JP2008118209A - Image processing program and image processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processing program and an image processor in which Retinex processing can be carried out at high speed with high precision. <P>SOLUTION: Since a pattern of the reference point in a mask for original image formed by preparation processing of a mask for original image corresponds to a pattern of the reference point in a mask for shrink image, distribution of reflectivity obtained from a shrink image is extremely close to distribution of reflectivity obtained from an original image. Consequently, an appropriate clip range (a clip range extremely approximate to a clip range obtained from an original image) can be obtained from a shrink image and Retinex processing can be carried out at high speed with high precision. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to an image processing program and an image processing apparatus, and more particularly to an image processing program and an image processing apparatus that can perform Retinex processing at high speed and with high accuracy.

  When a subject is imaged under backlight conditions, the image of the subject portion becomes an unclear backlight image with low brightness and contrast as the detailed mode becomes difficult to distinguish. In addition to such backlit images, poor images due to overexposure or underexposure, blurring or blurring during shooting, noise, insufficient light quantity, etc. are being used to improve image quality by improving brightness and contrast through image processing. Yes. As one method of such image processing, Retinex processing (Retinex processing) is known.

  In the Retinex process, the original image data is retained for the high-quality part, and the image quality is mainly improved for the low-quality part. In this Retinex processing, each pixel data of the original image is corrected to a value reflecting the pixel data of the peripheral pixels by a Gaussian filter, and the reflectance (reflectance of the original image) is calculated from the natural logarithm of the corrected pixel data. Luminance) component data is calculated, and the pixel component of the original image is divided by the pixel component of the original image to calculate the illuminance component data. That is, the original image is divided into two components, a reflectance component and an illuminance component. Then, after performing processing for correcting the brightness and gradation (contrast) such as gamma correction for the illuminance component, the corrected illuminance component and the reflectance component are combined to produce an original image. Image data in which the image quality of the low image quality portion such as the backlight image portion is improved can be generated. In order to obtain the reflectance component, the reflectance R (x, y) is normalized.

  Here, the normalization of the reflectance R (x, y) will be described with reference to FIG. The reflectance R (x, y) is calculated by the following equation.

ここで、ｘは、横方向の座標を、ｙは、縦方向の座標を、Ｉ（ｘ，ｙ）は、座標（ｘ，ｙ）における強度値を、Ｆ（ｘ，ｙ）は、座標（ｘ，ｙ）のフィルタ係数を、＊は、畳み込み演算（コンボリューション）をそれぞれ示し、この畳み込み演算により周辺平均輝度（周辺平均値）が求められる。なお、ｌｏｇは、ｅを底とする自然対数である。また、畳み込み演算では、マスクを構成する複数のフィルタ係数と画像を構成する画素値とが乗算され、その積が累算される。

Here, x is a coordinate in the horizontal direction, y is a coordinate in the vertical direction, I (x, y) is an intensity value at the coordinate (x, y), and F (x, y) is a coordinate ( The filter coefficient of x, y), * indicates a convolution operation (convolution), and the peripheral average luminance (peripheral average value) is obtained by this convolution operation. Note that log is a natural logarithm with e as the base. In the convolution operation, a plurality of filter coefficients constituting the mask are multiplied by the pixel values constituting the image, and the product is accumulated.

  FIG. 18 shows the normalization of an integer value in the range of 0 to 255, where the minimum value of the reflectance R (x, y) obtained by the above equation is 0 and the maximum value is 255 for a certain still image. It is the histogram which totaled appearance frequency.

  The median value M is obtained from this histogram, the upper limit value of the range including 45% of pixels from the median value to the larger side is U, and the lower limit value of the range including 45% of pixels from the median value to the smaller side is D. .

  The values of R (x, y) corresponding to the values of U and D are UpR and DownR, and the normalized reflectivity refle (x, y) is as follows when the value of R (x, y) is equal to or less than DownR: 0.0, when the value of R (x, y) is equal to or higher than UpR, 1.0, and when the value of R (x, y) is larger than DownR and smaller than UpR,

とする。

And

  Since refle (x, y) is obtained in this way, R (x, y) must be stored for all pixels in order to obtain the clipped range (between UpR and DownR). Moreover, R (x, y) is a numerical value obtained by logarithmic calculation, so it must be recorded using a decimal point. For floating point, it is 4 bytes, and for double precision real numbers, 8 bytes for each pixel. In order to memorize | store, a huge memory capacity is required.

  Japanese Patent Laid-Open No. 2001-69525 (Patent Document 1) discloses a solution to the problem that color balance is lost or color misregistration occurs when this Retinex process is performed independently for each of the RGB planes. Discloses a method of converting RGB values into a coordinate space composed of luminance components and color components such as YCbCr and YIQ, performing Retinex processing only on the luminance component Y, and returning to RGB while maintaining the color components. Has been. When this method is used, only the luminance component is adjusted, and the color component is not adjusted, so that the color balance is not lost and color shift does not occur. Further, in this method, since the Retinex process is performed only on the luminance component, the calculation amount is small and the process can be executed at a high speed as compared with the case where the Retinex process is performed on each of the RGB planes. In order to normalize x, y), it is not necessary to store each of RGB, and only the luminance needs to be stored, so that the necessary storage capacity is reduced.

Japanese Patent No. 3731577 (Patent Document 2) discloses a method for increasing the processing speed of the Retinex processing. In this method, a reduced image (low resolution) is formed from the original image by a method such as an average pixel method, and a peripheral average luminance image (blurred image) obtained by calculating a peripheral average luminance for each pixel of the reduced image is formed. A retinex processed image is formed from the enlarged image of the blurred image and the original image, and an output image is formed from the retinex processed image and the original image.
JP 2001-69525 A Japanese Patent No. 3731577

  However, in the processing disclosed in the conventional literature, the reflectance is calculated and stored for all the pixels of the original image, the normalization parameter is determined after obtaining the reflectance for all the pixels, and then the reflectance is calculated. Since the pixel value is corrected and output by the Retinex value obtained by normalizing the above, there is a problem that it takes a long time from the start of processing to the start of output of the pixel value.

  SUMMARY An advantage of some aspects of the invention is that it provides an image processing program and an image processing apparatus that can perform Retinex processing at high speed and with high accuracy.

  As a result of intensive studies to solve the above problems, the present inventor obtained the reflectance of the reduced image prior to calculating the reflectance of the original image, and normalized using the reflectance of the reduced image. It was conceived to determine the optimization parameters. In order to obtain the normalization parameter, using the reduced image can reduce the number of pixels for which the reflectance is to be calculated, so that processing can be performed at high speed. In addition, by determining the normalization parameters in advance, the reflectance calculated for the original image can be normalized and output immediately without waiting for the calculation of the reflectance of other pixels. Therefore, the time until the start of image output can be shortened, and the processing speed can be increased.

  As described above, when deriving the normalization parameter from the reduced image, there is a possibility that the image quality may deteriorate if the difference between the normalization parameter obtained using the reduced image and the normalization parameter obtained using the original image is large. It is desirable that the difference between these values is small.

  As a result of further studies, the present inventor has made the normalization parameter (that is, the original image) suitable even when the reduced image is used by equalizing the coverage of the reduced image mask and the coverage of the original image mask. It was found that a value closer to the normalization parameter obtained from the reflectance of the image was obtained, and a highly accurate output result was obtained.

  This will be described in more detail with reference to FIG. FIG. 19 is a diagram schematically showing a reduced image mask and an original image mask. The mask is a region where an image processing filter is applied to each pixel around the pixel when the peripheral average luminance of the pixel of interest is obtained, and the size of the region is the size of the mask. As described above, when calculating the reflectance for each pixel, the peripheral average luminance, which is the average value of the pixels around the target pixel, is calculated. At that time, the peripheral average luminance is set as the “peripheral pixel”. The area where the value is reflected is determined by the size of the mask. The mask coverage is the ratio of the mask size to the size of the image to be masked. For example, when the length of each side of the original image is 100, the mask of the original image mask If the length of each side is also 100, the coverage of the original image mask is 1.

  As shown in FIG. 19, making the coverage ratio equal means that the ratio of the size of the reduced image mask to the size of the reduced image is made equal to the ratio of the size of the original image mask to the size of the original image.

  For example, when the original image mask has a size that covers 64 (8 × 8) pixels of the original image, the reduced image reduced at the reduction ratio of 1/2 includes 16 (4 × 4) pieces in the reduced image. By using a reduced image mask of a size that covers these pixels, the coverage can be made equal. By making the coverage ratios equal in this way, the ranges that are referred to when calculating the reflectance of one pixel of interest match, so that a suitable normalization parameter can be obtained even when the reflectance of the reduced image is used. It is.

