JP2006345026A - Image processor and image processing method employing function expression - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processor and an image processing method capable of attaining processing of an image to have a format suitable for resolution conversion and/or compression of the image without causing deterioration to the image even when resolution conversion and compression are repetitively applied to the image. <P>SOLUTION: The image processor applies format conversion processing to an image whose pixel has a pixel value such as color information into a function expression image. A function expression section 11 converts the image into the function expression image on the basis of pixel values of the image. A coding section 12 applies entropy coding to a coefficient sequence on the basis of the function expression converted by the function expression section. The function expression section may use the shift linear interpolation method. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an image processing apparatus and an image processing method, and more particularly to an image processing apparatus and an image processing method suitable for simultaneously performing resolution conversion and image compression of an input image.

  As information is digitized and networked, image information as digital data is widely used. Since the resolutions of input / output devices in digital images are diverse, and the resolutions suitable for each device are different, it is necessary to convert the resolution of the image in order to make use of the performance of each input / output device. Here, the resolution conversion means increasing or decreasing the DPI (Dot Per Inch), which is the number of pixels per unit amount of the digital image, thereby enlarging / reducing the image (higher density / lower density of the image). ).

  Common image interpolation methods include a nearest neighbor method, a bilinear method, a bicubic method, and the like (Non-Patent Document 1). A method using a higher order polynomial or B-Spline function has also been proposed (Non-Patent Document 2). However, a high expansion accuracy is expected, but an unnatural result due to an increase in calculation amount or overfitting. There is also a problem such as showing.

  Many methods other than the above have been proposed for resolution conversion. For example, as a method to increase the magnification accuracy in consideration of edge information, a method of reconstructing isoluminance lines repeatedly using the level set method, a method of reconstructing a triangular mesh and adapting to edges, and a super-resolution method For example, a method using Bayesian estimation is used.

  Also, image data compression is generally performed in order to efficiently use a limited capacity of a storage device or to reduce traffic during transmission / reception on a network. Data compression is to reduce the capacity while maintaining the meaning of data according to a certain procedure. There are a lossless compression method and a lossy compression method. A typical lossy compression method is the JPEG format. Since the JPEG format is a lossy compression method, the image deteriorates every time data is stored. Degradation of an image in JPEG format appears in the image as block noise or mosquito noise.

  In addition, as a lossless compression method that can completely restore compressed data to the data before compression, LZW method, which is a kind of LZ method, and LHA and ZIP that combine LZ method and Huffman method are widely used. (Non-patent Document 3).

  In recent years, the RangeCoder method for realizing a compression rate equivalent to that of the arithmetic compression method at a higher speed, the Block-Sorting method for converting data into an efficient array for compression, and Prediction for compressing data as a probabilistic structure. Attention has been focused on more efficient compression methods such as the by Partial Matching (PPM) method.

  A typical image compression format for resolution conversion is JPEG2000, which compresses an image using multi-resolution analysis of wavelet transform (Non-Patent Document 4). In addition, there is Pixel Live which is an image compression format suitable for image enlargement processing for compressing a luminance change vector of an image by Seratem Technology Co., Ltd.

P. Thevenaz, T .; Blu and M.M. Unser, “Image Interpolation and Resampling”, Handbook of Medical Imaging, Processing and Analysis, I.D. N. Bankman, Ed. , Academic Press, San Diego, CA, USA, pp. 393-420, 2000 M.M. Unser, “Splines: A Perfect Fit for Signal and Image Processing”, IEEE Signal Processing Magazine, vol. 16, no. 6, pp. 22-38, Nov. 1999 Haruhiko Okumura, Toshi Yamazaki “LHA and ZIP” Softbank Publishing, Inc., 2003 M.M. W. Marcellin, M.M. J. et al. Gorish, A .; Bilgin, M.M. P. Boriek, “An Overview of JPEG-2000”, Data Compression Conference 2000, pp. 523-544, 2000

  With regard to resolution conversion, the nearest neighbor method and the bilinear method have high processing speed and can be used easily, but on the other hand, they are not practical because of their low accuracy. In addition, the nearest neighbor method has a problem that the contrast of the original image is maintained but jaggies are conspicuous, and the bilinear method has a problem that the smoothing effect is obtained and jaggies do not occur but the image is blurred. Furthermore, the bicubic method has a problem that the processing speed is slow although the accuracy is good.

  Also, JPEG2000 uses wavelet transform multi-resolution analysis, which is advantageous for reduction processing of image thumbnails and the like, but it is difficult to cope with high-magnification enlargement. Furthermore, in the PixelLive format, enlargement at a high magnification of 1200% is possible, but the compression ratio is not so high, so there remains a problem in terms of compression.

