JP2015041200A - Image processor, image forming apparatus, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To shorten the processing period required for correcting an image blur even when uneven blurs are generated on an image.SOLUTION: An image processor 1 comprises: an image acceptance section 11 for accepting input image data input from the outside; an image division section 12 for dividing the input image based on the input image data into a plurality of small areas; a small area selection section 13 for selecting one or more small areas from among the plurality of small areas of the input image, for allowing a parameter estimation section 14 to estimate a parameter value; the parameter estimation section 14 for estimating a parameter value which a function defined according to a blur generated in the image has; a blur correction section 15 for generating blur correction image data having the blur generated in the input image corrected on the basis of the parameter value estimated and the input image data; and an image output section 16 for outputting the blur correction image data to the outside.

Description

  The present invention relates to an image processing apparatus, an image forming apparatus, and a program.

  As a technique for correcting image blur, there is a technique using PSF (Point Spread Function). This technology utilizes the fact that a blurred image can be represented by convolving a PSF with the original sharp image, and restores the original sharp image by performing a PSF deconvolution operation on the blurred image. Technology. As an image restoration technique using PSF, for example, there is a Wiener filter that restores an original image on the assumption that the same blur is generated in the image when the PSF is known (see Non-Patent Document 1). ). Further, Richardson-Lucy devolution exists as a technique for restoring the original image when the PSF is known (see Non-Patent Document 2).

  Further, as a conventional technique described in the publication, an image processing apparatus that performs a shading correction process, a gamma correction process, a gradation process, and a process of selecting filter strengths in two directions orthogonal to an input image, There is an image processing device that adjusts the degree of image sharpness by switching the contents of filtering processing according to the area of the original, regardless of whether the input image is a pattern, text, line drawing, photograph, halftone image, etc. (See Patent Document 1).

  Furthermore, as a conventional technique described in other publications, an input image is divided into a plurality of small region images, a PSF is estimated for each small region image, a shape of the estimated PSF is identified, and a highly similar PSF shape There exists a subject separation device that separates subject images having similar blurring methods by classifying the small region images into the same group and integrating adjacent small region images classified into the same group for each group. . There is also an image restoration device that obtains a restored image by further obtaining a PSF based on a subject image that is separated by the subject separation device and has a similar blur (see Patent Document 2).

  Further, it is known that a natural image without blur has a distribution (sparse distribution) in which a gradient histogram indicating the degree of change in pixel value in an image is concentrated on a specific value (non-sparse distribution). (See Patent Documents 3 and 4). Non-Patent Document 4 shows an example of such a prior distribution of images as Equation (6) and Equation (7).

  Further, in statistics, an EM algorithm is known as a technique for obtaining a parameter that maximizes likelihood, which is a degree representing likelihood, by iterative calculation (see Non-Patent Documents 4 and 5). Non-Patent Document 5 also describes other methods such as a sampling method, a variational method, and a generalized EM algorithm as a method for obtaining parameters.

JP-A-9-247460 JP 2012-155456 A

Keiichi Tezuka, Tadahiro Kitahashi, Hideo Ogawa, "Digital Image Processing Engineering", 1st edition, Nikkan Kogyo Shimbun, June 15, 1985, p.54-57 W. Richardson, “Bayesian-based iterative method of image restoration”, Journal of the Optical Society of America, Vol.62, 1972, p.55-59 R. Fergus, et.al, `` Removing camera shake from a single photograph '', SIGGRAPH 2006 Papers, ACM, 2006, p.787-794 A.Levin, et.al, `` Efficient marginal likelihood optimization in blind deconvolution '', Proceeding of CVPR 2011, 2011, p.2657-2664 C.M.Bishop, "Pattern Recognition and Machine Learning, Statistical Prediction Based on Bayesian Theory", First Edition, 3rd edition, Springer Japan, June 21, 2009, p.139-171

  An object of the present invention is to reduce the processing time required to correct image blur even when nonuniform blur occurs in the image.