  In order to achieve the above object, an image processing program according to claim 1 is a program executed by an image processing apparatus that performs image correction processing, wherein each pixel of the reduced image obtained by reducing the original image is reduced image. A reduced image reflectance acquisition step of obtaining a peripheral average value using a mask for the image, and obtaining a reflectance from the peripheral average value and the value of each pixel of the reduced image, and the reflectance obtained by the reduced image reflectance acquisition step The normalization parameter setting step that sets the parameter value when performing normalization by summing up the frequency of each value in the entire range of values taken by, and the peripheral average value for each pixel of the original image using the mask for the original image And obtaining the reflectance from the peripheral average value and the pixel value of the original image, and the reflectance obtained by the original image reflectance obtaining step as the normality A correction step of obtaining a normalized Retinex value normalized based on the parameter value set by the parameter setting step, and performing correction based on the normalized Retinex value and the pixel value of the original image, The ratio between the size of the reduced image and the size of the reduced image filter is the same as the ratio between the size of the original image and the size of the original image filter.

  The image processing program according to claim 2 is the image processing program according to claim 1, wherein the reduced image filter defines a calculation area of a peripheral average value in the reduced image and includes p rows and q columns in the calculation area. The pixel is divided into a reference pixel whose value is reflected in the peripheral average value and a thinned pixel whose value is not reflected in the peripheral average value. The original image is divided into a plurality of blocks for each pixel in k rows and k columns. A reduced image forming step of dividing and consolidating each block into one pixel to form a reduced image, and pixels in p rows and q columns of the reduced image defined by the calculation area of the reduced image filter are the original image When the pixels of k, p rows, k, q columns of the original image defined by the calculation area of the filter for use are reduced, each block of the original image that is the source of each reference pixel in the reduced image has a small amount. Both those comprising 1 to the reference pixel is included, and an original image filter preparation step of preparing an original image filter for dividing the reference pixels and thinning pixels.

  According to a third aspect of the present invention, there is provided the image processing program according to the second aspect, wherein in the original image filter preparation step, pixels of p rows and q columns of the reduced image determined by the calculation area of the reduced image filter are calculated. , Each block of the original image that is the source of each thinned pixel in the reduced image when the pixels of k, p rows, k, and q columns of the original image defined by the calculation area of the original image filter are reduced An original image filter for distinguishing the reference pixel and the thinned pixel is prepared so that the inner pixel becomes the thinned pixel.

  The image processing program according to claim 4 is the image processing program according to claim 2 or 3, wherein the reduced image forming step extracts n pixels from each block (n is an integer of 1 or more), One pixel is generated based on the extracted n pixels to form a reduced image, and the original image filter preparation step includes p rows q of the reduced image determined by the reduced image filter. When the pixels in the column are in a relationship in which the pixels of k, p rows, k, and q columns of the original image defined by the original image filter are reduced, each of the original images that are the sources of the reference pixels of the reduced image An original image filter is provided for dividing a reference pixel and a thinned-out pixel so that at least one of the n extracted pixels in the block is a reference pixel.

  The image processing program according to claim 5 is the image processing program according to any one of claims 2 to 4, further comprising a reduced image filter preparation step for preparing the reduced image filter, and the original image filter preparation step. Is for generating an original image filter based on the reduced image filter prepared in the reduced image filter preparation step.

  The image processing apparatus according to claim 6, which performs image correction processing, obtains a peripheral average value for each pixel of a reduced image obtained by reducing the original image using a reduced image filter, and the peripheral average value thereof And the reduced image reflectance acquisition means for obtaining the reflectance from the value of each pixel of the reduced image, and the frequency of each value in the entire range of values taken by the reflectance obtained by the reduced image reflectance acquisition means, Normalization parameter setting means for setting parameter values for normalization, and for each pixel of the original image, a peripheral average value is obtained using the original image filter, and from the peripheral average value and the pixel value of the original image An original image reflectance acquisition unit for obtaining the reflectance, and a normalized retine that normalizes the reflectance acquired by the original image reflectance acquisition unit based on the parameter value set by the normalization parameter setting unit Correction means for performing correction based on the normalized Retinex value and the pixel value of the original image, and the ratio between the size of the reduced image and the size of the filter for the reduced image is The ratio is the same as the ratio between the size of the original image and the size of the original image filter.

  The image processing apparatus according to claim 7 is the image processing apparatus according to claim 6, wherein the reduced image filter defines a calculation area of a peripheral average value in the reduced image and includes p rows and q columns in the calculation area. The pixel is divided into a reference pixel whose value is reflected in the peripheral average value and a thinned pixel whose value is not reflected in the peripheral average value. The original image is divided into a plurality of blocks for each pixel in k rows and k columns. A reduced image forming unit that divides and aggregates each block into one pixel to form a reduced image, and pixels in p rows and q columns of the reduced image defined by the calculation area of the reduced image filter include the original image When the pixels of k, p rows, k, q columns of the original image determined by the calculation area of the filter for use are reduced, each block of the original image that is the source of each reference pixel in the reduced image has at least 1 Reference picture of To include, in which and an original image filter preparation means for preparing the original image filter for dividing the reference pixels and thinning pixels.

  The image processing device according to claim 8 is the image processing device according to claim 7, wherein the original image filter preparation means includes pixels of p rows and q columns of the reduced image defined by the calculation region of the reduced image filter. , Each block of the original image that is the source of each thinned pixel in the reduced image when the pixels of k, p rows, k, and q columns of the original image defined by the calculation area of the original image filter are reduced An original image filter for distinguishing the reference pixel and the thinned pixel is prepared so that the inner pixel becomes the thinned pixel.

  The image processing device according to claim 9 is the image processing device according to claim 7 or 8, wherein the reduced image forming unit extracts n pixels from each block (n is an integer of 1 or more), One pixel is generated on the basis of the extracted n pixels to form a reduced image, and the original image filter preparation means includes p rows q of the reduced image determined by the reduced image filter. When the pixels in the column are in a relationship in which the pixels of k, p rows, k, and q columns of the original image defined by the original image filter are reduced, each of the original images that are the sources of the reference pixels of the reduced image An original image filter is provided for dividing a reference pixel and a thinned-out pixel so that at least one of the n extracted pixels in the block is a reference pixel.

  The image processing apparatus according to claim 10, wherein the image processing apparatus according to claim 7 includes a reduced image filter preparation unit that prepares the reduced image filter, and the original image filter preparation unit. Is for generating an original image filter based on the reduced image filter prepared by the reduced image filter preparation means.

  According to the image processing program of claim 1, in the reduced image reflectance acquisition step, for each pixel of the reduced image obtained by reducing the original image, a peripheral average value is obtained using the reduced image filter, and the peripheral average value thereof The reflectance is obtained from the value of each pixel of the reduced image. Then, in the normalization parameter setting step, the frequency of each value in the entire range of values taken by the reflectance obtained in the reduced image reflectance acquisition step is aggregated, and parameter values for normalization are set. Then, in the original image reflectance acquisition step, a peripheral average value is obtained for each pixel of the original image using the original image filter, and a reflectance is obtained from the peripheral average value and the pixel value of the original image. Then, a normalizing retinex value obtained by normalizing the reflectance obtained by the original image reflectance obtaining step based on the parameter value set by the normalizing parameter setting step is obtained by the correcting step, and the normalized latency Correction is performed based on the neck value and the pixel value of the original image. Therefore, the number of operations for obtaining the normalization parameter is small, and the processing can be performed at high speed.

  In addition, since the ratio between the size of the reduced image and the size of the reduced image filter is the same as the ratio between the size of the original image and the size of the original image filter, even when the reflectance of the reduced image is used. A suitable normalization parameter can be obtained, and the Retinex processing can be performed with high accuracy.

  According to the image processing program of claim 2, in addition to the effect of the image processing program of claim 1, the pixels of p rows and q columns of the reduced image defined by the calculation area of the reduced image filter are used for the original image. At least one reference is made to each block of the original image that is the source of each reference pixel in the reduced image when the pixels of k, p rows, k, and q columns of the original image defined by the calculation area of the filter are reduced. Since pixels are included, the pixel value reflected in the reflectance of the reduced image approximates the pixel value reflected in the reflectance of the original image. Therefore, even when the reflectance of the reduced image is used, a suitable normalization parameter can be obtained, and there is an effect that the Retinex process can be performed with high accuracy. Note that “·” means multiplication.

  According to the image processing program of the third aspect, in addition to the effect produced by the image processing program of the second aspect, the pixels in the p rows and q columns of the reduced image defined by the calculation area of the reduced image filter are used for the original image. When pixels in k, p rows, k, and q columns of the original image defined by the calculation area of the filter are reduced, the pixels in each block of the original image that is the source of each thinned pixel in the reduced image are thinned pixels Therefore, the pixel value that is thinned out when the reflectance of the reduced image is calculated approximates the pixel value that is thinned out when the reflectance of the original image is calculated. In other words, the pixel value reflected in the reflectance of the reduced image approximates the pixel value reflected in the reflectance of the original image. Therefore, even when the reflectance of the reduced image is used, a suitable normalization parameter can be obtained, and there is an effect that the Retinex process can be performed with high accuracy.