  In view of such circumstances, the present invention provides an image processing apparatus and image processing capable of performing processing to an image format suitable for image resolution conversion and / or compression without deterioration even when image resolution conversion and compression are repeatedly performed. Is to provide a method.

  In order to achieve the above-described object of the present invention, the encoding side of the image processing apparatus according to the present invention converts the image into a function representation based on each pixel value of the image, and the function representation unit. And a coding unit that performs entropy coding on the coefficient sequence based on the function expression.

  Here, the function expression unit may use a shift linear interpolation method. In this case, the encoding unit further includes an optimization conversion unit that converts the function expression converted by the function expression unit into a coefficient sequence suitable for encoding, and the coefficient sequence converted by the optimization conversion unit. Is subjected to entropy coding.

At this time, the optimization conversion unit converts the coefficient sequence suitable for encoding by the following processing, that is,
Calculating a real vertex sequence s [n] (where n is a natural number) shifted by the shifted linear interpolation method;
The real vertex sequence s [n] is rounded and converted to an integer vertex sequence d [n].
FIC coefficient sequence D [n] that is a difference value from the previous vertex of integer vertex sequence d [n] is calculated,
Write a predetermined value to the m byte coefficient list (where m is a natural number) and / or the m + 1 byte coefficient list under the following conditions:
(1) When | D [n] | is larger than the maximum pixel value or when d [n] <0 or d [n] is larger than the maximum pixel value, the value of D [n] is set to m + 1. Write to the byte coefficient list and write a predetermined flag value to the m byte coefficient list. (2) When D [n] is equal to the flag value, write 0 to the m + 1 byte coefficient list and write the flag value to the m byte coefficient list. 3) When D [n] is equal to the value obtained by subtracting the maximum number of pixel values from the flag value, 0 is written to the m + 1 byte coefficient list and the flag value is written to the m byte coefficient list. (4) Above (1) to ( When D [n] <0 except for the condition of 3), the value obtained by adding the maximum number of pixel values to D [n] is written to the m-byte coefficient list. (5) Other than the conditions of (1) to (4) above The value of D [n] is m Write to the coefficient list.

  The encoding unit may use any one of the PPM encoding method, the Huffman encoding method, the arithmetic encoding method, and the RangeCoder encoding method.

  On the other hand, the decoding side of the image processing apparatus of the present invention is based on the entropy-encoded coefficient sequence, a decoding unit that performs entropy decoding, and the coefficient sequence decoded by the entropy decoding unit. A function expression reduction unit that reduces the function expression.

  Here, the function expression reduction unit may reduce the function expression to the function expression by the shift linear interpolation method. In this case, the decoding unit only needs to have a function expression coefficient sequence conversion unit that performs entropy decoding and further converts to a function expression based on a coefficient sequence before encoding.

At this time, the function expression coefficient sequence conversion unit converts into a function expression under the following conditions:
(1) When the value of the m byte coefficient list (where m is a natural number) is a predetermined flag value and the value of the m + 1 byte coefficient list is other than 0, the value of the m + 1 byte coefficient list is D [n]. 2) When the value of the m-byte coefficient list is a flag value and the m + 1 byte coefficient list is 0, if d [n] is larger than the maximum value of the pixel values, a value obtained by subtracting the maximum number of pixel values from the flag value D [n], and if d [n] is less than or equal to the maximum pixel value, the flag value is D [n]. (3) When the conditions other than the above conditions (1) to (2) are satisfied, d [n] If D is larger than the maximum pixel value, the value obtained by subtracting the maximum number of pixel values from the m-byte coefficient list is D [n]. If d [n] is less than or equal to the maximum pixel value, the m-byte coefficient list The value is D [n] where d [n] (however, Is an integer vertex row of natural numbers) is a real series of vertexes s to be shifted by the shift linear interpolation method [n], D [n] is the FIC coefficient sequence is a difference value between integral series of vertexes d [n].

  The decoding unit may use any one of the PPM decoding method, the Huffman decoding method, the arithmetic decoding method, and the RangeCoder decoding method.

  In the image processing apparatus and the image processing method of the present invention, a high-detail enlarged image can be generated by high-resolution resolution conversion and the format can be highly compressed. It is possible to transmit and output at an arbitrary resolution on the receiving side, and further, there is an advantage that reversible compression / reversible resolution conversion is possible without causing image deterioration even if these are repeated.