According to the first aspect of the present invention, there is provided a receiving unit that receives input image data, a dividing unit that divides the image data received by the receiving unit into a plurality of regions, and a blur that occurs in an image. A function defined to be expressed in accordance with the position of the acquisition unit, the acquisition unit that acquires the function having a plurality of variables, and the acquisition unit acquired from the plurality of regions divided by the division unit The selection unit that selects one or a plurality of regions for calculating the value of the variable included in the function, and the one or more regions selected by the selection unit, and the acquisition unit acquires the A calculation unit that calculates a value of a variable included in the function; a value of the variable calculated by the calculation unit; and the image data received by the reception unit; Based on an image processing apparatus and a correcting means for correcting the blur of the image in the image data.
The invention according to claim 2 is the image processing apparatus according to claim 1, wherein the function is defined by a function representing a Gaussian distribution.
The invention according to claim 3 is characterized in that the selecting means selects the area in which the distribution of pixel values in the area is non-uniform among the plurality of areas divided by the dividing means. An image processing apparatus according to claim 1.
According to a fourth aspect of the present invention, there is provided a receiving unit that receives input image data that is input image data, a dividing unit that divides the input image data received by the receiving unit into a plurality of regions, and an image. It is a function defined so that blur is represented according to the position in the image, from the plurality of regions divided by the dividing unit, an acquisition unit that acquires the function having a plurality of variables, Using the selection means for selecting one or a plurality of areas for calculating the value of the variable possessed by the function acquired by the acquisition means, and the one or more areas selected by the selection means, the acquisition A calculation means for calculating a value of a variable included in the function acquired by the means; a value of the variable calculated by the calculation means; Generating means for generating blur-corrected image data obtained by correcting image blur in the input image data based on the input image data, and a recording material based on the blur-corrected image data generated by the generating means The image forming apparatus includes an image forming unit that forms an image.
According to a fifth aspect of the present invention, there is provided a computer having a function of receiving input image data, a function of dividing the received image data into a plurality of areas, and a blur occurring in the image at a position in the image. A function defined to be expressed in accordance with the function to obtain the function having a plurality of variables, and to calculate the value of the variable that the obtained function has from the plurality of divided areas A function for selecting one or a plurality of areas, a function for calculating a variable value of the acquired function using the selected one or a plurality of areas, and accepting the calculated value of the variable And a function for correcting image blur in the image data based on the image data.

According to the first aspect of the present invention, compared with the case where the present configuration is not provided, even if non-uniform blur occurs in the image, the processing required to correct the blur of the image Time is shortened.
According to the second aspect of the present invention, when the degree of blur increases from the central part toward the peripheral part as compared with the case where the present configuration is not provided, it is easy to correct the blur of the image. Become.
According to the third aspect of the present invention, it is possible to improve the accuracy of correcting image blur as compared with the case where the present configuration is not provided.
According to the invention described in claim 4, the processing required for correcting the blur of the image even when the blur is uneven in the image as compared with the case where the present configuration is not provided. It is possible to shorten the time and form an image.
According to the fifth aspect of the present invention, even when non-uniform blur occurs in the image as compared with the case where the present configuration is not provided, the processing required for correcting the blur of the image A function for shortening the time can be realized by a computer.

It is the block diagram which showed the function structural example of the image processing apparatus which concerns on this Embodiment. It is the figure which showed an example of the small area | region in the input image which concerns on this Embodiment. It is the figure which showed an example of the production | generation model of the blurred input image which concerns on this Embodiment. (A) (b) (c) is a figure showing an example of blur by a Gaussian filter. It is the flowchart which showed an example of the procedure which the image processing apparatus which concerns on this Embodiment estimates an original image. 1 is a diagram illustrating an example of an image forming apparatus that realizes functions of an image processing apparatus according to an exemplary embodiment. It is the figure which showed an example of the hardware constitutions of the image processing apparatus which concerns on this Embodiment.

Embodiments of the present invention will be described below in detail with reference to the accompanying drawings.
<Description of Configuration of Image Processing Device>
FIG. 1 is a block diagram illustrating a functional configuration example of an image processing apparatus 1 according to the present embodiment.
The image processing apparatus 1 according to the present embodiment includes an image receiving unit 11 that receives image data (hereinafter referred to as input image data) input from the outside, and an image based on the input image data (hereinafter referred to as an input image). Is divided into a plurality of regions (hereinafter, this region is referred to as a small region), and a parameter estimation unit 14 (to be described later) among the plurality of small regions of the input image is used for estimating the parameter value. Or the small area selection part 13 which selects a some small area is provided. The image processing apparatus 1 also uses the parameter estimation unit 14 that estimates the parameter value of the function defined for the blur occurring in the image, and the input image data based on the estimated parameter value and the input image data. Are provided with a blur correction unit 15 that generates image data (hereinafter referred to as blur correction image data) obtained by correcting the blur generated in step 1 and an image output unit 16 that outputs the blur correction image data to the outside.