  According to the image processing program of the fourth aspect, in addition to the effect produced by the image processing program according to the second or third aspect, pixels of p rows and q columns of the reduced image determined by the reduced image filter are used for the original image. Of the n extracted pixels in each block of the original image that is the source of each reference pixel of the reduced image when the pixels of k, p rows, k, and q columns of the original image defined by the filter are reduced Since at least one pixel serves as a reference pixel, the pixel value reflected in the reflectance of the reduced image is more approximate to the pixel value reflected in the reflectance of the original image, and the reflectance of the reduced image is used. In this case, a suitable normalization parameter can be obtained, and the Retinex process can be performed with higher accuracy.

  According to the image processing program of claim 5, in addition to the effect produced by the image processing program according to any of claims 2 to 4, based on the reduced image filter prepared by the reduced image filter preparation step, Since the original image filter is generated, there is an effect that an appropriate normalization parameter can be obtained while maintaining the correspondence between the original image filter and the reduced image filter regardless of the reduction ratio. On the contrary, if the original image filter is reduced to generate a reduced image filter, if the reduction ratio is extremely large, that is, if the reduced image is extremely small relative to the original image, the reduced image filter is There is a problem that the reflectance of the reduced image cannot be appropriately calculated because it becomes too small.

  According to the image processing device of claim 6, the reduced average image reflectance obtaining unit obtains the peripheral average value for each pixel of the reduced image obtained by reducing the original image using the reduced image filter, and the peripheral average value thereof. The reflectance is obtained from the value of each pixel of the reduced image. Then, the normalization parameter setting means sums up the frequency of each value in the entire range of values obtained by the reflectance obtained by the reduced image reflectance acquisition means, and sets the parameter value for normalization. Then, the original image reflectance acquisition means calculates the peripheral average value for each pixel of the original image using the original image filter, and the reflectance is determined from the peripheral average value and the pixel value of the original image. Then, the correction means obtains a normalized Retinex value obtained by normalizing the reflectance acquired by the original image reflectance acquisition means based on the parameter value set by the normalization parameter setting means, and the normalized Retinex value is obtained. Correction is performed based on the value and the pixel value of the original image. Therefore, the number of operations for obtaining the normalization parameter is small, and the processing can be performed at high speed.

  In addition, since the ratio between the size of the reduced image and the size of the reduced image filter is the same as the ratio between the size of the original image and the size of the original image filter, even when the reflectance of the reduced image is used. A suitable normalization parameter can be obtained, and the Retinex processing can be performed with high accuracy.

  According to the image processing apparatus of the seventh aspect, in addition to the effect produced by the image processing apparatus of the sixth aspect, the pixels of p rows and q columns of the reduced image defined by the calculation area of the reduced image filter are used for the original image. At least one reference is made to each block of the original image that is the source of each reference pixel in the reduced image when the pixels of k, p rows, k, and q columns of the original image defined by the calculation area of the filter are reduced. Since pixels are included, the pixel value reflected in the reflectance of the reduced image approximates the pixel value reflected in the reflectance of the original image. Therefore, even when the reflectance of the reduced image is used, a suitable normalization parameter can be obtained, and there is an effect that the Retinex process can be performed with high accuracy.

  According to the image processing apparatus of the eighth aspect, in addition to the effect produced by the image processing apparatus of the seventh aspect, the pixels in the p rows and q columns of the reduced image defined by the calculation area of the reduced image filter are used for the original image. When pixels in k, p rows, k, and q columns of the original image defined by the calculation area of the filter are reduced, the pixels in each block of the original image that is the source of each thinned pixel in the reduced image are thinned pixels Therefore, the pixel value that is thinned out when the reflectance of the reduced image is calculated approximates the pixel value that is thinned out when the reflectance of the original image is calculated. In other words, the pixel value reflected in the reflectance of the reduced image approximates the pixel value reflected in the reflectance of the original image. Therefore, even when the reflectance of the reduced image is used, a suitable normalization parameter can be obtained, and there is an effect that the Retinex process can be performed with high accuracy.

  According to the image processing device of the ninth aspect, in addition to the effect produced by the image processing device according to the seventh or eighth aspect, the pixels of p rows and q columns of the reduced image defined by the reduced image filter are used for the original image. Of the n extracted pixels in each block of the original image that is the source of each reference pixel of the reduced image when the pixels of k, p rows, k, and q columns of the original image defined by the filter are reduced Since at least one pixel serves as a reference pixel, the pixel value reflected in the reflectance of the reduced image is more approximate to the pixel value reflected in the reflectance of the original image, and the reflectance of the reduced image is used. In this case, a suitable normalization parameter can be obtained, and the Retinex process can be performed with higher accuracy.

  According to the image processing device of claim 10, in addition to the effect produced by the image processing device according to any of claims 7 to 9, based on the reduced image filter prepared by the reduced image filter preparation means, Since the original image filter is generated, there is an effect that an appropriate normalization parameter can be obtained while maintaining the correspondence between the original image filter and the reduced image filter regardless of the reduction ratio. On the other hand, if the original image filter is reduced to generate a reduced image filter, the reduced image filter is used when the reduction ratio is extremely large, that is, when the reduced image is extremely small with respect to the original image. As a result, the problem arises that the reflectance of the reduced image cannot be calculated appropriately.

  Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings. FIG. 1 is a block diagram illustrating an electrical configuration of a printer 1 having a function of performing image processing according to the present embodiment. In this embodiment, an image processing program installed in the printer 1 is stored in image data (original image data) input from a personal computer (hereinafter referred to as “PC”) 2, a digital camera 21, or an external medium 20. Retinex processing (Retinex processing) or the like is executed, and correction of a low image quality region such as a backlight image portion of image data is executed.

  As shown in FIG. 1, the printer 1 includes a CPU 11, a ROM 12, a RAM 13, a print head, and the like, and includes a printing unit 15 that performs printing (output) on a print medium (for example, a paper medium), an output image size, and the like. And an operation panel 16 having a user operation unit (for example, a numeric keypad) through which the user can input the input value.

  The printer 1 includes an interface (hereinafter referred to as “I / F”) 17 that can be connected to the PC 2 via the cable 5, an I / F 18 that can be connected to the digital camera 21 via the cable 6, and an external medium 20. And an external media slot 19 that can be detachably mounted (for example, a flash memory such as an SD memory card or a memory stick). As a communication method performed by these I / Fs 17 and 18, USB (Universal Serial Bus) is used.

  Therefore, the printer 1 can input the image data stored in the PC 2 via the cable 5 and the I / F 17, and the image data captured by the digital camera 21 can be input via the cable 6 and the I / F 18. It is possible to input via. Further, it is possible to input image data stored in the external medium 20 from the external medium 20 mounted in the external media slot 19.

  The CPU 11 is an arithmetic processing device that controls the entire printer 1. The ROM 12 stores various control programs executed by the CPU 11, fixed values used when executing the programs, and the like. The image processing program stores an image processing program that performs image processing such as Retinex processing. A memory 12a, a print control program memory 12b for storing a print control program for performing printing, a lookup table memory (LUT memory) 12c, and the like are provided.

  The RAM 13 has a working area for storing various register groups required when the control program is executed by the CPU 11, a temporary area for temporarily storing data being processed, and the like, and can be accessed at random. An original image memory 13a for storing original image data; a reduced Retinex image memory 13b for storing the reflectance required in the Retinex processing of the reduced image after reducing the original image; and each pixel in the histogram processing A histogram memory 13c for storing the frequency of values, an original image size memory 13d for storing the size of the original image, a print parameter memory 13e for storing print parameters for printing set by the user, etc. Reduced image size determined based on parameters etc. It includes a reduction algorithm memory 13g and the reduced image size memory 13f which small algorithms are stored, respectively, the mask memory 13h for storing the mask (filter) used in obtaining the surrounding average luminance of the target pixel.

  The original image data memory 13 a stores image data input from the PC 2, the digital camera 21, and the external media 20 via the I / F 17, I / F 18, and external media slot 19, respectively. When original image data is stored in units of one line and data of the number of lines that can form a reduced image is read, a reduced image is formed. In the present embodiment, both the original image data and the output image data are composed of RGB values, and each RGB value is represented by a data size of 8 bits (1 byte). ”To“ 255 ”.

  The RGB value is a value having as constituent components an R value representing red, which is the three primary colors of light, a G value representing green, and a B value representing blue. Since various colors are generated by mixing the three primary colors of light, the color of each pixel of the original image is one color (hue, gradation, etc.) by a combination of R value, G value, and B value (RGB value). ) Is displayed. The luminance (lightness) increases as the RGB value increases.

  The reduced Retinex image memory 13b is a memory that stores the reflectance Rs (x, y) obtained in the Retinex process for only the luminance signal of the reduced image by reducing the original image. As a method of reducing the original image, the nearest neighbor (nearest) that samples (extracts) the value of one pixel closest to the position of the original image corresponding to the pixel of the image to be reduced and uses the value of the sampled pixel as it is. Neighbor) method, bilinear method that samples the values of four pixels around the position of the original image corresponding to the pixels of the image to be reduced, and performs interpolation using the sampled pixel values, the number according to the reduction rate The average pixel method in which the pixel values are sampled and the average value is used as the value of one pixel of the reduced image, the values of 16 pixels are sampled, and the values are interpolated to obtain the value of one pixel of the reduced image. The bicubic method is known. When the processing speed is prioritized over the image quality, the nearest neighbor method is used, and when the image quality is prioritized over the processing speed, the bicubic method or the average pixel method is used.