  The best mode for carrying out the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram for explaining an image processing apparatus according to the present invention. FIG. 1A is a block diagram of an encoding side for encoding an input image, and FIG. 1B is input. FIG. 3 is a block diagram on the decoding side that performs decoding of a decoded image. The encoding side of the image processing apparatus according to the present invention mainly includes an input image unit 10, a function expression unit 11, and an encoding unit 12. The image data input to the input image unit 10 is an image having pixel values such as color information such as RGB and luminance information for each pixel. This may be a general JPEG image or the like. The pixel values are not limited to those described above, and various pixel values such as CMYK, XYZ, HSV, LAB, YIQ, and YUV can be used depending on the color space.

  The function expression unit 11 converts the input image from the input image unit 10 into an image that has been function-expressed. This is based on the pixel value of the input image, and is divided into RGB colors, and the pixel value (0 to 255) for each pixel is expressed by a certain function on the pixel coordinates. For example, a CSRBF method, a linear interpolation method, a shift linear interpolation method, or the like can be used to represent a function passing through a point in each pixel. Once function expression is performed, it can be converted to an arbitrary resolution by setting a predetermined sampling interval (sampling interval) according to the number of pixels required for the output image using this function expression. Become. In the present invention, the resolution conversion is always performed by expressing the image as a function. However, depending on the sampling interval, the DPI may not change before and after the conversion. In this specification, it is expressed as “resolution conversion” including this.

  In the image processing apparatus of the present invention, the encoding unit 12 encodes the image data that has been functionally expressed in this way. In this function, entropy coding is performed using a derived coefficient sequence in a function obtained by the function expression method. Since the present invention focuses on lossless compression, lossless encoding is used in the encoding here, but the present invention is not limited to this, and of course, lossy encoding may be used. I do not care. The encoding unit 12 can compress the image data converted into the function expression.

  As the encoding, various encoding methods such as a PPM encoding method, a Huffman encoding method, an arithmetic encoding method, and a RangeCoder encoding method can be applied, and these are considered in consideration of compression capability and processing speed. The method may be appropriately selected and used.

  As described above, the image processing apparatus of the present invention can generate an image in a format that can be compressed while performing resolution conversion of the image. There is no deterioration even when resolution conversion and compression are repeated, and it is possible to generate an image converted to a resolution suitable for various devices by preparing a single image data. It is also convenient to distribute on top. Further, in order to enhance the security, an encryption process or the like may be performed as necessary.

  Next, the decoding side for decoding the image data in the format generated as described above will be described. As shown in FIG. 1B, the decoding side of the image processing apparatus according to the present invention mainly includes a decoding unit 20, a function expression reduction unit 21, and an output image unit 22. Data encoded on the encoding side of the image processing apparatus according to the present invention is received by the decoding unit 20. The decoding unit 20 performs entropy decoding based on the encoding method used in the encoding. Then, the function expression reduction unit 21 performs a process of reducing the decoded coefficient sequence to the original function expression. Thereafter, the output image unit 22 performs sampling at a predetermined sampling interval so as to obtain a predetermined resolution suitable for the target device by using this function expression, and performs general JPEG or the like using an RGB image or the like. Generate an image. At this time, of course, various formats other than JPEG, color space conversion, and the like may be performed.

  In this way, the image information that has been function-expressed and further encoded is decoded and further reduced to a function representation, thereby enabling reversible conversion of the image data. According to the image processing apparatus of the present invention, since an arbitrary resolution can be obtained, it is possible to realize an image format suitable for resolution conversion and also suitable for compression by encoding.

  As decoding, various decoding methods such as a PPM decoding method, a Huffman decoding method, an arithmetic decoding method, and a RangeCoder decoding method can be applied depending on the method used in the encoding. These may be appropriately selected and used in consideration of the compression capability and processing speed. If encryption is performed on the encoding side, processing corresponding to the encryption may be performed.

  Hereinafter, in the image processing apparatus of the present invention, an embodiment using the shift linear interpolation method as a function expression technique in the function expression unit 11 will be described in more detail. Here, the shifted linear interpolation method realizes the accuracy equivalent to the cubic function interpolation with the same amount of calculation as the linear interpolation by shifting the basis function of the linear interpolation method by a certain amount. FIG. 2 is a block diagram for explaining the details of the encoding unit and the decoding unit when the shift linear interpolation method is used as the function expression unit and the function expression reduction unit of the image processing apparatus of the present invention. FIG. 2A is a block diagram of the encoding unit, and FIG. 2B is a block diagram of the decoding unit. Each process can be realized by a personal computer or the like.

  When the shift linear interpolation method is used, the encoding unit 12 includes an optimization conversion unit 120 and an entropy encoding unit 121 as illustrated in FIG. The optimization conversion unit 120 converts the function expression converted by the function expression unit 11 using the shift linear interpolation method into a coefficient sequence suitable for encoding. Then, the entropy encoding unit 121 performs entropy encoding on the coefficient sequence thus converted.

