying the NxN filter coefficient value by a certain parameter value in the high-frequency gain emphasis filter 9 is set as a new filter coefficient value, and the parameter value is made variable. Thus, a high-frequency emphasizing image quality adjusting means may be provided. That is,
For example, image quality adjustment such as the degree of edge enhancement can be performed by multiplying each of the filter coefficients f _{1 to} f ₉ in FIG. 19 by a parameter value p and setting it as a new filter coefficient value.

The present invention can be applied to a case where a function of zooming a thumbnail image or a digital image frequently used in an electronic still camera, a printer, or the like is realized. Specific examples of the application include an electronic camera, portable / mobile image transmission / reception, and the like. Terminal (PD
A), a printer, a satellite image, a medical image, or a software module thereof.

[0060]

According to the present invention, the input image is up-sampled by a factor of U (U is an integer), the up-sampled image is filtered, and the filtered image is 1 / D
By downsampling the input image by a factor of two (D is an integer), the input image can be enlarged by an arbitrary rational number, that is, U / D times.
Resolution conversion with reduction can be performed.

This makes it possible to easily realize the resolution conversion, which has heretofore been easily achieved only by simple integer multiplication or reduction by a factor of one, by an arbitrary rational number. , D, it is possible to always obtain a high-quality resolution-converted image without distortion.

For filtering, digital filter means for changing a transfer function in accordance with a resolution conversion magnification so as not to cause distortion is used. The transfer function of the digital filter means includes an upsample magnification U and a downsample The transfer function H _U (z) of the filter for preventing imaging at the time of up-sampling and the transfer function H _D (z) of the filter for preventing aliasing at the time of down-sampling are determined. with combined and the resulting composite transfer function _{H UD (z) (= H} U (z) H D (z)), redundancy is eliminated, there is an effect requires less processing time.

Further, when the value of the magnification U of the up sample is larger than the value of the magnification D of the down sample, the filtering is performed using a zero-order hold method and a down sampling method, and the filtering is performed. When the value of the magnification D of the sample is larger than the value of the magnification U of the up sample, the filtering is performed by using the zero-order hold method and the averaging method, thereby achieving a simple configuration and a short processing time. Then, the resolution conversion of arbitrary rational number times,
It can be realized while maintaining high image quality.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a basic configuration of a resolution conversion device for performing an arbitrary rational multiple of an image as a first embodiment of the present invention;

FIG. 2 is a diagram for explaining the concept of a zero-order hold method.

FIG. 3 is a diagram for explaining the concept of a linear interpolation method.

FIG. 4 is a diagram for explaining the concept of the down-sampling method.

FIG. 5 is a diagram for explaining the concept of the average value interpolation method.

FIG. 6 is a block diagram showing respective basic configurations for U-time image enlargement and 1 / D image reduction.

FIG. 7 is a diagram for explaining the concept of triple upsampling;

FIG. 8 is a diagram for explaining a frequency band after up-sampling.

FIG. 9 is a diagram for explaining a frequency band of a U-fold enlarged image.

FIG. 10 is a diagram for explaining the concept of 1/4 times downsampling;

FIG. 11 is a diagram for explaining a frequency band of an ideal low-pass filter used for 1 / D-fold image reduction.

FIG. 12 is a diagram for explaining a frequency band of a 1 / D-fold reduced image.

FIG. 13 is a block diagram showing a configuration for performing U / D-fold resolution conversion by combining respective configurations for U-fold image enlargement and 1 / D image reduction.

FIG. 14 is a diagram for explaining elimination of redundancy in the case of 1/2 image reduction.

FIG. 15 is a diagram for explaining elimination of redundancy in the case of double image enlargement.

FIG. 16 is a diagram for describing filter characteristics when a 3 / 2-fold image is enlarged.

FIG. 17 is a diagram for describing filter characteristics at the time of image reduction of 2/3 times.

FIG. 18 is a block diagram illustrating an example of a configuration of an image resolution conversion device provided with a high-frequency gain emphasis filter.

FIG. 19 is a diagram for describing filter coefficients of a high-frequency gain emphasis filter.

[Explanation of symbols]

1,6 U-fold upsampler, 2,4,7,8 digital filter, 3,5 1 / D downsampler

 ────────────────────────────────────────────────── ─── Continued on the front page (72) Inventor Seiji Kimura 6-7-35 Kita-Shinagawa, Shinagawa-ku, Tokyo Inside Sony Corporation (72) Inventor Hitoshi Kiya 1-1, Minami-Osawa, Hachioji-shi, Tokyo Faculty of Engineering Department of Electronics and Information Engineering F-term (Reference) 5B057 CB08 CD07 CD09 CH18 DC22 5C076 AA21 AA22 AA32 BA03 BA04 BA06 BB06 BB24

Claims (10)

[Claims] 1. An up-sampling means for up-sampling an input image by a factor of U (U is an integer); a filter means for filtering an up-sampled image; , D is an integer), and performs downsampling means for downsampling the input image, and performs U / D times resolution conversion on the input image. 2. An image resolution converting apparatus according to claim 1, wherein the transfer function of said filter means is determined by a magnification U of an up sample and a magnification D of a down sample. 3. The filter means combines a transfer function H _U (z) of a filter for preventing imaging during up-sampling and a transfer function H _D (z) of a filter for preventing aliasing during down-sampling. Transfer function H _UD (z) obtained by
2. The image resolution conversion apparatus according to claim 1, wherein the image resolution conversion apparatus has the following characteristics. 4. When the value of the magnification U of the up-sample is larger than the value of the magnification D of the down-sample, the filter means is constituted by using zero-order hold means and down-sample means. The image resolution conversion apparatus according to claim 1, wherein: 5. When the value of the magnification D of the down sample is larger than the value of the magnification U of the up sample, the filter means is constituted by using zero-order hold means and averaging means. The image resolution conversion apparatus according to claim 1, wherein: 6. An up-sampling step of up-sampling an input image by a factor of U (U is an integer), a filtering step of filtering an up-sampled image, and a step of dividing the filtered image by 1 / D (= 1 / D). , D is an integer), and performing a U / D-fold resolution conversion on the input image. 7. The method according to claim 6, wherein in the filtering step, a transfer function thereof is determined by a magnification U of an up sample and a magnification D of a down sample. 8. The filtering step includes the step of transferring a transfer function H of a filter for preventing imaging during up-sampling.
7. A characteristic of a combined transfer function H _UD (z) obtained by combining _U (z) and a transfer function H _D (z) of a filter for preventing aliasing at the time of down-sampling. The resolution conversion method of the described image. 9. When the value of the magnification U of the up sample is larger than the value of the magnification D of the down sample, the filtering step is performed using a zero-order hold method and a down sample method. The method according to claim 6, wherein the resolution of the image is converted. 10. When the value of the magnification D of the down sample is larger than the value of the magnification U of the up sample, the filtering step is performed using a zero-order hold method and an averaging method. The method according to claim 6, wherein the resolution of the image is converted.