JP4775289B2 - Image processing apparatus and image processing method - Google Patents

Description

  The present invention relates to an image processing apparatus and an image processing method for performing dodging processing on a digital photographic image taken with a digital camera or the like.

  In general, for color images of backlight scenes and strobe scenes shot on film by an optical camera, only the shadow part is corrected brightly while maintaining the gradation of the background, or only the highlight part is corrected darkly. Thus, there has been proposed an image processing apparatus that realizes dodging processing that simultaneously reproduces the gradations of both the main subject and the background by digital image processing.

  For example, Patent Document 1 discloses a luminance that creates a histogram indicating the relationship between luminance data and the frequency as an image processing device that corrects the contrast of an image based on image data for each different color of each pixel constituting the image. Luminance data belonging to one of the density sections is divided into the data calculation means and the density range of the brightness data divided into a first density section corresponding to the low brightness portion and a second density section corresponding to the high brightness portion based on the histogram. There has been proposed an image processing apparatus including a luminance data changing unit that changes only the brightness.

  However, an 8-bit image photographed with a digital camera or the like has an inherently narrow gradation width as compared with an image photographed on a film, and preferable correction characteristics can be obtained in the correction processing corresponding to the film image as described above. It was difficult.

  Therefore, as a technique for generating a brightness correction function capable of performing different brightness corrections corresponding to the shadow part and the highlight part on a color image taken with a digital camera or the like as a look-up table, Patent Document 2 As described above, it is conceivable to construct an image analysis unit such as a neural network.

  For example, various image characteristics of the sample image are input to the input layer, for example, the maximum luminance, minimum luminance, average luminance, etc. of a predetermined area such as the entire image, shadow portion, highlight portion, main subject portion, etc. as input teacher data, Learning for a large number of sample images so that the multiple characteristic values output from the output layer at that time match the output teacher data, which is the characteristic value after the sample image is appropriately corrected by manual operation or the like. The neural network obtained by doing is constructed.

The characteristic value output by such a neural network is configured to be a representative point constituting a luminance correction look-up table.
JP 2002-125130 A JP 09-168095 A

  However, when the representative points of the two look-up tables of the highlight part and the shadow part are obtained using the above-mentioned neural network, the number of outputs becomes very large, and a sufficient number of samples for learning cannot be prepared. Then, there is a problem that an overlearning state in which only a specific path among a plurality of paths from the input layer to the output layer is used, and appropriate data is not output.

  In addition, in the dodging process, linearity or the like is applied to the input / output distribution of the R (red), G (green), and B (blue) color components of the pixels constituting the image corrected by manual operation and the image before correction. There is also a problem that it is difficult to construct a look-up table for luminance correction using characteristic values output by a neural network because a constant correlation is not observed and the values tend to vary.

  The present invention has been made in view of the above-described conventional problems, and can properly capture the correlation between a plurality of characteristic values of a sample image and the characteristic values after the dodging correction is properly performed on the sample image. An object of the present invention is to provide an image processing apparatus and an image processing method capable of appropriately deriving and correcting a correction parameter for.

In order to achieve the above-mentioned object, the characteristic configuration of the image processing apparatus according to the present invention is as described in claim 1 of the document of the claims, the sample image corresponding to a plurality of characteristic values of the input sample image. The image analysis unit learned on the basis of a plurality of sample images so that correction coefficients for the highlight portion and the shadow portion that have been properly dodging corrected are output, and the plurality of characteristic values of the input target image A brightness conversion function generating unit that generates a brightness conversion function of the highlight part and the shadow part based on correction coefficients of the highlight part and the shadow part of the target image output from the image analysis unit, and the brightness conversion function e Bei luminance conversion processing unit for correcting processes the brightness of generator said target image based on the luminance conversion function generated by the luminance conversion function generation unit, the bright highlight portion There luminance conversion function conversion function and a shadow portion that generates a look-up table based on the following equation.

  According to the above-described configuration, the value output from the image analysis unit with respect to the plurality of characteristic values of the input target image is a single value that is the correction coefficient of the highlight portion and the shadow portion of the target image. Therefore, the image analysis unit can appropriately perform learning without preparing an extremely large number of sample images, and the luminance conversion function generation unit can perform the learning based on such correction coefficients. Thus, an appropriate luminance conversion function is generated for the highlight portion and the shadow portion.