A procedure for expressing the function by the shift linear interpolation method and optimizing the coefficient sequence based on the function expression by the optimization conversion unit will be specifically described below. A sample value sequence obtained by discretizing a function f (x) based on the pixel value of the input image at the sampling interval T is assumed to be f [n] = f (nT). At this time, the continuous function f _L (x) obtained by the linear interpolation method is expressed by the following equation.
However, n is a natural number.

FIG. 3 is a diagram for explaining the functions of the linear interpolation method and the shift linear interpolation method for the sample values. As shown by the broken line in FIG.
It is expressed as follows.

Next, the function f _S (x) obtained by the shift linear interpolation method from Equation 1 obtained by the linear interpolation method is expressed by the following equation.
Derived.
This is indicated by the solid line in FIG. In FIG. 3, the shift parameter τ = 0.4 is shown, but the value of τ is generally recommended to be about 0.21, and in a more specific embodiment, τ = 0.21 was used. In this manner, image data such as an RGB image is converted into a function expression using a shift linear interpolation method.

A method for converting the coefficient sequence expressed in the function by the shift linear interpolation method as described above into a coefficient sequence suitable for encoding in the optimization conversion unit 120 will be described in detail below. First, sample
It is calculated by the following equation using c [n].
This coefficient sequence s [n] is a point indicated by x in FIG.

The
Thus, it is an integer vertex sequence. At this time, the shift linear function does not pass through the original pixel completely. However, since the original pixel can be completely restored by performing the discretization process again in the process of function reconstruction later, there is no problem in this point. .

Further, if the difference value from the previous vertex of the integer vertex sequence d [n] is defined as the FIC coefficient sequence, FI
Abbreviation for Image Compression.

  Next, a method will be described in which the FIC coefficient sequence D [n] derived by the above equation 7 is converted into a coefficient sequence suitable for compression in the next step and, for example, the PPM encoding method is used as the encoding method. . First, the FIC coefficient sequence D [n] will be examined. FIG. 4 shows the FIC coefficient distribution of 8-bit image data of a human figure having 256 pixels × 256 pixels. In the figure, the horizontal axis represents the value of the FIC coefficient, and the vertical axis represents the number of the FIC coefficient. From FIG. 4, it can be seen that there is a maximum peak at 0 and the coefficients are concentrated near 0. It can be seen that the coefficient values are almost distributed from -100 to 100. In addition, the feature is that the integer vertex sequence d [n] usually falls within the range of 0 to 255. The characteristics of these coefficient distributions are characteristics that generally hold for all image data. From these characteristics, the optimization conversion process is performed by the following method.

First, a 1-byte coefficient list and a 2-byte coefficient list are prepared. Then, predetermined values are written in these coefficient lists under the following conditions.
(1) When | D [n] |> 255 or d [n] <0 or d [n]> 255, the value of D [n] is written to the 2-byte coefficient list and 128 is written to the 1-byte coefficient list. (2) When D [n] = 128, write 0 to the 2-byte coefficient list and write 128 to the 1-byte coefficient list. (3) When D [n] = − 128, write 0 to the 2-byte coefficient list. Write 128 to the 1-byte coefficient list. (4) When D [n] <0 except for the above conditions (1) to (3), write a value obtained by adding 256 to D [n] to the 1-byte coefficient list. (5) Write the value of D [n] to the 1-byte coefficient list when conditions other than the above conditions (1) to (4) are not met

  Here, in the case of 8-bit image data, the range of pixel values is 0 to 255, the value of 255 in (1) above corresponds to the maximum value of the pixel values, and the value of 256 in (4) is the pixel value. It corresponds to the maximum number of. The value (2) 128 is a flag value that serves as a flag indicating that data is written in the 2-byte coefficient list, and is, for example, ½ of the maximum number of pixel values. In the specific embodiment of the present application, 128 which is ½ of the maximum number is set to 128. However, the present invention is not limited to this, and any numerical value may be used as long as the number of appearances is small after the processing. . The value of −128 in (3) corresponds to a value obtained by subtracting the maximum number of pixel values from the flag value. Although the case of 8-bit image data is shown here, the present invention is naturally applicable to the case of 16-bit image data. In this case, the pixel value range is 0 to 65535. In this case, for example, the flag value can be 32768. Furthermore, in the case of 8-bit image data, a value of 1 byte or more is generated by the shift τ, so 2 bytes are required to express all the FIC coefficient sequences. Similarly to this, 16-bit image data In some cases, 3 bytes are required.