  An image receiving unit 11 as an example of a receiving unit receives input image data from the outside. In the present embodiment, the input image based on the input image data is a target for correcting image blur. Then, the image reception unit 11 transmits the received input image data to the image division unit 12 and the blur correction unit 15.

  An image dividing unit 12 as an example of a dividing unit divides an input image based on input image data transmitted from the image receiving unit 11 into a plurality of small regions. Here, the image dividing unit 12 divides the input image into m × n small regions vertically and horizontally. Then, the image dividing unit 12 divides the input image into a plurality of small areas, and then transmits the input image data to the small area selecting unit 13.

  The small region selection unit 13 as an example of a selection unit selects one or a plurality of small regions suitable for parameter estimation by the parameter estimation unit 14 from the plurality of small regions of the input image divided by the image dividing unit 12. To do. As a small region suitable for parameter estimation, for example, a region in which the distribution of pixel values in the small region is not uniform corresponds. For an image with a uniform distribution of pixel values, even if blurring occurs in the image, it may not be possible to determine whether blurring has occurred because the distribution of image values is uniform. For this reason, the small region selection unit 13 selects an image with a nonuniform pixel value distribution as an image for which it is easy to determine the presence or absence of blur in the image. Then, the small region selection unit 13 transmits the selected one or more small regions to the parameter estimation unit 14.

  FIG. 2 is a diagram showing an example of a small area in the input image according to the present embodiment. The input image shown in FIG. 2 is divided into m × n small areas vertically and horizontally by the image dividing unit 12 (m = 9, n = 8 in the input image shown in FIG. 2). This small region is used for parameter estimation by the parameter estimation unit after being selected by the small region selection unit 13, and the size of the small region depends on, for example, the degree of change in blur in the small region. It can be determined. For example, it is conceivable to reduce the small area if the blur change in the small area is large, and to increase the small area if the blur change in the small area is small.

  In addition, the small region selection unit 13 calculates entropy indicating the degree of distribution of pixel values for each small region, for example, in order to select a small region with a nonuniform pixel value distribution. Select in order. In the example illustrated in FIG. 2, the small region selection unit 13 selects the top three small regions (r1 to r3 indicated by diagonal lines) having the largest entropy. In addition, for example, the small area selection unit 13 may select a small area whose entropy is larger than a predetermined threshold. Further, for example, the small region selection unit 13 may select a small region based on the difference between the maximum pixel value and the minimum pixel value in each small region, the average of the absolute value of the gradient in each small region, or the like. good.

  The parameter estimation unit 14 as an example of an acquisition unit and a calculation unit is a function that is defined so that blur occurring in an image is represented according to a position in the image. The parameter value of the function is estimated using one or a plurality of small regions selected by the region selection unit 13. Details of the function will be described later.

  The blur correction unit 15 as an example of a correction unit generates blur correction image data based on the parameter value estimated by the parameter estimation unit 14 and the input image data transmitted from the image reception unit 11. The image based on the blur-corrected image data is obtained by correcting the blur of the input image, and is estimated as an original image (hereinafter referred to as an original image) before the blur occurs. Then, the blur correction unit 15 transmits the generated blur correction image data to the image output unit 16.

  The image output unit 16 outputs the blur correction image data transmitted from the blur correction unit 15 to the outside.

<Description of blurred image generation model>
Next, a model for generating a blurred input image will be described. FIG. 3 is a diagram showing an example of a blurred input image generation model according to the present embodiment. As shown in FIG. 3, an original image before blur is represented as an original image x, and an input image after blur is represented as y. In general, the blur characteristic can be expressed by PSF. If PSF is expressed as k, generation of blur is formulated as a convolution operation between the filter kernel k, which is a discrete expression of PSF, and the original image x. . As a model for generating a blurred image, it is realistic to consider noise n. For this reason, the input image y is expressed as shown in Equation 1.