Before Symbol image correction coefficient H _level of the highlight and shadow output from the analysis _unit, the highlight and shadow of the gamma value hγ the S _level as a variable, Sadamari is S [gamma], highlights based on the gamma value of the respective Luminance conversion function Hi for the area and luminance conversion function Si for the shadow area are obtained.

The second feature structure, as described in the claim 2, in addition to the first feature configuration described above, the image analysis unit comprises an input layer having a plurality of characteristic values of the target image is input, the intermediate layer And an output layer that outputs correction coefficients for the highlight portion and the shadow portion of the target image, and when the characteristic value of the sample image is input as input teacher data to the input layer, the sample image is appropriately The weighting coefficient of each layer is calculated by repeatedly calculating the teacher data obtained from a plurality of sample images so that the correction coefficient for the dodging corrected highlight part and shadow part is output from the output layer as output teacher data. It is composed of a neural network configured with a learning unit to learn, and the output layer is composed of two units, so it is appropriate even without an extreme number of samples Learning is to converge.

The third feature structure, as described in the claim 3, in addition to the first or second feature configuration described above, the luminance conversion processing unit, based on the luminance conversion function of the highlight and shadow Then, a value obtained by weighting each pixel luminance of the target image subjected to luminance conversion based on a value of a low frequency component of the target image is calculated as a corrected pixel luminance.

  According to the above-described configuration, the two luminance values of each pixel of the target image subjected to luminance conversion based on the luminance conversion functions of the highlight portion and the shadow portion are weighted by the luminance value of the original low frequency component of the target image. As a result, a new luminance is generated, so that the correction for the shadow portion is strongly reflected in the region where the luminance value of the low frequency component is low, and the correction for the highlight portion is strong in the region where the luminance value of the low frequency component is high. Will be reflected.

The first characteristic configuration of the image processing method according to the present invention is the highlight portion in which the sample image is appropriately dodging corrected with respect to the plurality of characteristic values of the input sample image as described in claim 4 And a learning step for learning the image analysis unit based on a plurality of sample images so that a correction coefficient for the shadow portion is output, and the characteristic value of the target image is input to the image analysis unit learned in the learning step. A correction coefficient output step for obtaining correction coefficients for the highlight and shadow portions of the target image, and the brightness of the highlight and shadow portions based on the correction coefficients for the highlight and shadow portions calculated in the correction coefficient output step. A luminance conversion function generation step for generating a conversion function; and the target based on the luminance conversion function generated by the luminance conversion function generation step. E Bei luminance conversion processing for correction of the image, the luminance conversion function generation step is a luminance conversion function of the luminance conversion function and shadow portions of the highlight portion to the point of generating a look-up table based on the following equation.

In the second feature configuration, as described in claim 5 , in addition to the first feature configuration described above, the brightness conversion processing step includes brightness conversion based on each brightness conversion function of the highlight portion and the shadow portion. A value obtained by weighting each pixel luminance of the target image based on a weighting factor based on a value of a low frequency component of the target image is calculated as a corrected pixel luminance.

  As described above, according to the present invention, the correlation between a plurality of characteristic values of a sample image and the characteristic values after the dodging correction of the sample image is properly captured, and correction parameters for the target image are appropriately derived. An image processing apparatus and an image processing method that can be corrected can be provided.

  Embodiments of a photographic image processing apparatus incorporating an image processing apparatus according to the present invention will be described below. As shown in FIG. 2, the photographic image processing apparatus 1 performs photographic paper 2 exposure processing based on output image data and develops exposed photographic paper, and developed photographic film. A media driver 32 that reads image data from an image data storage medium M such as a memory card that stores image data taken by a film scanner 31 or a digital still camera that reads an image from F, or a general-purpose computer as a controller 33 And an operation station 3 for setting and inputting print order information for a photographic image as an input original image and performing various image correction processes including image processing according to the present invention. The print data edited from the image is output to the photo printer 2. Desired photographic print is produced.