  The original image is reduced by any one of these methods, and the reflectance is calculated for each pixel (pixel) of the reduced image using Equation 1 described above, and stored in the reduced Retinex image memory 13b.

  The histogram memory 13c is a memory for forming a histogram by summing up the frequencies of the reflectances Rs (x, y) of each pixel of the reduced image. When a histogram is created, an upper limit value and a lower limit value (clip range), which are parameters for performing normalization, are determined based on the histogram.

  The original image size memory 13d stores the size of the original image. When the original image data is read from the PC 2 or the like, it is read along with the original image data and stored in the original image size memory 13d. When determining the size of the reduced image and the reduction algorithm, the original image size stored in the original image size memory 13d is referred to.

  The print parameter memory 13e determines whether the print mode for printing is a high-quality photo mode or a normal image quality normal mode, and the recording medium type is glossy paper, inkjet paper, or plain paper. The user sets whether the size of the recording medium is A4, B5, letter, or the like, and stores the set parameters. When the printing mode is selected, the printing resolution and droplet size set in the selected printing mode are stored as parameters, and when printing is performed, printing is performed according to these parameters.

  Based on these print parameters and the size of the original image, the size of the reduced image and the algorithm to be reduced are determined and stored in the reduced image size memory 13f and the reduced algorithm memory 13g, respectively. Mainly, when the image quality is given priority over the processing speed, a large size is set as the reduced image size, and when the processing speed is given priority over the image quality, a small size is set as the reduced image size. The reduction algorithm is selected from the above-mentioned nearest neighbor method, bilinear method, average pixel method, and the like.

  These print parameters are set by the user in the PC 2 and input to the printer 1, and can also be set by an operator provided on the operation panel 16.

  The mask memory 13h is a memory for storing a mask (filter) used when obtaining the peripheral average luminance of each pixel, and a reduced image mask to be applied to the reduced image and an original image mask to be applied to the original image (see FIG. 8). In the present embodiment, the reduced image mask is stored in advance in the ROM 12, and is described as being read out to the mask memory 13h as necessary. On the other hand, it is assumed that the original image mask is dynamically generated based on the reduced image mask read from the ROM 12 and stored in the mask memory 13h.

  The operation panel 16 includes an LCD for displaying printing parameters and the like, and various operators for setting printing parameters and parameters relating to image processing, and instructing execution of image processing and printing processing.

  Next, various printing parameters set in the PC 2 connected to the printer 1 will be described with reference to FIG. FIG. 2 shows a print parameter setting screen 28 displayed on the display unit when the PC 2 selects the print parameter setting.

  On the print parameter setting screen 28, a paper type selection box 28a for selecting and setting a printing paper as a recording medium for printing, a paper size setting box 28b for selecting and setting the size of the printing paper, and a printing mode for setting a printing mode. A setting box 28c and a number setting box 28d for setting the number of copies to be printed are displayed.

  The paper type selection box 28a has a display area for displaying the selected paper type, and an icon in which a triangle pointing downward is drawn on the right end of the area. When clicked, a pull-down menu is displayed as shown in FIG. When the cursor is moved to one of the items in the next displayed pull-down menu and the switch provided on the mouse is clicked, the item designated by the cursor is selected.

  In this embodiment, either plain paper or glossy paper can be selected as the type of printing paper, and the example shown in FIG. 2 shows a state in which plain paper is selected.

  Similarly, the paper size setting box 28b has an area for displaying the size of the selected paper, and an icon for instructing display of a pull-down menu for selection. The paper sizes include A4, letter, Either 5 "x 7" or 4 "x 6" can be selected.

  Similarly, the print mode setting box 28c has an area for displaying the selected print mode and an icon for instructing display of a pull-down menu for selection, and the print mode is selected from the photo mode and the normal mode. Can be selected. The photo mode is a mode for printing with high image quality compared to the normal mode, and printing is performed with high resolution. For example, the resolution of the photo mode is 1200 × 1200 dpi, and the resolution of the normal mode is a low resolution of 600 × 600 dpi. Note that in an inkjet printer, the droplet size and the type of ink used may differ depending on the resolution.

  The number of copies setting box 28d consists of an area for displaying the set number of copies as a numerical value. On the right side of the area, an increment icon displaying an upward triangle for increasing the numerical value and a downward pointing for decreasing the numerical value. A decrement icon displaying a triangle is provided, and the number of copies can be set by moving the cursor to these icons and operating the mouse.

  In addition to the above icons, the print parameter setting screen 28 includes a radio button for setting the print direction with respect to the paper, an OK button for enabling the setting and instructing the erasing of the setting screen, and a setting screen for disabling the setting. A cancel button for instructing deletion and a help button for instructing display of a setting explanation screen are displayed.

  The print parameters set by the PC 2 in this way are input to the printer 1 via the cable 5 and the I / F 17 and stored in the print parameter memory 13e.

  Next, image processing executed by the CPU 11 will be described with reference to FIGS. 3 and 4. 3 and 4 are flowcharts showing image processing. The process shown in FIG. 3 is a process called pre-processing. In this process, the original image is reduced, and a clip range for normalizing the reflectance R is set based on the reduced image.

  In this preprocessing, first, data indicating the size of the original image attached to the original image data is stored in the original image size memory 13d (S1). The data indicating the size of the image is a value represented by the number of vertical and horizontal pixels that normally form a rectangular image.

  Next, parameters necessary for setting the reduced image size and the reduction algorithm are read out from the print parameters stored in the print parameter memory 13e (S2). In this embodiment, among the print parameters, the print mode, paper type, and paper size are necessary parameters.

  Next, the reduced image size and the reduction algorithm are determined based on these parameters and the like, and the determined reduced image size is stored in the reduced image size memory 13f and the reduction algorithm in 13g, respectively (S3). This process will be described later with reference to FIGS.

  Next, a reduced image mask is prepared (S4). Specifically, the reduced image mask is read from the ROM 12 and stored in the mask memory 13h.

  Next, a reduced image is formed. In the process of forming a reduced image, first, it is read in units of one line from the PC 2 or the like storing the original image and stored in the original image memory 13a of the RAM 13 (S6). Image data formed by a digital camera or the like is compressed and stored by a compression method such as JPEG, and is stored in the order of lines of pixels in the horizontal direction of a rectangular image.

  Next, it is determined whether or not the number of lines of image data stored in the original image memory 13a is stored so that the number of lines can be reduced by the reduction algorithm stored in the reduced image algorithm memory 13g (S7). ) If the number of lines that can be reduced is not stored (S7: No), the process returns to S4. If the number of lines that can be reduced is stored (S7: Yes), the reduced image data is stored according to the reduction algorithm. Then, the reduced image is stored in the RAM (S9). The reduction algorithm is 1 line when the nearest neighbor method is used, 2 lines when the bilinear method is used, 3 lines when the bicubic method is used, and according to the reduction rate in the case of the average pixel method. The original image data of the number of lines corresponds to “the number of lines that can be reduced”. When the reduced image data is stored in the RAM and then the original image for one line is read and stored in the original image memory 13a, the previously stored data is overwritten. As a result, the storage capacity of the original image memory 13a can be reduced.

  Next, it is determined whether or not all the lines of the original image have been scanned to form a reduced image (S10). If unprocessed lines still remain (S10: No), the process of S4 is performed. Returning to FIG. 5, when the processing is completed for all lines (S10: Yes), each pixel of the reduced image is converted into a luminance signal Y and color signals Cb and Cr (S11).

  The luminance signal Y and the color signals Cb and Cr are calculated from the RGB values of the original pixels by the following formula.

輝度信号Ｙと色信号Ｃｂ，Ｃｒの値をそれぞれ記憶し、後の演算で使用するようにしてもよいが、記憶容量が小さい場合は、ＲＧＢ値のみを記憶し、必要に応じて演算して求めるようにしてもよい。

The values of the luminance signal Y and the color signals Cb and Cr may be stored and used in later calculations, but if the storage capacity is small, only the RGB values are stored and calculated as necessary. You may make it ask.

  Next, the reflectance Rs (x, y) is calculated for the luminance signal Y calculated by the above equation (S12). Note that “s” is a suffix for the reduced image, and no suffix is added to the original image. The reflectance Rs is calculated by the following equation, assuming that the luminance value of each pixel of the reduced image is Is (x, y) and the filter for the reduced image is Fs (x, y).

なお、「＊」は、畳み込み演算（コンボリューション）を示し、この畳み込み演算により周辺平均輝度（周辺平均値）が求められる。なお、ｌｏｇは、ｅを底とする自然対数である。

Note that “*” indicates a convolution operation (convolution), and the peripheral average luminance (peripheral average value) is obtained by this convolution operation. Note that log is a natural logarithm with e as the base.

  Next, the reflectance Rs (x, y) obtained by the above calculation is stored in the reduced Retinex image memory 13b of the RAM 13 (S13).