ただし、ｋ＊ｘは、ｋとｘとの畳み込み演算を表す。

However, k * x represents a convolution operation between k and x.

  In the present embodiment, PSF (k) and noise n are defined for the generation model shown in FIG. 3, and it is assumed that a blurred input image is generated by this generation model. The original image x is estimated.

  First, as the noise n, for example, noise according to a Gaussian distribution with an average of 0 and a variance of γ is added to each pixel independently.

  Also, for example, when shooting with a device that is not of high quality such as a camera of a portable information terminal, the vicinity of the optical axis is in focus, but the degree of blur often increases toward the periphery. Therefore, as the PSF, the blur occurring on the image is not represented by a fixed PSF, but is represented by a PSF whose characteristics change according to the position on the image. Then, the blur of the image is regarded as a substantially uniform blur locally. For example, the blur generated at the position p on the image is determined by the PSF of the Gaussian filter which is a function representing a Gaussian distribution as shown in Equations 2 and 3. Define as required. Here, the position p is assumed to be the center position in the Gaussian filter.

  However, in Equations 2 and 3, k is a PSF filter kernel, λ is a normalization coefficient for setting the sum of the filter kernels to 1, and z is a filter kernel from the kernel center (ie, position p which is the center). The distance to each element, v, represents the dispersion of blur, that is, the degree of blur. Since the variance v is not a fixed value but changes according to the position p on the image, the degree of blur changes according to the position on the image.

  Further, the variance v is expressed as a function of the position p on the image and the optical axis position c, and the variance increases as the distance from the optical axis position c increases, that is, the blur of the image increases. . || p−c || is the Euclidean distance between p and c, and a is a coefficient for controlling the gradient of the blur. Furthermore, δ is a value provided because Equation 2 cannot be calculated when the dispersion v on the optical axis position becomes 0, and is set appropriately as a value that is larger than 0 but very small. For example, when the position p on the image is the optical axis position c, the variance is only δ, and k is expressed as a filter when there is almost no blur.

  Various blurs are uniformly described by the definition of the blur by the Gaussian filter shown in Formula 2 and Formula 3. 4A to 4C are diagrams illustrating an example of blur caused by a Gaussian filter.

  The image shown in FIG. 4A is a general image in which blur has occurred, and an example of an image in which the optical axis position c is at the center of the image and the blur increases as the distance from the optical axis position c increases. It is. The coefficient a is set to a value corresponding to the degree of blur. The image shown in FIG. 4B is an example of an image read by a flat bed scanner or the like. In this case, since the optical axis position c is at the center of the image and hardly blurs, the coefficient a is set to a value close to zero. The image shown in FIG. 4C is an example of an image obtained by cutting out a part of the blurred image. The optical axis position is at the lower left of the image, and the coefficient a is set to a value corresponding to the degree of blur. The blurring of the image shown in FIGS. 4A to 4C is represented as blurring by the Gaussian filters of Formula 2 and Formula 3.

  Further, for the original image x, an image prior distribution p (x) is defined as a property of the original image x, and the original image x follows the prior distribution p (x). For example, in a natural image without blur, it is assumed that the gradient histogram has a distribution (sparse distribution) that concentrates on a specific value. Using this property, the prior distribution p (x) is determined.

  As described above, by defining PSF (k) and noise n and determining the prior distribution of the original image x, among the parameters of Equation 1, the coefficient a, the optical axis position c, and the variance γ in the noise n are unknown. Parameter. These unknown parameters are probabilistically estimated from the input image y, and are estimated by using a known procedure. Then, using a known procedure based on the estimated unknown parameter, the blur of the input image is corrected and the original image is estimated.

<Explanation of unknown parameter estimation procedure>
Next, a procedure for the parameter estimation unit 14 to estimate the coefficient a, the optical axis position c, and the variance γ of the noise n as unknown parameters will be described. As this estimation procedure, a known one may be used, and an outline thereof will be described here.