  As shown in FIGS. 2 and 3, the photographic printer 2 has two systems of photographic paper magazines 21 containing roll-shaped photographic paper P, and photographic paper P drawn from the photographic paper magazine 21 with a predetermined print. A sheet cutter 22 that cuts into a size; a back print unit 23 that prints print information such as a frame number on the back of the cut photographic paper P; an exposure unit 24 that exposes the photographic paper P based on the print data; A development processing unit 25 including a plurality of processing tanks 25a, 25b, and 25c filled with processing solutions for developing, bleaching, and fixing the exposed photographic paper P is disposed along the transport path of the photographic paper P. The laterally-feeding conveyor 26 that discharges the photographic paper P that has been dried after the development process, and the sheet-paper (photo print) P that is stacked on the laterally-feeding conveyor 26 is sorted in order units. Configured to include the data 27.

  The exposure unit 24 receives a laser beam bundle of RGB three colors modulated based on the print data in the main scanning direction orthogonal to the conveyance direction with respect to the photographic paper P conveyed in the sub-scanning direction by the conveyance mechanism 28. An exposure head 24a for outputting and exposing is accommodated.

  A transport mechanism 28 including a plurality of roller pairs that transport the photographic printing paper P at a process speed corresponding to the exposure unit 24 and the development processing unit 25 disposed along the transport path is disposed before and after the exposure unit 24. Is provided with a chucker-type transport mechanism 28a capable of transporting photographic paper P in multiple rows.

  The controller 33 provided in the operation station 3 is installed with an application program that operates under the control of a general-purpose operating system and executes various controls of the photographic processing apparatus 1, and a monitor 34 as an operation interface with the operator. A keyboard 35, a mouse 36, and the like are connected.

  The photographic processing process executed by the cooperation of the hardware and software of the controller 33 will be described in functional blocks. As shown in FIG. 4, photographic image data read by the film scanner 31 and the media driver 32 is received. An image input unit 40 that performs predetermined preprocessing and transfers it to a memory 41, which will be described later, and print order information and image editing information are displayed on the screen of the monitor 34, and an operation for inputting necessary data for them. A graphic user interface unit (hereinafter referred to as a “GUI unit”) that generates a graphic operation screen for displaying an icon for use or generates various control commands based on input operations from the keyboard 35 and mouse 36 to the displayed graphic operation screen. 42) and the picture. Photo image data transferred from the input unit 40, photo image data after correction processing by the image processing unit 47, which will be described later, correction parameters at that time, and set print order information are partitioned into predetermined areas and stored. A memory 41, an order processing unit 43 for generating print order information, an image processing unit 47 for performing density correction processing and contrast correction processing on each photographic image data stored in the memory 41, and the graphic user A display control unit 46 including a video RAM for displaying the image data and various input / output graphic data developed in the memory 41 based on a display command from the interface unit 42 on the monitor 34; Print for outputting the final corrected image after the correction processing to the photographic printer 2 A print data generating unit 44 that generates over data, and the like formatter 45 which converts the final corrected image according to customer orders to a file format for writing in a storage medium such as a CD-R.

  The film scanner 31 is configured to operate in two modes: a pre-scan mode that reads an image recorded on the film F at a high speed although it has a low resolution, and a main scan mode that reads a high-resolution image at a low speed. Various correction processes are performed in the pre-judge mode (described in detail in the pre-judge process described later) on the low resolution image read in the can mode, and the correction parameters stored in the memory 41 at that time are changed. Based on this, the final correction processing for the high-resolution image read in the main scan mode is executed and output to the printer 2.

  Similarly, the image file read from the media driver 32 includes a high-resolution captured image and its thumbnail image, and various correction processes described later are performed on the thumbnail image. Based on the stored correction parameters, a final correction process is performed on the high-resolution captured image. When the image file does not include a thumbnail image, the image input unit 40 generates a thumbnail image from the high-resolution captured image and transfers it to the memory 41. In this way, the calculation load of the controller 33 is reduced by executing various editing processes that are frequently trial and error on low-resolution images.

  As shown in FIG. 1, the image processing unit 47 includes an enlargement / reduction processing unit 11 that converts the input original image data into an image size suitable for the size of a photographic print, and a dodging process for the enlarged / reduced image. A dodging processing unit 10 that performs image processing, a density correction unit 12 that adjusts the density of the dodged image, a color correction processing unit 13 that corrects the color tone of the image, and a sharp image that enhances the edges of the image and reduces noise. The conversion processing unit 14 and the like are provided.