  Next, Rs (x, y) is compared with the maximum value and the minimum value (S14). Specifically, the reflectance Rs (x, y) obtained for the first coordinate is set as the maximum value and the minimum value, and the reflectance Rs (x, y) obtained from the next is compared with the maximum value and the minimum value, respectively. When the reflectance Rs (x, y) obtained this time is larger than the maximum value, the reflectance Rs (x, y) obtained this time is set as a new maximum value, and the reflectance Rs (x, y) obtained this time is obtained. Is smaller than the minimum value, the reflectance Rs (x, y) obtained this time is set as a new minimum value, and the reflectance Rs (x, y) obtained this time is smaller than the maximum value and larger than the minimum value. Is a process of not changing the maximum value and the minimum value.

  Next, it is determined whether or not the processing of S12 to S14 has been performed for all coordinates of the reduced image (S15). If there are still unprocessed coordinates (S15: No), the processing returns to S12 and all of them are performed. When the process for the coordinates of (1) is finished (S15: Yes), the reflectance Rs (x, y) is normalized based on the maximum value and the minimum value obtained by the process of S14, and a histogram is formed (S16). .

  Next, a median value is obtained from the formed histogram (S17), and a clipped range (clip range) of the reflectance Rs (x, y) is determined from the median value and the histogram (S18). This clip range is, for example, a value larger than the median value, and an upper limit value that includes 45% of all samples is the upper limit value of the clip range, and a lower value that is less than the median value and includes 45% of all samples. The value is the lower limit value of the clip range.

  Through the above pre-processing, a reduced Retinex image composed of the reduced image reflectance obtained by reducing the original image is formed, and the clip range for normalizing the reflectance of the original image is obtained from the reduced Retinex image. It is done. As a result, the number of calculations for obtaining the clip range can be made much smaller than the number of calculations for the original image, and the processing speed is increased. Further, since the Retinex image of the reduced image is stored, there is an advantage that the processing can be performed with a small storage capacity compared with the case of storing the Retinex image of the original image. Further, there is no large difference between the clip range obtained using the reduced image and the clip range obtained using the original image. The following table shows the difference between the clip range (upper limit and lower limit) obtained from the reduced image and the clip range obtained from the original image.

この表が示すように、元画像により求めた最大値は、２．７１２、最小値は、−４．０６３であり、縮小画像により求めた最大値は、１．７２９、最小値は、−２．６０７である。よって、元画像により求めた最大値と縮小画像により求めた最大値との差は、０．９８３であり、元画像により求めた最小値と縮小画像により求めた最小値との差は、１．４５６であって、かなり大きな差が認められる。

As shown in this table, the maximum value obtained from the original image is 2.712, the minimum value is −4.063, the maximum value obtained from the reduced image is 1.729, and the minimum value is −2 .607. Therefore, the difference between the maximum value obtained from the original image and the maximum value obtained from the reduced image is 0.983, and the difference between the minimum value obtained from the original image and the minimum value obtained from the reduced image is 1. 456, and a considerable difference is recognized.

  On the other hand, the upper limit value of the clip range obtained from the original image is 0.825, the lower limit value is −0.822, the upper limit value obtained from the reduced image is 0.742, and the lower limit value is −0.755. It is. Therefore, the difference between the upper limit value obtained from the original image and the upper limit value obtained from the reduced image is 0.083, and the difference between the lower limit value obtained from the original image and the lower limit value obtained from the reduced image is 0. 0. 067, and it can be seen that these differences are small.

  Next, the Retinex process of the original image, which is a post process performed based on the clip range obtained by the process shown in FIG. 3, will be described.

  FIG. 4 is a flowchart showing the original image Retinex processing. In this post-processing, each pixel of the original image is processed, and the processed pixel value is sequentially output to the printing unit 15.

  First, an original image mask is prepared (S20). The details of the original image mask preparation processing will be described later with reference to FIGS. Next, the reflectance R (x, y) is calculated for the original image using Equation 1 (S21). In this calculation, the filter F (x, y) is a mask for the original image stored in the mask memory 13h.

  Next, based on the clip range obtained using the reduced image, R (x, y) is normalized to obtain a normalized reflectivity refle (x, y) (S22).

  Next, using this refle (x, y), a pixel value Out (x, y) subjected to Retinex processing for luminance is obtained by the following equation (S23). Of the terms in Equation 5, I (x, y) / refle (x, y) corresponds to the normalized Retinex value.

次に、このＯｕｔ（ｘ，ｙ）と、色信号Ｃｂ、Ｃｒとに基づいて、ＲＧＢ値に変換する（Ｓ２４）。この変換は、次式により算出される。

Next, based on this Out (x, y) and the color signals Cb and Cr, conversion into RGB values is performed (S24). This conversion is calculated by the following equation.

以上の処理により元画像の輝度信号にレティネックス処理が施され、色信号に基づいてＲＧＢ値に戻した画素値が求められる。次に、この処理された画素値を印刷部１５に出力する（Ｓ２５）。よって、前処理により、正規化を行うクリップ範囲が特定されているので、後処理では、順次各画素について処理を行い印刷部１５に出力することができる。その結果、処理を開始してから、印刷が開始されるまでの時間を短縮することができる。

Through the above processing, the Retinex process is performed on the luminance signal of the original image, and the pixel value returned to the RGB value based on the color signal is obtained. Next, the processed pixel value is output to the printing unit 15 (S25). Therefore, since the clip range to be normalized is specified by the preprocessing, each pixel can be sequentially processed and output to the printing unit 15 in the postprocessing. As a result, the time from the start of processing to the start of printing can be shortened.

  Next, it is determined whether or not the process has been completed for all the pixels of the original image (S26). If there is a pixel that has not been processed yet (S26: No), the process returns to the process of S21 and all the pixels are processed. When the process is finished for (S26: Yes), the process is finished.

  Next, the process of determining the reduced image size and the reduction algorithm, which is the process of S3, will be described with reference to FIGS. The original image is reduced, the clip range is determined using the reduced image, and each pixel of the original image is corrected based on the clip range, so that the clip range can be determined quickly and the storage capacity is small. Can be processed. However, when the size of the reduced image is small or the image quality of the reduced image is greatly deteriorated, an appropriate clip range may not be set. Further, if the reduced image size is increased more than necessary or the image quality of the reduced image is increased more than necessary, the processing time takes longer and the processing speed decreases. Accordingly, it is necessary to select an appropriate reduced image size and reduction algorithm according to the print parameters set by the user.

  5 and 6 are look-up tables for setting a reduced image size and a reduction algorithm in accordance with the print mode, paper type, paper size, and original image size. FIG. FIG. 6 is for the normal mode. These tables are stored in the table memory 12c of the ROM 12, and are referred to in the process of the flowchart shown in FIG.

  FIG. 5 is a table selected when the print mode is the photo mode, and is first classified according to the type of paper. In this embodiment, the paper is classified into two types, glossy paper and plain paper, but in addition to these, the paper may be classified according to inkjet paper or the like.

  Next to the paper type, it is classified by the paper size. The paper sizes are classified into four types of 4 ″ × 6 ″, 5 ″ × 7 ″, letter, and A4 in order from the smallest paper size, and further classified according to the original image size for each paper size. The size of the original image is classified into three types of 600 × 800, 1200 × 1600, and 2400 × 3200 (pixels), and a reduced image size and a reduction algorithm are set for each size.

  In this embodiment, there are three types of reduced image sizes of 150 × 200, 300 × 400, and 450 × 600, and three reduction algorithms are the average pixel method (ME), the bilinear method (BL), and the nearest neighbor method (NN). One of the classes is set. FIG. 6 is a table that is selected when the print mode is the normal mode. Like the photo mode, the table is classified according to the paper type, paper size, and original image size, and the reduced image size and the reduction algorithm are set. The

  FIG. 7 is a flowchart showing details of the process of S3 in the flowchart shown in FIG. In this process, it is first determined whether the print mode selected as the print parameter is a photo mode or a normal mode (S31). If the print mode is the photo mode, a table for the photo mode is selected (S32). If the print mode is the normal mode, the table for the photo mode is selected (S33) and set as a print parameter. The set reduced image size and the reduction algorithm are read out from the set paper type, the paper size, and the original image size stored in the original image size memory 13d, and the reduced image size memory 13f and the reduction algorithm memory 13g are read out. (S34).

  Next, the original image mask preparation processing (S20) will be described with reference to FIGS.

  First, the size of the original image mask generated in the original image mask preparation process (S20) will be described with reference to FIG. FIG. 8A is a schematic diagram showing the relationship between the original image size and the reduced image size, and FIG. 8B shows the relationship between the size of the original image mask and the size of the reduced image mask. It is a schematic diagram. The “original image size” or “reduced image size” corresponds to the number of pixels constituting each image, and in the present embodiment, it is represented by the number of pixels in the horizontal direction × the number of pixels in the vertical direction.

  The “original image mask size” or “reduced image mask size” is the size of the calculation area when calculating the peripheral average luminance. In this embodiment, the horizontal direction included in the calculation area The size of each mask is expressed by the number of pixels × the number of pixels in the vertical direction. Here, the shape of the mask is not limited to a rectangle, but may be a shape other than a rectangle such as a circle. In this case, the size of a rectangular area inscribed in the mask is referred to as a mask size. .