  First, the parameter estimation unit 14 sets the center coordinates of the i-th small region ri as pi and the original image of the small region ri as xi for the small region selected by the small region selection unit 13 as shown in FIG. In this case, the distribution of the input image yi in the small region ri is based on the following equation (4) based on the equation 1 representing the input image y and the Gaussian distribution in which the noise n is 0 on average and the variance is γ Expressed with a distribution.

ただし、記号∝は比例関係を表す記号であり、θは数５式のように、未知パラメータａ、ｃ、γを表すものである。

However, the symbol 記号 is a symbol representing a proportional relationship, and θ represents the unknown parameters a, c, and γ as shown in Equation 5.

  Here, it is assumed that the prior distribution of the original image x is constant at p (x) regardless of the position on the image. At this time, from the well-known Bayes' theorem, the simultaneous distribution of xi and yi is expressed as in Equation 6.

  In addition, when using the property that the gradient histogram has a distribution that concentrates on a specific value as p (x) as described above, xi and yi are not represented by the image itself but represented by the gradient. This will reduce the calculation process. However, since the expression itself is the same as Expression 4 and Expression 5, the same expression is used here, including the case where xi and yi are expressed as gradients.

  Next, assuming that there are a total of R small regions selected by the small region selection unit 13 (three in the example of FIG. 2), each of the small regions is assumed to be independent. Using Equation 6, the simultaneous distribution of y = {yi, y2,..., YR} and x = {x1, x2,..., XR} is expressed as Equation 7. However, p = {p1, p2,..., PR}.

  Next, the parameter estimation unit 14 obtains an unknown parameter θ. In this embodiment, as an example, a procedure using an EM algorithm, which is a known procedure, will be described. The EM algorithm is a technique for obtaining an unknown parameter that maximizes the likelihood by iterative calculation. In the present embodiment, the likelihood is a function related to θ of the probability that the input image y observed by the blurred input image generation model of FIG. 3 is generated, and is expressed by a conditional distribution P (y | θ). The That is, the unknown parameter θ that gives the highest probability that the actually observed input image y is generated is obtained by the generation model shown in FIG.

Hereinafter, each step of the EM algorithm will be described. First, initialization in the EM algorithm is performed. Here, the parameter estimation unit 14 initializes an unknown parameter with an expected value, and sets this as θ ^OLD as shown in Equation 8. In the EM algorithm, the finally obtained value (unknown parameter θ in the present embodiment) may depend on the value determined by initialization. For this reason, the value determined by initialization should be an expected value, but any value may be determined if it cannot be predicted.

Next, the E step in the EM algorithm is performed. Here, the parameter estimation unit 14 obtains the posterior distribution q (x) of x with respect to the current unknown parameter θ ^{OLD using} the original image x as an unknown latent variable. The posterior distribution q (x) is expressed as Equation 9.

Next, the parameter estimation unit 14 obtains an expected value for q of the complete data log likelihood log p (x, y | p, θ), and this is expressed as Q (θ, θ ^OLD ) as shown in Equation 10. To do. Θ in the complete data log likelihood is a variable, not the current value θ ^OLD (fixed value).

Next, the M step in the EM algorithm is performed. Here, the parameter estimation unit 14 ^obtains θ that maximizes Q (θ, θ ^OLD ) of Formula 10 obtained in E step, and sets this as the updated unknown parameter θ ^new . θ ^new is expressed as in Expression 11.

When the parameter estimation unit 14 obtains θ ^new represented by Equation 11, it determines whether the obtained θ ^new is a converged value. If θ ^new is not a converged value, the parameter estimation unit 14 substitutes θ ^new for θ ^OLD in Equation 9, and repeats from step E again. As a method for determining the convergence of θ, for example, there is a method of confirming whether or not the rate of change of each element of θ is below a predetermined value. Assuming that θ ^new after convergence is θ ^* , θ ^* is expressed as shown in Equation 12.

As described above, the parameter estimation unit 14 estimates the unknown parameter θ, that is, the coefficient a, the optical axis position c, and the variance γ of the noise n. In the E step and the M step, for example, when the prior distribution p (x) of the original image x is a Gaussian distribution, the unknown parameter θ is calculated, but q (x) and Q (θ, θ ^OLD ) In some cases, θ may not be calculated. In such a case, the parameter estimation unit 14 may obtain the unknown parameter θ using, for example, a known sampling method, variation method, generalized EM algorithm, or the like.