  Now, by writing into each coefficient list under such conditions, the FIC coefficients that were all expressed in 2 bytes were converted into 1-byte coefficients and a small number of 2-byte coefficients. When the above-described optimization conversion process is performed on the image data of a person image having 256 pixels × 256 pixels, the 2-byte coefficient before processing is 65,792, whereas the 1-byte coefficient after processing is 65,79. 792 and the 2-byte coefficient was 40. The distribution of 1-byte coefficients at this time is shown in FIG. In the figure, the horizontal axis represents the value of the FIC coefficient, and the vertical axis represents the number of the FIC coefficient. From FIG. 5, it can be seen that there are peaks at 0 and 255, and the coefficients are concentrated in the vicinity of zero. Also, a low peak at 128 used as a flag value for reference in the 2-byte coefficient list can be confirmed.

  As described above, the optimization conversion process according to the present invention can efficiently convert the FIC coefficient into the 1-byte coefficient list and the 2-byte coefficient list, and can express the image as a function with a small amount of information. . In the encoding unit of the present invention, the 1-byte coefficient list and the 2-byte coefficient list are combined and encoded by the entropy encoding unit. Since the 1-byte coefficient list is biased in coefficient information, it is considered that the compression effect by entropy coding is high.

  Next, a description will be given of a technique for decoding a coefficient sequence expressed by a function using the shift linear interpolation method and encoded through the optimization conversion process as described above. When the shift linear interpolation method is used, the decoding unit 20 includes an entropy decoding unit 200 and a function expression coefficient sequence conversion unit 201 as shown in FIG. The entropy decoding unit 200 decodes what is encoded by the entropy encoding unit 121. Then, the function expression coefficient string conversion unit 201 converts (reduces) the decoded coefficient string into a function expression based on the coefficient string before encoding.

More specifically, the function expression is converted under the following conditions.
(1) When the value of the 1-byte coefficient list is 128 and the value of the 2-byte coefficient list is other than 0, the value of the 2-byte coefficient list is set to D [n]. (2) The value of the 1-byte coefficient list is 128. And when the 2-byte coefficient list is 0, if d [n]> 255, −128 is set to D [n], and if d [n] ≦ 255, 128 is set to D [n] (3) When conditions other than the above (1) and (2) are satisfied, if d [n]> 255, the value obtained by subtracting 256 from the 1-byte coefficient list is D [n], and if d [n] ≦ 255, 1 The value of the byte coefficient list is D [n] where d [n] (where n is a natural number) is an integer vertex sequence of a real vertex sequence s [n] shifted by the shift linear interpolation method, D [n] Is an FIC coefficient sequence which is a difference value of the integer vertex sequence d [n].

  In this case as well, the case of 8-bit image data has been described with specific numerical values, but the present invention is not limited to this, and even in the case of 16-bit image data as described above, it depends on the maximum pixel value and the like. It can be applied by changing the conditions.

  With the above processing, the coefficient sequence before encoding can be restored, and by using the equations 1 to 7 in reverse, it is possible to completely reduce the function expression converted on the encoding side. Therefore, compression is performed on the encoding side, the capacity is reduced, the data is transmitted to a network, etc., decoding is performed on the receiving side, the original function expression is reduced, and an output resolution is determined using a predetermined sampling interval. An image can be output.

  Here, an image output after resolution conversion and compression processing by the image processing apparatus of the present invention will be described. 6A and 6B are diagrams for viewing changes due to image compression / enlargement (actual image data is color), FIG. 6A is a sample original image, and FIG. 6B is an image processing apparatus of the present invention. An image compressed and enlarged using FIG. 6C is an image enlarged by the bicubic method after performing the conventional JPEG compression. First, an original image of 150 pixels × 150 pixels was compressed and interpolated using the image processing apparatus of the present invention to generate an image of 450 pixels × 450 pixels (FIG. 6B). Similarly, after the same original image was compressed by JPEG, image interpolation was performed by the bicubic method to generate an image of 450 pixels × 450 pixels (FIG. 6C). As can be seen from the figure, an image compressed by JPEG generated block noise during compression, and further appeared so that this block noise was emphasized during image interpolation by the bicubic method. On the other hand, the image subjected to the compression and image interpolation according to the present invention can capture the features of the image such as edges well and can naturally enlarge the fine portion. This is because the compression according to the present invention is reversible compression so that there is no image degradation, and there is no image degradation even if resolution conversion is performed by function expression.

  Next, the evaluation of three points of image enlargement accuracy, calculation time required for resolution conversion, and image compression rate will be described by comparing the present invention with the prior art. In addition, the environment where various evaluations were performed used Microsoft (registered trademark) Windows (registered trademark) XP as an OS on a computer having a configuration of Intel (registered trademark) Pentium (registered trademark) 4 3.2 GHz, 2 GB RAM. Is.