  As shown in FIG. 8A, when the reduced image is the original image reduced by a reduction ratio of 1/2 (that is, the original image is divided into a plurality of blocks every 2 × 2 pixels, In the case of forming one pixel of the reduced image from the block), in the original image mask preparation process (S20) described later, the original image obtained by doubling the mask size (reciprocal of the reduction ratio) compared to the reduced image mask A mask is generated. That is, the original image mask is generated so that the ratio between the size of the reduced image and the size of the reduced image mask is the same as the ratio between the size of the original image and the size of the original image mask. In this way, since the calculation areas for calculating the peripheral average luminance coincide with each other, a preferable normalization parameter can be obtained even when the reflectance of the reduced image is used.

  As described above, each of the reduced image mask and the original image mask defines a calculation area when calculating the peripheral average luminance, and has a filter coefficient for all pixels included in the area, If the convolution operation is performed, the amount of calculation becomes enormous. Therefore, the reduced image mask and the original image mask divide the pixels in the calculation region determined by each into reference pixels and thinned pixels, and reflect only the value of the reference pixels in the peripheral average luminance, Reducing the amount of calculation has been done.

  The thinning pattern in the reduced image mask and the original image mask will be described with reference to FIGS. 9 to 11 are diagrams schematically showing a reduced image mask and an original image mask, in which elements (thinning points) obtained by thinning out filter coefficients are shown in white, and elements (reference points) having filter coefficients are shown. It is the figure shown in the gray color. The reduced image mask and the original image mask described with reference to FIGS. 9 to 11 are examples of masks applied when the original image is reduced to ¼ to form a reduced image. As described above, since the ratio of the mask size is equal to the ratio between the reduced image and the original image, the reduced image mask has a size (1/4) of the size of the original image mask (for example, 16 × 16 pixels) ( For example, 4 × 4 pixels).

  In the present embodiment, the leftmost and uppermost elements of the reduced image mask and the original image mask are (0, 0), the horizontal direction is the X-axis direction, and the vertical direction is the Y-axis direction. It shall be expressed in coordinates. For example, the second element in the X-axis direction and the third element in the Y-axis direction from (0, 0) is represented as (2, 3).

  First, the reduced image mask will be described with reference to FIG. By performing a convolution operation on the reduced image while shifting the reduced image mask, the reflectance is obtained for each pixel of the reduced image. When a reflectance of 1 is obtained, a reduced image mask having a size of 4 × 4 pixels is associated with 4 × 4 pixels of the reduced image, and a pixel corresponding to a reference point (hereinafter referred to as a reference pixel). The value is multiplied by the filter coefficient of the reference point, and the value is reflected in the reflectance. On the other hand, the values of pixels corresponding to the thinning points (hereinafter referred to as thinning pixels) are thinned out, and the values are not reflected when calculating the reflectance. In this way, arithmetic processing can be reduced by thinning out values of some pixels.

  Similarly, with respect to the original image mask, by performing a convolution operation on the original image while shifting the original image mask, the reflectance is obtained for each pixel of the original image. When obtaining a reflectance of 1, an original image mask having a size of 16 × 16 pixels is made to correspond to 16 × 16 pixels of the original image, and the reference pixel value corresponding to the reference point is referred to by the reference The point filter coefficient is multiplied, and the value of the thinned pixel corresponding to the thinned point is thinned.

  Here, in order to approximate the reflectance obtained from the reduced image and the reflectance obtained from the original image as much as possible, the reference point pattern in the original image mask is changed to the reference point pattern in the reduced image mask. It is assumed that it corresponds to.

  The original image mask shown in FIG. 9 is an original image mask generated when the average pixel method is selected as a reduction algorithm when forming a reduced image. When a reduced image is formed at a reduction ratio of 1/4, the original image is divided into a plurality of blocks for each 4 × 4 pixel, and one pixel of the reduced image is formed for each block. Here, when a reduced image is formed by the average pixel method, all 4 × 4 pixels constituting one block of the original image are extracted, and one pixel of the reduced image is extracted from the extracted pixel values. It is formed.

  Therefore, when the 4 × 4 pixels of the reduced image determined by the calculation area of the reduced image mask are in a relationship in which the 16 × 16 pixels of the original image determined by the calculation area of the original image filter are reduced, the reduced image The reference point of the mask for the original image is determined so that the 4 × 4 pixels of the original image extracted at the time of forming the reference pixel in FIG. In this way, the pixel value reflected in the reflectance of the reduced image approximates the pixel value reflected in the reflectance of the original image. As a result, a suitable normalization parameter can be obtained even when the reflectance of the reduced image is used, and the Retinex process can be performed with high accuracy.

  The original image mask shown in FIG. 10 is an original image mask when the bilinear method (BL method) is selected as the reduction algorithm. When a reduced image is formed at a reduction ratio of 1/4 by the bilinear method, the original image is divided into a plurality of blocks for every 4 × 4 pixels, and 4 pixels adjacent to each other are extracted from each block. One pixel of the reduced image is formed by interpolating the values.

  Therefore, when the 4 × 4 pixels of the reduced image determined by the calculation area of the reduced image mask are in a relationship in which the 16 × 16 pixels of the original image determined by the calculation area of the original image filter are reduced, the reduced image The reference point of the mask for the original image so that four pixels extracted at the time of forming the reduced image among the 4 × 4 pixels of the original image that are the source of the reference pixel in FIG. To decide. In this way, the pixel value reflected in the reflectance of the reduced image and the pixel value reflected in the reflectance of the original image can be approximated, which is preferable even when the reflectance of the reduced image is used. Normalization parameters can be obtained.

  The original image mask shown in FIG. 11 is an original image mask when the nearest neighbor method (NN method) is selected as the reduction algorithm. When a reduced image is formed at a reduction ratio of 1/4 by the nearest neighbor method, the original image is divided into a plurality of blocks every 4 × 4 pixels, and one pixel is extracted from each block, and the value of the one pixel is This is used as one pixel of the reduced image as it is.

  Therefore, the reference point of the mask for the original image is set so that one of the 4 × 4 pixels corresponding to the reference pixel in the reduced image becomes the reference pixel of the original image. decide. In this way, the pixel value reflected in the reflectance of the reduced image and the pixel value reflected in the reflectance of the original image can be approximated, which is preferable even when the reflectance of the reduced image is used. Normalization parameters can be obtained.

  The original image mask preparation process (S20) will be described with reference to FIG. FIG. 12 is a flowchart showing the original image mask preparation processing. This original image mask preparation process is a process for preparing an original image mask based on a reduction algorithm selected when a reduced image is formed.

  First, the reduction algorithm selected at the time of reduced image formation is read from the reduction algorithm memory 13g (see FIG. 1) (S201). Next, the reduction rate is read (S202). Here, the reduction ratio is a value corresponding to a ratio of one side length of the reduced image to one side length of the original image.

  Next, the size of the reduced image mask is read from the mask memory 13h (see FIG. 1) (S203). Then, the size of the original image mask is determined from the size of the reduced image mask and the reduction ratio (S204). Specifically, a value obtained by multiplying the size of the reduced image mask by the reciprocal of the reduction ratio is set as the size of the original image mask. In this way, the size of the original image mask is set so that the ratio between the size of the reduced image and the size of the reduced image filter is the same as the ratio between the size of the original image and the size of the original image mask. Can be determined (see FIG. 8).

  Next, the process branches based on the reduction algorithm read in the process of S201 (S205). When the reduction algorithm is the average pixel method (S205: average pixel method), the process proceeds to subroutine A (S206), and the reference point of the original image mask described with reference to FIG. 9 is determined. When the reduction algorithm is the bilinear method (S205: bilinear method), the process proceeds to subroutine B (S208) to determine the reference point of the original image mask described with reference to FIG. When the reduction algorithm is the nearest neighbor method (S205: nearest neighbor method), the process proceeds to subroutine C (S210), and the reference point of the original image mask described with reference to FIG. 11 is determined.

  Next, the generated original image mask is stored in the mask memory 13h (see FIG. 1) (S212), and the original image mask preparation process is terminated.

  The subroutine A (S206) will be described with reference to FIG. FIG. 13 is a flowchart showing the subroutine A. Subroutine A (S206) is a process executed when the average pixel method is selected as the reduction algorithm.

In the subroutine A, first, the size of the original image mask determined in the process of S204 (see FIG. 12) is read (S2061). Next, the magnification BR of the original image mask size with respect to the reduced image mask size is calculated (S2062). The magnification BR is a value calculated by the following equation (1) and rounded up after the decimal point.
BR = original image mask vertical size / reduced image mask vertical size (1)
Next, the reference point number M of the original image mask is calculated from the following equation (2) (S2063).
Number of mask reference points for original image M = number of mask reference points for reduced image m × (number of pixels of original image extracted when one pixel of reduced image is formed) (2)
When the reduced image is formed by the average pixel method, one pixel of the reduced pixel is formed from the ^two (BR) pixels of the original image. Therefore, in the subroutine A, (one pixel of the reduced image) in the expression (2) The number of pixels of the original image extracted for formation) is (BR) ² .