  In the image processing unit 47, first, each correction process is performed on a low-resolution image (for example, 256 × 384 pixels), and the correction parameter determined at that time, for example, in the dodging process, a highlight unit And the brightness conversion function of the shadow portion are stored in the memory 41, and then various correction processes are executed based on the parameters stored for the high-resolution image (for example, 2560 × 3840 pixels) to generate an output image. Then, the output image is output to the photographic printer 2 to generate a photographic print.

  Hereinafter, the dodging process for a digital image in which pixels are composed of 8 bits each of R (red), G (green), and B (blue) input from the media driver 32 will be described in detail.

  The dodging processing unit 10 outputs a plurality of sample images so that correction coefficients for a highlight portion and a shadow portion in which the sample image is appropriately dodging-corrected with respect to a plurality of characteristic values of the input sample image are output. Based on the correction coefficient of the highlight portion and the shadow portion of the target image output from the image analysis portion 10a with respect to the plurality of characteristic values of the input target image. A luminance conversion function generation unit 10b for generating a luminance conversion function for the highlight part and the shadow part, and a luminance conversion for correcting the luminance of the target image based on the luminance conversion function generated by the luminance conversion function generation part 10b. A color correction function using a gamma value calculated from the pixel luminance before and after the luminance conversion by the processing unit 10c and the luminance conversion processing unit 10c as a color correction variable. Zui in each pixel R (red), G (green), is configured to include a color correction processing unit 10d for correcting the B (blue) color components.

  The image analysis unit 10a includes an input layer including 89 neurons (I1, I2,..., I88, I89) to which a plurality of characteristic values of the sample image are input, and 45 neurons (M1, M1, M2). M2..., M45) and an output layer including two neurons (O1, O2), and a learning unit that learns the coupling coefficient of neurons between each layer based on a teacher signal It consists of a neural network with

  The learning unit is a highlight unit that is output from the output layer (O1, O2) when the characteristic value of the sample image is input as input teacher data to the input layer (I1, I2,..., I88, I89). And back-propagation using the steepest descent method so that the correction coefficients Hlevel and Slevel of the shadow part become the output teacher data, that is, the correction coefficient when the sample image is appropriately dodged by manual correction. To determine the coupling coefficient of neurons between each layer.

  Specifically, a characteristic value calculation unit for extracting characteristic values of each original image is provided for a plurality of sample images shot under various shooting conditions such as a backlight scene, a strobe scene, and a snowy mountain scene. 6 (a) to 6 (e), the average luminance, maximum luminance, and minimum luminance of the entire input image, and the average luminance, maximum luminance, and minimum luminance of the central portion and the peripheral portion where the main subject of the input image is located. , Average brightness and maximum brightness and minimum brightness of the left and right sides of the input image, average brightness and maximum brightness and minimum brightness of the upper and lower sides of the input image, and each block obtained by dividing the input image into blocks with a predetermined pixel size Average value of difference in average luminance of blocks adjacent to each other in the upper and lower regions, average luminance deviation of entire image, maximum value and minimum value of R (red), G (green), and B (blue) components of the entire image; Distribution of histograms for R (red), G (green), and B (blue) components Total 89 pieces of characteristic values consisting of (total count of pixel component contained in each area gradation shaft 20 equal portions of 0 to 255) is calculated as the input teacher data.

Here, the luminance is the luminance component of the YCC signal indicating the luminance color difference calculated based on [Equation 5 ] from the R (red), G (green), and B (blue) components of the pixels constituting the input image. Y means, for example, the average luminance of the entire input image indicates the average value of the luminance components of each pixel.

  The characteristic value calculation unit is further provided with a manual correction unit that performs dodging correction on these images by manual operation, and gradation conversion is performed on the highlight portion and the shadow portion by manual operation. The average gamma value of the highlight and shadow areas is calculated from the pixel brightness of the sample image that has been appropriately tone converted and the pixel brightness before correction, and the maximum and minimum widths of the average gamma value for multiple sample images are calculated. The normalized values are multiplied by a predetermined normalization coefficient to calculate the correction coefficients Hlevel and Level of the highlight part and the shadow part as output teacher data. As a result, correction coefficients Hlevel and Slevel are calculated as normalized values that take values between 0 and 100, respectively.