  First, the image enlargement accuracy will be described with reference to FIG. FIG. 7 is a result (actual image data is color) in each method when the portion (32 pixels × 32 pixels) indicated by the black frame in FIG. 7A is enlarged to 8 times (256 pixels × 256 pixels). 7 (b) is based on the nearest neighbor method, FIG. 7 (c) is based on the bilinear method, FIG. 7 (d) is based on the bicubic method, and FIG. 7 (e) is based on the image processing apparatus of the present invention. Is. As can be seen from FIG. 7B, the nearest neighbor method has noticeable jaggy and lacks smoothness of the image. In the bilinear method of FIG. 7C, no jaggy is generated, but the image is totally blurred. Both the bicubic method shown in FIG. 7D and the present invention shown in FIG. 7E generate a clear enlarged image. The image according to the present invention is slightly better in image contrast. As described above, it can be seen that the image processing apparatus according to the present invention has an enlargement accuracy equivalent to that of the bicubic method and a contrast superior to that of the bicubic method.

Next, a result of measuring the calculation time required for resolution conversion when two original images of 256 pixels × 256 pixels and 512 pixels × 512 pixels are enlarged to two types of 1024 pixels × 1024 pixels and 2048 pixels × 2048 pixels using various methods, respectively. Is shown in Table 1.
As can be seen from Table 1, for all the results, it can be seen that the calculation takes time in the order of the nearest neighbor method, the bilinear method, the method of the present application, and the bicubic method. Combined with the result of the above-described enlargement accuracy, it can be seen that the present invention can provide the same enlargement accuracy as the bicubic method at a calculation speed that is intermediate between the bilinear method and the bicubic method.

Further, the compression rate is evaluated. FIG. 8 shows sample images (actual image data is color) of a person or landscape used for evaluating the compression rate. Table 2 shows the results of compression rates when these sample images were reversibly compressed using TIFF using LZW compression, TIFF using ZIP compression, PNG compression, lossless JPEG, lossless JPEG2000, and the present invention. Show. The compression rate is the file size (percentage) of the compressed image with respect to the file size of the original image.
From Table 2, it can be seen that there are images suitable for compression and unsuitable for the conventional compression method. TIFF (LZW), TIFF (ZIP), and PNG are used to compress images with a small number of colors and regularity in color distribution, such as original image (b), original image (d), and original image (e). Although it is suitable, it can be seen that it is not suitable for compression of an image having a large number of colors such as the original image (a) and the original image (c) and having no regular distribution. JPEG-LS and JPEG2000 are suitable for compression of images such as an original image (a), an original image (c), an original image (d), and an original image (f) in which colors change smoothly. In an image having a large color change as in e), the compression rate is lowered. On the other hand, in the present invention, there is little orientation failure with respect to all images, and a stable compression rate is always shown. Furthermore, the original image (b) achieves a higher compression rate than the conventional compression method.

  As described above, according to the present invention, it is possible to convert the image format into an image format which has an enlargement accuracy comparable to that of the conventional image, has a fast resolution conversion process, and has an excellent compression ratio on average.

  The image processing apparatus and the image processing method of the present invention are not limited to the above illustrated examples and calculation formulas, and various changes can be made without departing from the scope of the present invention. is there.

FIG. 1 is a block diagram for explaining an image processing apparatus according to the present invention. FIG. 1A is a block diagram of an encoding side for encoding an input image, and FIG. 1B is input. FIG. 3 is a block diagram on the decoding side that performs decoding of a decoded image. FIG. 2 is a block diagram for explaining the details of the encoding unit and the decoding unit when the shift linear interpolation method is used as the function expression unit and the function expression reduction unit of the image processing apparatus of the present invention. FIG. 3 is a diagram for explaining a function based on a linear interpolation method and a shift linear interpolation method on sample values. FIG. 4 is an FIC coefficient distribution graph of image data of a certain person image. FIG. 5 is a distribution graph of 1-byte coefficients of data processed by the image processing apparatus of the present invention. FIG. 6 is a comparative diagram for viewing changes due to image compression / enlargement. FIG. 7 is a comparative diagram for viewing the enlargement accuracy of an image. FIG. 8 is a diagram showing a sample image for evaluating the compression rate of the image.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Input image part 11 Function expression part 12 Encoding part 20 Decoding part 21 Function expression reduction part 22 Output image part 120 Optimization conversion part 121 Entropy encoding part 200 Entropy decoding part 201 Function expression coefficient sequence conversion part

Claims (24)