Next, the variable i is set to 1 (S2064), and the i-th reference point (Xsi, Ysi) in the reduced image mask is read (S2065). Then, a reference reference point MainPoint of the original image mask is obtained from the read reference points (Xsi, Ysi) (S2066). This calculation is represented by the following formula (3).
MainPoint (x) = Xsi · BR
MainPoint (y) = Ysi · BR (3)

Next, the counter CountX = 0 and the counter CountY = 0 are set (S2067). Then, the reference point (X _L , Y _L ) of the original image mask is obtained according to the following equation (4) (S2068).
X _L = MainPoint (x) + CountX
Y _L = MainPoint (y) + CountY (4)

Then, with respect to the obtained reference point (X _L , Y _L ), a filter coefficient is calculated by a Gaussian function (S2069) and stored in the RAM 13 (S2070).

  Next, “1” is added to the value of the counter CountX (S2071), and it is determined whether or not the counter CountX ≧ BR is satisfied (S2072). While the counter CountX is smaller than BR (S2072: No), the processing returns to S2068 and a point adjacent in the X-axis direction is obtained as a reference point.

  If the counter CountX = BR is satisfied while the processing is repeated in this way (S2072: Yes), then the counter CountX = 0 and “1” is added to the value of the counter CountY (S2073). It is determined whether or not the counter CountY ≧ BR is satisfied (S2074). While the counter CountY is smaller than BR (S2074: No), the processing returns to S2068 to calculate the next reference point.

If the counter CountY = BR is satisfied while the processing is repeated in this way (S2074: Yes), that is, ^two reference points of the original image mask BR are determined for one reference point of the reduced image mask. Then, “1” is added to the variable i (S2075), and it is determined whether or not the variable i is larger than the reference point number m of the reduced image mask (S2076). While the variable i is equal to or less than m (S2076: No), the process returns to S2065, and the next reference point of the reduced image mask is read.

  When the variable i becomes equal to m while repeating the process in this manner (S2076: Yes), that is, for all the reference points of the reduced image mask, the corresponding reference points of the original image mask are calculated. If so, the subroutine A is terminated. According to the subroutine A, as described with reference to FIG. 9, the reference point of the original image mask and its filter coefficient are determined.

  The subroutine B (S208) will be described with reference to FIG. FIG. 14 is a flowchart showing the subroutine B. Subroutine B (S208) is a process executed when a reduced image is formed by the bilinear method. Of the steps shown in the flowchart shown in FIG. 14, the same steps as those shown in the flowchart of FIG.

  In the subroutine B, first, the size of the original image mask is read (S2061), and the magnification BR of the original image mask size with respect to the reduced image mask size is calculated (S2062).

  Next, the reference point number M of the original image mask is calculated (S2063). When the reduced image is formed by the bilinear method, one pixel of the reduced image is formed from the four pixels of the original image. Therefore, the pixel of the original image extracted in the formation of one pixel of the reduced image in the above formula (2) “4” is assigned to (number).

  Next, the variable i is set to 1 (S2064), and the i-th reference point (Xsi, Ysi) in the reduced image mask is read (S2065). Then, a reference reference point MainPoint of the original image mask is obtained from the read reference points (Xsi, Ysi) (S2066).

Next, the counter CountX = 0 and the counter CountY = 0 are set (S2067). Then, the reference point (X _L , Y _L ) of the original image mask is calculated according to the equation (4) (S2068). Then, for the calculated reference point (X _L , Y _L ), a filter coefficient is calculated by a Gaussian function (S2069) and stored in the RAM 13 (S2070).

  Next, “1” is added to the value of the counter CountX (S2081), and it is determined whether or not the counter CountX ≧ 2 is satisfied (S2082). While the counter CountX is smaller than 2 (S2082: No), the process returns to S2068, and a point adjacent in the X-axis direction is calculated as a reference point.

  If the counter CountX = 2 is reached while the processing is repeated in this way (S2082: Yes), then the counter CountX = 0 and “1” is added to the value of the counter CountY (S2083). Then, it is determined whether or not the counter CountY ≧ 2 is satisfied (S2084). While the counter CountY is smaller than 2 (S2084: No), the processing returns to S2068 to calculate the next reference point.

  When the counter CountY = 2 is reached while repeating the processing in this way (S2084: Yes), that is, four reference points of the original image mask are determined with respect to one reference point of the reduced image mask. Then, “1” is added to the variable i (S2085), and the reference point of the corresponding original image mask is calculated until the variable i = m, that is, for all the reference points of the reduced image mask. The process is repeated until the subroutine B is completed. According to the subroutine B, as described with reference to FIG. 10, the reference point of the original image mask and its filter coefficient are determined.

  The subroutine C (S210) will be described with reference to FIG. FIG. 15 is a flowchart showing the subroutine C. Subroutine C (S210) is a process executed when a reduced image is formed by the nearest neighbor method. Of the steps shown in the flowchart shown in FIG. 15, the same steps as those shown in the flowchart of FIG.

  In the subroutine C, first, the size of the original image mask is read (S2061), and the magnification BR of the original image mask size with respect to the reduced image mask size is calculated (S2062).

  Next, the reference point number M of the original image mask is calculated (S2063). When a reduced image is formed by the nearest neighbor method, one pixel of the reduced image is formed from one pixel of the original image. Therefore, the original image extracted in (one pixel formation of the reduced image) in the above formula (2) is used. “1” is substituted into the number of pixels).

  Next, the variable i is set to 1 (S2064), and the i-th reference point (Xsi, Ysi) in the reduced image mask is read (S2065). Then, a reference reference point MainPoint of the original image mask is obtained from the read reference points (Xsi, Ysi) (S2066).

Next, the reference point (X _L , Y _L ) of the original image mask is obtained (S2108). Here, since the reference point of the original image mask corresponding to one reference point of the reduced image mask is only the reference reference point MainPoint, the reference reference point MainPoint calculated in the processing of S2066 is used as the reference point (XL _). , Y _L ).

Then, for the reference point (X _L , Y _L ), a filter coefficient is calculated by a Gaussian function (S2069) and stored in the RAM 13 (S2070).

  Next, “1” is added to the variable i (S2075), and the reference point of the corresponding original image mask is calculated until the variable i = m, that is, for all the reference points of the reduced image mask. The process is repeated until the subroutine C is completed. According to the subroutine C, as described with reference to FIG. 11, the reference point of the original image mask and its filter coefficient are determined.

  According to the original image mask prepared by the original image mask preparation processing of the present embodiment, as shown in FIGS. 9 to 11, the reference point pattern in the original image mask and the reference point in the reduced image mask Therefore, the reflectance distribution obtained from the reduced image is very close to the reflectance distribution obtained from the original image. As a result, an appropriate clip range (a clip range that is very close to the clip range obtained from the original image) can be obtained from the reduced image, and the Retinex process can be performed at high speed and with high accuracy.

  Note that the reduced image reflectance acquisition step and the reduced image reflectance acquisition means described in the claims correspond to the processing of S12 in the flowchart shown in FIG. 3, and the normalization parameter setting step and the normalization parameter setting means are the same as those in FIG. 4 corresponds to the process of S16 to S18 of the flowchart shown in FIG. 4, the original image reflectance acquisition step and the original image reflectance acquisition means correspond to the process of S21 of the flowchart shown in FIG. 4, and the correction step and the correction means correspond to FIG. 3, the reduced image forming step and the reduced image forming unit correspond to the process of S8 in the flowchart shown in FIG. 3, and the original image filter preparing step and the original image filter preparing unit are shown in FIG. 3 corresponds to the processing shown in the flowchart shown in FIG. 3, and the reduced image filter preparation step and the reduced image filter preparation means are shown in FIG. S4 processing of over chart are true.

  As described above, the present invention has been described based on the embodiments. However, the present invention is not limited to the above embodiments, and various improvements and modifications can be made without departing from the spirit of the present invention. It can be easily guessed.

  For example, in the original image mask preparation process (S20) of the above embodiment, the reference point of the original image mask is determined according to the reduction algorithm, but the reference point determination process is not limited to this.

  FIG. 16 is a flowchart showing a modification of the original image mask preparation process (S20). In the flowchart shown in FIG. 16, the same steps as those in the original image mask preparation process (S20) of the embodiment described with reference to FIG. According to the original image mask preparation process (S20) of the modification shown in FIG. 16, the reference point of the original image mask is determined by the subroutine C (S210), so that the original shown in FIG. An image mask is generated.

  According to the original image mask generated in this way, the p × q pixels of the reduced image determined by the calculation area of the reduced image mask are k · p of the original image determined by the calculation area of the original image mask. When the relationship of × k · q pixels is reduced, at least one reference pixel is included in each block of the original image that is the source of each reference pixel in the reduced image. Further, the pixels in each block of the original image that are the sources of the thinned pixels in the reduced image are thinned pixels.