  The learning unit described above repeats learning thousands of times until the coupling coefficient converges based on these input teacher data and output teacher data.

Note that the manual correction unit may be mounted on the photo processing apparatus, or may be incorporated in an external image processing apparatus and the result may be input to the photo processing apparatus. The characteristic values input to the neural network are not limited to those described above, and average values of R (red), G (green), and B (blue) components are used instead of the luminance. May be used, or Cb and Cr data YCC-converted by [Equation 5 ] may be used instead of the data on each component of R (red), G (green), and B (blue). In any case, any correction coefficient Hlevel and Level may be learned so as to be appropriately separated and output with respect to an arbitrary characteristic value of the input image.

  That is, the image analysis unit 10a includes an input layer to which a plurality of characteristic values of the target image are input, an intermediate layer, and an output layer from which correction coefficients for the highlight and shadow portions of the target image are output. When the characteristic value of the sample image is input as input teacher data to the input layer, the correction coefficient for the highlight portion and the shadow portion in which the sample image is appropriately dodging corrected is output from the output layer as output teacher data. As shown in the figure, it is composed of a neural network that includes a learning unit that repeatedly calculates the teacher data obtained from a plurality of sample images and learns the weighting coefficient of each layer.

  When the characteristic value of the target image is input to the image analysis unit 10a that has been learned in this manner, appropriate correction coefficients Hlevel and Level for the highlight and shadow portions for the target image are output.

  Hereinafter, the flow of image processing by the dodging processor 10 will be described based on the flowchart shown in FIG.

  When a low-resolution target image is input (SA1), the above-described image characteristic value is calculated by the characteristic value calculation unit (SA2), and the image characteristic value is input to the image analysis unit 10a. Appropriate correction coefficients Hlevel and Level of the shadow part are output (SA3).

Next, the brightness conversion function Hi (i) and the brightness conversion function Si (i) of the shadow part are looked up by the brightness conversion function generation unit 10b based on [Expression 6 ] and [Expression 7 ]. (SA4). FIG. 7 shows an example of the luminance conversion function Hi (i) of the highlight portion and the luminance conversion function Si (i) of the shadow portion when hγ = sγ = 3.0. It is shown that the luminance is converted so that

In the luminance conversion processing unit 10c, a minimum value that can represent the maximum value and the minimum value for each of the R (red), G (green), and B (blue) components of each pixel constituting the input image ( 0) and the maximum value (255 for 8 bits, 4095 for 12 bits), the gradation expression range of each color component is extended and converted. For example, the R (red) component is converted by the conversion formula shown in [Expression 8 ] (SA5).

Further, the R (red), G (green), and B (blue) components R ′, G ′, and B ′ converted by [Equation 8 ] by the luminance conversion processing unit 10c according to [Equation 5 ]. A luminance value Y is calculated, and a luminance value image is generated (SA6).

Next, a dodging area, that is, a low-frequency image is generated from the generated luminance value image. Specifically, the luminance value image is subjected to bilateral filter processing, and a blurred image leaving an edge is generated as a low-frequency image represented by a new luminance value Y ′ (SA7). On the other hand, the luminance conversion processing of the highlight portion and the shadow portion is performed based on the look-up table representing the conversion characteristic by the luminance conversion function (SA8, SA9), and the luminance value of each pixel of the target image subjected to the luminance conversion is calculated. The luminance value Y ″ fused based on [Equation 9 ] is calculated as the corrected pixel luminance value (SA10).

  That is, based on the value of the low frequency component of the target image, in detail, the luminance width of the maximum luminance (here 255) and the minimum luminance (here 0) that each pixel luminance Y ′ of the low frequency image can take. A value obtained by weighting based on the normalized weight coefficient W is calculated as the corrected pixel luminance.

Subsequently, in the color correction processing unit 10d, the gamma value calculated by [Equation 10 ] from the pixel luminance before and after the luminance conversion by the luminance conversion processing unit 10c is obtained as a color correction variable of each pixel (SA11).