<Description of estimation procedure of original image>
Next, a procedure for the blur correction unit 15 to estimate the original image x will be described. As this estimation procedure, a known one may be used, and an outline thereof will be described here.
First, when the expected value of x related to the posterior distribution of x shown in Equation 9 is obtained, or when x that maximizes the posterior distribution is obtained, the blur correction unit 15 performs Equation 13 and Equation 14 respectively. Thus, these values may be used as the estimated value x ^* of the original image.

In this embodiment, the above-described EM algorithm estimates the blur PSF parameters (a ^* , c ^* ) and the variance γ ^{* of} noise n. Therefore, the blur correction unit 15 may estimate the original image using a known Wiener filter.

Similarly, since the parameters (a ^* , c ^* ) of the blur PSF are estimated by the EM algorithm, the blur correction unit 15 is a Richardson-, which is a general devolution algorithm when the PSF is known. The original image may be estimated using Lucy decomposition.

<Description of Flowchart for Estimating Original Image>
Next, a flow in which the image processing apparatus 1 estimates an original image will be described. FIG. 5 is a flowchart illustrating an example of a procedure by which the image processing apparatus 1 according to the present embodiment estimates an original image.

  First, the image receiving unit 11 receives input image data from the outside (step 101). Then, the image receiving unit 11 transmits the input image data to the image dividing unit 12 and the blur correction unit 15. The image dividing unit 12 divides the input image based on the input image data transmitted from the image receiving unit 11 vertically and horizontally into m × n small regions (step 102). Then, the image dividing unit 12 divides the input image into a plurality of small areas, and then transmits the input image data to the small area selecting unit 13.

  The small region selection unit 13 calculates entropy indicating the degree of distribution of pixel values for each small region of the input image divided by the image dividing unit 12 (step 103). Next, the small region selection unit 13 selects the top K small regions having the largest entropy based on the calculated entropy (step 104). Then, the small region selection unit 13 transmits the selected K small regions to the parameter estimation unit 14.

The parameter estimation unit 14 performs K small transmissions transmitted from the small region selection unit 13 on the PSF (k) in the generation model of the blurred input image, the function defined for the noise n, and the prior distribution of the original image x. Using the region, unknown parameters (a ^* , c ^* , γ ^* ) are estimated (step 105). Here, the parameter estimation unit 14 represents the distribution of the input image yi as a conditional distribution, where pi is the center coordinate of the i-th small region ri, xi is the original image of the small region ri. Further, the parameter estimation unit 14 obtains a simultaneous distribution of xi and yi using Bayes' theorem. Furthermore, the parameter estimation unit 14 estimates an unknown parameter (a ^* , c ^* , γ ^* ) that maximizes the likelihood using the EM algorithm.

The blur correction unit 15 generates blur correction image data from the input image data transmitted from the image reception unit 11 using the unknown parameters (a ^* , c ^* , γ ^* ) estimated by the parameter estimation unit 14, An original image is estimated (step 106). Here, the blur correction unit 15 may use the K small regions transmitted from the small region selection unit 13 in the estimation of the original image. Thereafter, the image output unit 16 outputs the blur-corrected image data to the outside, and this processing flow ends.

  As described above, the image processing apparatus 1 according to the present embodiment estimates the PSF and the original image from the information of the blurred input image by estimating the parameters of the function defined for the blur occurring in the image. Do. When estimating the PSF or the original image, the image processing apparatus 1 may estimate few parameters such as the coefficient a and the optical axis position c by using the selected small area among the small areas obtained by dividing the input image. Further, the small area selected by the image processing apparatus 1 may be any number that is sufficient to estimate the parameters. Therefore, for example, the processing time for correcting the blur of the input image and estimating the original image is shortened as compared with the configuration in which the PSF is estimated for all the small regions of the input image. In addition, the image processing apparatus 1 uses a function representing a PSF that changes according to a position on the image, instead of using a fixed PSF for blurring that occurs in the input image. For this reason, even when non-uniform blur occurs in the image, the processing time for estimating the original image is shortened.