An image processing apparatus that performs a format conversion process on an image having a pixel value such as color information in each pixel into a function-represented image,
A function representation unit that converts the image into a functional representation based on each pixel value of the image;
An encoding unit that performs entropy encoding on the coefficient sequence based on the function expression converted by the function expression unit;
An image processing apparatus comprising:   The image processing apparatus according to claim 1, wherein the function expression unit uses a shift linear interpolation method.   The image processing apparatus according to claim 2, wherein the encoding unit further includes an optimization conversion unit that converts the function expression converted by the function expression unit into a coefficient sequence suitable for encoding, An image processing apparatus that performs entropy coding on a coefficient sequence converted by an optimization conversion unit. The image processing apparatus according to claim 3, wherein the optimization conversion unit converts into a coefficient sequence suitable for encoding by the following processing:
Calculating a real vertex sequence s [n] (where n is a natural number) shifted by the shifted linear interpolation method;
The real vertex sequence s [n] is rounded to be converted into an integer vertex sequence d [n].
FIC coefficient sequence D [n], which is a difference value from the previous vertex of the integer vertex sequence d [n], is calculated,
Write a predetermined value to the m byte coefficient list (where m is a natural number) and / or the m + 1 byte coefficient list under the following conditions:
(1) When | D [n] | is larger than the maximum pixel value or when d [n] <0 or d [n] is larger than the maximum pixel value, the value of D [n] is set to m + 1. Write to the byte coefficient list and write a predetermined flag value to the m byte coefficient list (2) When D [n] is equal to the flag value, write 0 to the m + 1 byte coefficient list and write the flag value to the m byte coefficient list (3) When D [n] is equal to the value obtained by subtracting the maximum number of pixel values from the flag value, 0 is written in the m + 1 byte coefficient list and the flag value is written in the m byte coefficient list. (4) When D [n] <0 other than the condition (3), a value obtained by adding the maximum number of pixel values to D [n] is written to the m-byte coefficient list. (5) Conditions (1) to (4) above Otherwise, the value of D [n] An image processing apparatus that writes to an m-byte coefficient list.   5. The image processing apparatus according to claim 1, wherein the encoding unit uses any one of a PPM encoding method, a Huffman encoding method, an arithmetic encoding method, and a RangeCoder encoding method. An image processing apparatus that performs encoding. An image processing device that performs format conversion processing on an image having a pixel value such as color information in each pixel and converted into an image that is expressed as a function into an image having pixel values such as color information in each pixel. The device is
A decoding unit that performs entropy decoding on the entropy-encoded coefficient sequence;
A function expression reduction unit that reduces the function expression based on the coefficient sequence decoded by the entropy decoding unit;
An image processing apparatus comprising:   The image processing apparatus according to claim 6, wherein the function expression reduction unit reduces the function expression to a function expression by a shift linear interpolation method.   The image processing apparatus according to claim 7, wherein the decoding unit includes a function expression coefficient sequence conversion unit that performs entropy decoding and further converts into a function expression based on a coefficient sequence before encoding. An image processing apparatus. The image processing apparatus according to claim 8, wherein the function expression coefficient sequence conversion unit converts into a function expression under the following conditions:
(1) When the value of the m byte coefficient list (where m is a natural number) is a predetermined flag value and the value of the m + 1 byte coefficient list is other than 0, the value of the m + 1 byte coefficient list is D [n]. 2) When the value of the m-byte coefficient list is a flag value and the m + 1 byte coefficient list is 0, if d [n] is larger than the maximum value of the pixel values, a value obtained by subtracting the maximum number of pixel values from the flag value D [n], and if d [n] is less than or equal to the maximum pixel value, the flag value is D [n]. (3) When the conditions other than the above conditions (1) to (2) are satisfied, d [n] If D is larger than the maximum pixel value, the value obtained by subtracting the maximum number of pixel values from the m-byte coefficient list is D [n]. If d [n] is less than or equal to the maximum pixel value, the m-byte coefficient list The value is D [n] where d [n] (however, Is FIC coefficient sequence is a difference value between integral series of vertexes of a real series of vertexes s [n] to be shifted, D [n] is the integer vertex sequence d [n] by a natural number) is shifted linear interpolation method,
An image processing apparatus.   10. The image processing device according to claim 6, wherein the decoding unit uses any one of a PPM decoding method, a Huffman decoding method, an arithmetic decoding method, and a RangeCoder decoding method. An image processing apparatus that performs decoding.   10. The image processing apparatus according to claim 4 or 9, wherein in the case of 8-bit image data, the pixel value ranges from 0 to 255 and m = 1, and in the case of 16-bit image data, the pixel value. A value range is 0 to 65535, and m = 2.   10. The image processing apparatus according to claim 4, wherein the flag value is ½ of the maximum number of pixel values. 11. An image processing method for performing a format conversion process on an image having pixel values such as color information in each pixel into a function-represented image, the method comprising:
Based on each pixel value of the image, the process of converting the image into a functional representation,
Encoding the coefficient sequence based on the function expression converted by the process of converting to the function expression;
An image processing method comprising:   14. The image processing method according to claim 13, wherein the step of converting into the function expression uses a shift linear interpolation method.   15. The image processing method according to claim 14, wherein the encoding step further includes a step of converting the function expression converted by the process of converting into the function expression into a coefficient sequence suitable for encoding. An image processing method, wherein entropy coding is performed on the coefficient sequence converted by the process. 16. The image processing method according to claim 15, wherein the process of converting into a coefficient sequence suitable for encoding is converted into a coefficient sequence suitable for encoding by the following processing:
Calculating a real vertex sequence s [n] (where n is a natural number) shifted by the shifted linear interpolation method;
The real vertex sequence s [n] is rounded to be converted into an integer vertex sequence d [n].
FIC coefficient sequence D [n], which is a difference value from the previous vertex of the integer vertex sequence d [n], is calculated,
Write a predetermined value to the m byte coefficient list (where m is a natural number) and / or the m + 1 byte coefficient list under the following conditions:
(1) When | D [n] | is larger than the maximum pixel value or when d [n] <0 or d [n] is larger than the maximum pixel value, the value of D [n] is set to m + 1. Write to the byte coefficient list and write a predetermined flag value to the m byte coefficient list (2) When D [n] is equal to the flag value, write 0 to the m + 1 byte coefficient list and write the flag value to the m byte coefficient list (3) When D [n] is equal to the value obtained by subtracting the maximum number of pixel values from the flag value, 0 is written in the m + 1 byte coefficient list and the flag value is written in the m byte coefficient list. (4) When D [n] <0 other than the condition (3), a value obtained by adding the maximum number of pixel values to D [n] is written to the m-byte coefficient list. (5) Conditions (1) to (4) above Otherwise, the value of D [n] An image processing method comprising writing to an m-byte coefficient list.   The image processing method according to any one of claims 13 to 16, wherein the entropy encoding is performed by any one of a PPM encoding method, a Huffman encoding method, an arithmetic encoding method, and a RangeCoder encoding method. An image processing method characterized by being used. An image processing method for performing an image format conversion process on an image having a pixel value such as color information in each pixel and converted into an image having a function representation of an image having a pixel value such as color information in each pixel. The method
Performing entropy decoding on the entropy-encoded coefficient sequence;
A process of reducing to a function expression based on the coefficient sequence decoded by the process of performing the entropy decoding;
An image processing method comprising:   19. The image processing method according to claim 18, wherein the process of reducing to the function expression is reduced to a function expression by a shift linear interpolation method.   20. The image processing method according to claim 19, wherein the step of performing entropy decoding includes a step of performing entropy decoding and further converting the function into a function expression based on a coefficient sequence before encoding. Processing method. The image processing method according to claim 20, wherein the step of converting into the function representation is performed under the following conditions:
(1) When the value of the m byte coefficient list (where m is a natural number) is a predetermined flag value and the value of the m + 1 byte coefficient list is other than 0, the value of the m + 1 byte coefficient list is D [n]. 2) When the value of the m-byte coefficient list is a flag value and the m + 1 byte coefficient list is 0, if d [n] is larger than the maximum value of the pixel values, a value obtained by subtracting the maximum number of pixel values from the flag value D [n], and if d [n] is less than or equal to the maximum pixel value, the flag value is D [n]. (3) When the conditions other than the above conditions (1) to (2) are satisfied, d [n] If D is larger than the maximum pixel value, the value obtained by subtracting the maximum number of pixel values from the m-byte coefficient list is D [n]. If d [n] is less than or equal to the maximum pixel value, the m-byte coefficient list The value is D [n] where d [n] (however, Is FIC coefficient sequence is a difference value between integral series of vertexes of a real series of vertexes s [n] to be shifted, D [n] is the integer vertex sequence d [n] by a natural number) is shifted linear interpolation method,
An image processing method.   The image processing method according to any one of claims 18 to 21, wherein the entropy decoding is performed by any one of a PPM decoding method, a Huffman decoding method, an arithmetic decoding method, and a RangeCoder decoding method. An image processing method characterized by being used.   19. The image processing method according to claim 13, wherein in the case of 8-bit image data, the pixel value ranges from 0 to 255 and m = 1, and in the case of 16-bit image data, a pixel. A value range is 0 to 65535, and m = 2. 19. The image processing method according to claim 13, wherein the flag value is ½ of the maximum number of pixel values.