Further, using the calculated gamma value cγ as a variable, a color correction function shown in [Equation 11 ] is generated as a lookup table (SA12), and R (red), G (green), converted by [Equation 8 ], The B (blue) components R ′, G ′, B ′ are calculated as new component values R ″, G ″, B ″ based on [Equation 11 ] (SA14).

  FIG. 8 shows an example of such a color correction function, and the R (red), G (green), and B (blue) color components of the target image with respect to pixels whose luminance is converted to be low by the luminance conversion processing unit 10c. The brightness difference between the R (red), G (green), and B (blue) color components of the target image is corrected with respect to the pixel whose brightness is converted to be high by the brightness conversion processing unit 10c. It will be corrected to spread.

In the color correction function, the brightness difference between the R (red), G (green), and B (blue) color components of the target image is narrowed with respect to the pixel whose luminance is converted to be low by the luminance conversion processing unit 10c. A function that corrects the brightness difference of the R (red), G (green), and B (blue) color components of the target image with respect to a pixel whose luminance is converted to be high by the luminance conversion processing unit 10c. If there is, it is not limited to the one shown in [Equation 11 ].

  In this way, the luminance conversion function Hi (i) of the highlight portion and the luminance conversion function Si (i) of the shadow portion, which have been subjected to the dodging process, are used as correction parameters for all the input low-resolution images. It is stored in the memory 41 (SA15), and parameters for correction processing in the density correction unit 12, the color correction processing unit 13, the sharpening processing unit 14 and the like are stored in the memory 41, and then based on the parameters. Image processing is performed on the high-resolution target image (SA16), and a photographic print is generated.

  Note that in step SA7 described above, in order to speed up the arithmetic processing, the input pixels are blocked by the number of pixels of a predetermined size, and the mosaic is obtained by adding and averaging the luminance values of the pixels included in the block as new luminance values. A reduced image is generated by processing, a bilateral filter process is performed on the reduced image to generate a low-frequency image, and the above-described steps SA8 to SA10 are performed on the low-frequency image, and then the original image is enlarged. It may be a thing.

FIG. 10 (b) shows an image obtained by correcting the original image (a) by the dodging process according to the present invention, and FIG. 10 (c) shows the brightness conversion by the process up to step SA10 in the dodging process according to the present invention. This is an image of R, G, B components calculated by inverse transformation of [Equation 5 ] after multiplying the conversion ratio Y ″ / Y of Y ″ of the pixel by the Cb, Cr component. For example, referring to the skirt portion of the model, in the case of FIG. 10B, the color saturation as shown in FIG. 10C is reduced and the color reproducibility is excellent.

  Hereinafter, another embodiment will be described. In the above-described embodiment, the image analysis unit is configured using a neural network. However, the image analysis unit is not limited to the one configured using a neural network. It is also possible to configure with an arithmetic unit that performs multiple regression analysis, which is an example of analysis, and it is also possible to configure using a support vector machine or the like.

  When the automatic density correction is performed after the dodging process, the shadow area and the highlight part as shown in FIG. 7 are surrounded by the look-up table so that the effect of the dodging process is not changed by the automatic density correction. It is only necessary to correct sγ and hγ so that the area does not change and the target correction value by the automatic density correction does not change due to the dodging process.

In the embodiment described above, R (red), G (green), and B () of each pixel based on a color correction function using a gamma value calculated from pixel luminance before and after luminance conversion by the luminance conversion processing unit as a color correction variable. (Blue) Although the color correction processing unit for correcting the color component is described to constitute the dodging processing unit, the dodging processing unit according to the present invention does not necessarily include the above-described color correction processing unit, and luminance conversion is performed. Even if the conversion ratio Y ″ / Y of the pixel Y ″ is multiplied by the Cb and Cr components, the R, G and B components may be calculated by the inverse transformation of [Equation 5 ]. Good.

  In the above-described embodiment, the image processing apparatus and the image processing method according to the present invention are applied to the photographic image processing apparatus. However, the present invention is not limited to those incorporated in such a dedicated machine. The image processing method according to the present invention can be configured by a personal computer, a digital camera, or the like installed with an application program.