  Further, the image processing apparatus 1 estimates the original image from the information of the input image even if the PSF is not known. Therefore, the image processing apparatus 1 is also applied to, for example, an image sharing service in which a device that inputs an image to the image processing apparatus cannot be specified. Furthermore, the image processing apparatus 1 models the PSF by a function, and thus, the PSF and the original image are estimated with high accuracy and high speed as compared with, for example, a configuration in which the shape of the PSF is arbitrarily set and the original image is estimated. In addition, the image processing apparatus 1 performs high-precision blur correction using PSF, for example, as compared with a configuration in which image enhancement processing such as edge enhancement is performed according to a document area.

  In the present embodiment, if the Gaussian filter that defines the blur is increased, the accuracy of blur correction is improved. However, the larger the Gaussian filter, the longer the processing time for estimating unknown parameters and original images. For this reason, the size of the Gaussian filter is determined based on the required correction accuracy and processing time.

  Further, in the present embodiment, the variance of the Gaussian filter that defines the blurring of the image is expressed as in Equation 3, but the present invention is not limited to this. If the dispersion is expressed by the equation (3), the blur of the image is an isotropic blur in which the same blur occurs at a position where the distance from the optical axis position c is equal, such as a blur caused by a lens. expressed. On the other hand, for example, if the variance of the Gaussian filter is expressed as in Expression 15, the blur of the image is an anisotropic blur in which different blurs occur even at a location where the distance from the optical axis position c is equal. Represented as:

ただし、数１５式において、ｐは画像上の位置、ｃは光軸位置、ｂはぼけの大きさの勾配を制御するための係数である。また、δは数３式と同様に０よりは大きいが非常に小さい値が設定される。さらに、Σは分散共分散行列であり、ｔは転置、Σ^-１はΣの逆行列であることを表す。

In Equation 15, p is a position on the image, c is an optical axis position, and b is a coefficient for controlling the gradient of blur. Also, δ is set to a very small value that is larger than 0 as in Equation 3. Furthermore, Σ is a variance-covariance matrix, t is transposed, and Σ ⁻¹ is an inverse matrix of Σ.

  In the present embodiment, the blurred PSF of the image is defined by the Gaussian filter. However, the present invention is not limited to this. For example, the blurred PSF may be defined by another function such as an averaging filter.

<Description of Applicable Image Forming Apparatus>
The functions of the image processing apparatus 1 according to the embodiment of the present invention may be realized in an image forming apparatus including the image processing apparatus 1. Therefore, a configuration example of the image forming apparatus 20 will be described on the assumption that the processing of the image processing apparatus 1 is realized by the image forming apparatus 20. FIG. 6 is a diagram illustrating an example of the image forming apparatus 20 that implements the functions of the image processing apparatus 1 according to the present embodiment.

  An image forming apparatus 20 shown in FIG. 6 is a multi-function machine that is provided with, for example, a copying function, a scanner function, a printing function, a facsimile function, and the like, and forms an image on a sheet based on image data of each color. An image forming unit 21 as an example of a unit, a control unit 22 that controls the operation of the entire image forming apparatus 20, a communication unit 23 that communicates with a personal computer (PC), a scanner, and the like and receives image data, and a public And a facsimile (FAX) unit 24 that transmits and receives images through a line. Further, the image forming apparatus 20 is an example of a generation unit that generates blur correction image data by performing blur correction processing in the image processing apparatus 1 on the image data transferred from the communication unit 23 or the FAX unit 24. An image processing unit 25 is provided.

  In the image forming apparatus 20, under the operation control by the control unit 22, image forming processing by the following process is performed. First, the image processing unit 25 generates blur correction image data for the transferred image data, and then the generated blur correction image data is sent to the image forming unit 21. The image forming unit 21 forms and outputs an image on a sheet by an electrophotographic method or an ink jet method based on the blur correction image data.

<Description of applicable computers>
By the way, the processing of the image processing apparatus 1 in the embodiment of the present invention may be realized by a general-purpose computer. Therefore, the hardware configuration will be described assuming that this processing is realized by a computer.