  Therefore, the pixel value reflected in the reflectance of the reduced image approximates the pixel value reflected in the reflectance of the original image. As a result, a suitable normalization parameter can be obtained even when the reflectance of the reduced image is used, and the Retinex process can be performed with high accuracy. In addition, since the number of reference pixels can be reduced, the reflectance can be calculated at high speed.

  FIG. 17 is a diagram schematically showing a modification of the original image mask. As shown in FIG. 17, even when one pixel is randomly selected from the pixels in each block of the original image that is the source of the reference pixel in the reduced image and the pixel is used as the reference pixel, high-precision output The result can be obtained.

  The arrangement of the reference points in the reduced image masks shown in FIGS. 9 to 11 and 17 is arbitrary and can be changed as appropriate.

  In the above embodiment, the color expression system is the RGB format, but the present invention may be applied to other color expression systems other than the RGB format, for example, the CMY format.

  In the above embodiment, the image processing program of the present invention is executed by the CPU 11 incorporated in the printer 1, but is supplied as an application to the personal computer and executed by the CPU incorporated in the personal computer. You may make it do.

  The Retinex process may be SSR (single scale method) or MSR (multiscale method).

In the image processing of the above embodiment, the Retinex processing is performed by the CPU 11, but a DSP (Digital
(Signal Processor). When a DSP is used, processing such as product-sum operation can be executed at higher speed.

1 is a block diagram showing an electrical configuration of a printer equipped with an image processing program according to an embodiment of the present invention. It is a figure which shows the screen which sets a printing parameter. It is a flowchart which shows the pre-processing performed by an image processing program. It is a flowchart which shows the post-process performed following a pre-process. It is a photo mode table that is referred to when the print mode is a photo mode. It is a normal mode table that is referred to when the print mode is the normal mode. It is a flowchart which shows the process which refers to a table. (A) is a schematic diagram showing the relationship between the original image size and the reduced image size, and (b) is a schematic diagram showing the relationship between the size of the original image mask and the size of the reduced image mask. . It is a figure which shows typically the mask for reduced images and the mask for original images in case the average pixel method is selected as a reduction algorithm. It is a figure which shows typically the mask for reduction images, and the mask for original images in case the bilinear method is selected as a reduction | restoration algorithm. It is a figure which shows typically the mask for reduction images, and the mask for original images in case the nearest neighbor method is selected as a reduction | restoration algorithm. It is a flowchart which shows the mask preparation process for original images. It is a flowchart which shows the subroutine A. 3 is a flowchart showing a subroutine B. 3 is a flowchart showing a subroutine C. It is a flowchart which shows the mask preparation process for original images of a modification. It is a figure which shows typically the mask for reduced images and the mask for original images in a modification. It is a histogram figure for demonstrating the clip range for performing normalization. It is a figure which shows typically the mask for reduction images, and the mask for original images.

Explanation of symbols

1 Printer (image processing device)
2 Personal computer 11 CPU
12 ROM
12a Image processing program memory 13 RAM
13a Original image memory 13b Retinex image memory 13h Mask memory

Claims (10)

In an image processing program executed by an image processing apparatus that performs image correction processing,
A reduced image reflectance acquisition step for obtaining a peripheral average value for each pixel of the reduced image obtained by reducing the original image using a reduced image filter, and obtaining a reflectance from the peripheral average value and the value of each pixel of the reduced image; ,
A normalization parameter setting step for setting the parameter values when performing normalization by totalizing the frequency of each value in the entire range of values taken by the reflectance obtained by the reduced image reflectance acquisition step;
For each pixel of the original image, a peripheral average value is obtained using a filter for the original image, and an original image reflectance acquisition step for obtaining a reflectance from the peripheral average value and the pixel value of the original image;
A normalized Retinex value obtained by normalizing the reflectance acquired in the original image reflectance acquisition step based on the parameter value set in the normalization parameter setting step is obtained, and the normalized Retinex value and the original image A correction step for performing correction based on the pixel value,
An image processing program characterized in that a ratio between the size of the reduced image and the size of the reduced image filter is the same as a ratio between the size of the original image and the size of the original image filter. The reduced image filter defines a calculation area of the peripheral average value in the reduced image, and the pixels of p rows and q columns in the calculation area are referred to as a reference pixel whose value is reflected in the peripheral average value and a peripheral average value. It is divided into thinned pixels that do not reflect the value,
A reduced image forming step in which the original image is divided into a plurality of blocks for each pixel in k rows and k columns, and each block is aggregated into one pixel to form a reduced image;
Relationship in which pixels in p rows and q columns of the reduced image determined by the calculation region of the reduced image filter are reduced in pixels of k, p rows and k and q columns of the original image determined by the calculation region of the original image filter The original image filter for dividing the reference pixel and the thinned-out pixel is prepared so that at least one reference pixel is included in each block of the original image that is the source of each reference pixel in the reduced image. The image processing program according to claim 1, further comprising an original image filter preparation step. The original image filter preparation step includes:
Relationship in which pixels in p rows and q columns of the reduced image determined by the calculation region of the reduced image filter are reduced in pixels of k, p rows and k and q columns of the original image determined by the calculation region of the original image filter A filter for an original image for dividing a reference pixel and a thinned pixel so that a pixel in each block of the original image that is a source of each thinned pixel in the reduced image becomes a thinned pixel The image processing program according to claim 2, wherein: In the reduced image forming step, n pixels are extracted from each block (n is an integer equal to or greater than 1), one pixel is generated based on the extracted n pixels, and a reduced image is formed. Is,
The original image filter preparation step includes:
When the pixels of p rows and q columns of the reduced image determined by the reduced image filter are in a relation obtained by reducing the pixels of k, p rows and k and q columns of the original image determined by the original image filter, the reduction For an original image that classifies reference pixels and thinned pixels so that at least one of the n extracted pixels in each block of the original image that is the source of each reference pixel of the image is a reference pixel The image processing program according to claim 2, wherein a filter is prepared. A reduced image filter preparation step of preparing the reduced image filter;
5. The original image filter preparation step generates an original image filter based on the reduced image filter prepared in the reduced image filter preparation step. 5. An image processing program according to any one of the above. In an image processing apparatus that performs image correction processing,
Reduced image reflectance acquisition means for obtaining a peripheral average value for each pixel of the reduced image obtained by reducing the original image using a reduced image filter, and obtaining a reflectance from the peripheral average value and the value of each pixel of the reduced image; ,
A normalization parameter setting means for summing up the frequency of each value in the entire range of values taken by the reflectance obtained by the reduced image reflectance acquisition means, and setting a parameter value for normalization;
For each pixel of the original image, a peripheral average value is obtained using a filter for the original image, and an original image reflectance acquisition means for obtaining a reflectance from the peripheral average value and the pixel value of the original image;
A normalized Retinex value obtained by normalizing the reflectance acquired by the original image reflectance acquisition unit based on the parameter value set by the normalization parameter setting unit is obtained, and the normalized Retinex value and the original image Correction means for performing correction based on the pixel value,
The image processing apparatus according to claim 1, wherein a ratio between the size of the reduced image and the size of the reduced image filter is the same as a ratio between the size of the original image and the size of the original image filter. The reduced image filter defines a calculation area of the peripheral average value in the reduced image, and the pixels of p rows and q columns in the calculation area are referred to as a reference pixel whose value is reflected in the peripheral average value and a peripheral average value. It is divided into thinned pixels that do not reflect the value,
A reduced image forming unit that divides an original image into a plurality of blocks for each pixel of k rows and k columns and aggregates each block into one pixel to form a reduced image;
Relationship in which pixels in p rows and q columns of the reduced image determined by the calculation region of the reduced image filter are reduced in pixels of k, p rows and k and q columns of the original image determined by the calculation region of the original image filter The original image filter for dividing the reference pixel and the thinned-out pixel is prepared so that at least one reference pixel is included in each block of the original image that is the source of each reference pixel in the reduced image. The image processing apparatus according to claim 6, further comprising original image filter preparation means. The original image filter preparation means includes
Relationship in which pixels in p rows and q columns of the reduced image determined by the calculation region of the reduced image filter are reduced in pixels of k, p rows and k and q columns of the original image determined by the calculation region of the original image filter A filter for an original image for dividing a reference pixel and a thinned pixel so that a pixel in each block of the original image that is a source of each thinned pixel in the reduced image becomes a thinned pixel The image processing apparatus according to claim 7, wherein: The reduced image forming unit extracts n pixels from each block (n is an integer of 1 or more), generates one pixel based on the extracted n pixels, and forms a reduced image. Is,
The original image filter preparation means includes
When the pixels of p rows and q columns of the reduced image determined by the reduced image filter are in a relation obtained by reducing the pixels of k, p rows and k and q columns of the original image determined by the original image filter, the reduction For an original image that classifies reference pixels and thinned pixels so that at least one of the n extracted pixels in each block of the original image that is the source of each reference pixel of the image is a reference pixel The image processing apparatus according to claim 7, wherein a filter is prepared. A reduced image filter preparation means for preparing the reduced image filter;
10. The original image filter preparation means generates an original image filter based on the reduced image filter prepared by the reduced image filter preparation means. An image processing apparatus according to claim 1.

Images