  An application program is a computer that outputs a plurality of samples so that correction coefficients for a highlight portion and a shadow portion obtained by properly covering and correcting the sample image with respect to a plurality of characteristic values of the input sample image are output. Based on the correction coefficients of the highlight portion and the shadow portion of the target image output from the image analysis portion for the plurality of characteristic values of the input target image, and the image analysis portion learned based on the image, A luminance conversion function generation unit that generates a luminance conversion function of the highlight unit and the shadow unit, and a luminance conversion processing unit that corrects the luminance of the target image based on the luminance conversion function generated by the luminance conversion function generation unit This is an image processing program to function.

  Note that the above-described embodiment is merely an example of the present invention, and it is needless to say that the specific configuration and the like of each block can be changed and designed as appropriate within the scope of the effects of the present invention.

Functional block diagram of the image processing unit Explanatory drawing of the external configuration of the photographic image processing apparatus Illustration of photo printer Functional block diagram of photographic image processing device Illustration of the neural network that composes the image analysis unit Area explanatory diagram of target image for which image characteristic value is calculated Characteristic diagram of brightness conversion function for highlight and shadow areas Characteristic of color correction function Flow chart showing dodging process Explanatory drawing of a photographic image subjected to dodging processing according to the present invention (a color print of this drawing is submitted by the property submission form)

1: Photo image processing apparatus 10: Dodge processing unit 10a: Image analysis unit 10b: Luminance conversion function generation unit 10c: Luminance conversion processing unit 10d: Color correction processing unit 47: Image processing unit

Claims (5)

An image learned based on a plurality of sample images so that correction coefficients for a highlight portion and a shadow portion in which the sample image is appropriately dodging corrected with respect to a plurality of characteristic values of the input sample image are output. An analysis unit;
Generates brightness conversion functions for the highlight and shadow parts based on the correction coefficients for the highlight and shadow parts of the target image output from the image analysis unit for a plurality of characteristic values of the input target image A luminance conversion function generator to
E Bei luminance conversion processing unit for correcting processing the luminance of the target image based on the luminance conversion function generated by the luminance conversion function generation unit,
The luminance conversion function generation unit is an image processing apparatus that generates a luminance conversion function of a highlight unit and a luminance conversion function of a shadow unit as a lookup table based on the following formula.

The image analysis unit includes an input layer to which a plurality of characteristic values of the target image are input, an intermediate layer, and an output layer from which correction coefficients for the highlight and shadow portions of the target image are output, and the characteristics of the sample image When a value is input to the input layer as input teacher data, a correction coefficient for a highlight portion and a shadow portion in which the sample image is properly dodging corrected is output from the output layer as output teacher data. repeatedly calculated by the learning unit image processing apparatus according to claim 1 Symbol mounting is composed of a configured neural network comprises a learning weighting coefficient of each layer with respect to the teacher data obtained from a plurality of sample images. The luminance conversion processing unit corrects a value obtained by weighting each pixel luminance of the target image obtained by luminance conversion based on each luminance conversion function of the highlight unit and the shadow unit based on a value of a low frequency component of the target image. the image processing apparatus according to claim 1 or 2 Symbol placement is calculated as the pixel brightness. Based on the plurality of sample images, the image analysis unit is configured to output correction coefficients for the highlight portion and the shadow portion in which the sample image is appropriately dodging-corrected with respect to the plurality of characteristic values of the input sample image. Learning steps to learn,
A correction coefficient output step of inputting the characteristic value of the target image into the image analysis unit learned in the learning step to obtain the correction coefficient of the highlight part and the shadow part of the target image;
A brightness conversion function generating step for generating a brightness conversion function for the highlight part and the shadow part based on the correction coefficient for the highlight part and the shadow part calculated in the correction coefficient output step;
E Bei luminance conversion processing step of correcting processing said target image based on the luminance conversion function generated by the luminance conversion function generation step,
The luminance conversion function generating step is an image processing method for generating a luminance conversion function of a highlight portion and a luminance conversion function of a shadow portion as a lookup table based on the following formula.

In the luminance conversion processing step, each pixel luminance of the target image obtained by performing luminance conversion based on each luminance conversion function of the highlight portion and the shadow portion is calculated based on a weighting factor based on a value of a low frequency component of the target image. the image processing method according to claim 4 Symbol mounting calculates a weighting calculation value as the corrected pixel brightness.