  FIG. 7 is a diagram illustrating an example of a hardware configuration of the image processing apparatus 1 according to the present embodiment. As shown in the figure, the image processing apparatus 1 includes a CPU (Central Processing Unit) 31 that is a calculation means, a main memory 32 that is a storage means, and a magnetic disk device (HDD: Hard Disk Drive) 33. Here, the CPU 31 executes various software such as an OS (Operating System) and an application, and realizes each function of the image processing apparatus 1. The main memory 32 is a storage area for storing various types of software and data used for executing the software, and the magnetic disk device 33 stores programs for realizing the functions of the image processing apparatus 1 according to the present embodiment. Storing. Each function is realized by loading this program into the main memory 32 and executing processing based on this program by the CPU 31. Further, the image processing apparatus 1 includes a communication I / F 34 for performing communication with the outside.

  Specifically, in the image processing apparatus 1, an input image is divided, a small area is selected, an unknown parameter is estimated, an original image is estimated, and the like according to an instruction given by the CPU 31. In addition, reception of input image data and output of blur correction image data are performed via the communication I / F 34.

<Description of the program>
The processing performed by the image processing apparatus 1 according to the present embodiment described above is prepared as a program such as application software, for example.

  Therefore, the processing performed by the image processing apparatus 1 includes a function for receiving input image data, a function for dividing the received image data into a plurality of areas, and a position in the image where blur is generated in the image. 1 for calculating a value of a variable possessed by the acquired function from a plurality of divided areas and a function for obtaining a function having a plurality of variables. Or a function for selecting a plurality of areas, a function for calculating a value of a variable of the acquired function using the selected one or a plurality of areas, a value of the calculated variable, and received image data, Based on the above, it can also be understood as a program for realizing a function of correcting image blur in image data.

  The program for realizing the embodiment of the present invention can be provided not only by a communication means but also by storing it in a recording medium such as a CD-ROM.

DESCRIPTION OF SYMBOLS 1 ... Image processing apparatus, 11 ... Image reception part, 12 ... Image division part, 13 ... Small area | region selection part, 14 ... Parameter estimation part, 15 ... Blur correction part, 16 ... Image output part

Claims (5)

Receiving means for receiving input image data;
Dividing means for dividing the image data received by the receiving means into a plurality of regions;
An acquisition means for acquiring the function having a plurality of variables, the function being defined so that blur occurring in the image is represented according to a position in the image;
Selecting means for selecting one or a plurality of areas for calculating a value of a variable included in the function acquired by the acquiring means from the plurality of areas divided by the dividing means;
A calculating unit that calculates a value of a variable included in the function acquired by the acquiring unit, using the one or more regions selected by the selecting unit;
An image processing apparatus comprising: a correction unit that corrects image blur in the image data based on the value of the variable calculated by the calculation unit and the image data received by the reception unit.   The image processing apparatus according to claim 1, wherein the function is defined by a function representing a Gaussian distribution.   3. The image according to claim 1, wherein the selection unit selects, from among the plurality of regions divided by the division unit, the region having a non-uniform distribution of pixel values in the region. Processing equipment. Receiving means for receiving input image data, which is input image data;
Dividing means for dividing the input image data received by the receiving means into a plurality of regions;
An acquisition means for acquiring the function having a plurality of variables, the function being defined so that blur occurring in the image is represented according to a position in the image;
Selecting means for selecting one or a plurality of areas for calculating a value of a variable included in the function acquired by the acquiring means from the plurality of areas divided by the dividing means;
A calculating unit that calculates a value of a variable included in the function acquired by the acquiring unit, using the one or more regions selected by the selecting unit;
Generating means for generating blur-corrected image data obtained by correcting image blur in the input image data based on the value of the variable calculated by the calculating means and the input image data received by the receiving means;
An image forming apparatus comprising: an image forming unit that forms an image on a recording material based on the blur correction image data generated by the generating unit. On the computer,
A function to accept input image data;
A function of dividing the received image data into a plurality of regions;
A function defined so that blur occurring in the image is represented according to the position in the image, and a function of acquiring the function having a plurality of variables;
A function of selecting one or a plurality of regions for calculating a value of a variable included in the acquired function from the plurality of divided regions;
A function of calculating a value of a variable included in the acquired function using the selected one or more regions;
A program for realizing a function of correcting blurring of an image in the image data based on the calculated value of the variable and the received image data.

Images