JP5771430B2 - Wrinkle removal image processing apparatus, wrinkle removal image processing method, and wrinkle removal image processing program - Google Patents

Description

  The present invention relates to a wrinkle removal image processing apparatus, a wrinkle removal image processing method, and a wrinkle removal image processing program for removing wrinkles from an image.

  When taking a landscape photograph, it often appears that a mountain is far away. Thus, Non-Patent Document 1, for example, has been published as a technique for estimating the density of soot and removing the soot. In this paper, the flaws are removed by the following procedure. First, the input image is corrected by dividing RGB of each pixel of the input image by RGB of the atmospheric light. For the corrected input image, the density of wrinkles is estimated using the minimum RGB values in a local region called a dark channel. From this hull density, the transmittance with coarse accuracy is obtained. Furthermore, a fine accuracy transmittance is obtained by a technique called soft matting. Using this transmittance, a process for removing wrinkles from the image is performed.

K. He, J. Sun and X. Tang, “Single Image Haze Removal Using Dark Channel Prior” CVPR, 2009

  The local region used for obtaining the dark channel is desirably a region in which the transmittance is regarded as substantially constant. However, in the method of Non-Patent Document 1, since a square having a predetermined size centered on a pixel for which a dark channel is to be obtained is always used as the local region, the transmittance differs within the local region. A case occurs. Therefore, the estimated transmittance will be of coarse accuracy. Therefore, soft matting is performed in order to improve the transmittance accuracy. However, soft matting requires a huge amount of calculation, and there is a problem that it takes too much time for practical use.

  The present invention has been made in view of the above points, and an object thereof is to provide a wrinkle removal image processing device, a wrinkle removal image processing method, and a wrinkle removal image processing program that remove wrinkles in an image at high speed and with high accuracy. And

One aspect of the image processing apparatus of the present invention, the atmospheric exposure compensation unit that corrects the pixel value of the atmospheric optical input image, the target pixel of the input image, such as a pixel of interest unrealized transmittance considered to be constant A region calculation unit for calculating a local region, a minimum pixel value selection unit for selecting a minimum pixel value in the local region in the input image corrected by the pixel value of the atmospheric light, and transmission of the target pixel from the minimum pixel value And a transmittance calculating unit that calculates the rate.
Another aspect of the image processing apparatus of the present invention calculates the similarity between the atmospheric light correction unit that corrects the input image with the pixel value of the atmospheric light, and the target pixel of the input image and the peripheral pixels of the target pixel, said similarity area calculation unit that the pixel of interest unrealized transmittance to calculate the local region that considered to be constant with the minimum pixel value in the local region of the input image corrected by the pixel value of airglow A minimum pixel value selection unit to select, a transmittance calculation unit to calculate the transmittance of the target pixel from the minimum pixel value, and a wrinkle removal image generation unit to remove the wrinkle component from the input image using the transmittance It is characterized by.

Another embodiment of the image processing method of the present invention includes the steps of correcting an input image pixel values of atmospheric light, such as the target pixel force image, the target pixel unrealized transmittance considered to be constant local A step of calculating a region, a step of selecting a minimum pixel value in the local region in the input image corrected by the pixel value of light, a step of calculating the transmittance of the target pixel from the minimum pixel value, and transmission Removing the wrinkle component from the input image using the rate.

Another embodiment of the image processing program of the present invention includes the steps of correcting an input image pixel values of atmospheric light, such as the target pixel of the input image, a pixel of interest unrealized transmittance considered to be constant local A step of calculating a region, a step of selecting a minimum pixel value in the local region in the input image corrected by the pixel value of the atmospheric light, a step of calculating a transmittance of the target pixel from the minimum pixel value, and a transmittance Removing the wrinkle component from the input image using a computer, and causing the computer to exhibit.

  According to the present invention, a wrinkle removal image processing apparatus that removes wrinkles in an image at high speed and with high accuracy by obtaining a region used for obtaining a dark channel for each pixel for which a dark channel is to be obtained, and flaw removal An image processing method and a wrinkle removal image processing program can be provided.

1 is a configuration diagram of a wrinkle removal image processing apparatus according to a first embodiment. The graph which shows the relationship between a dark channel and the transmittance | permeability. The flowchart of the area | region determination by an area | region calculation part. The figure which shows the example of the conversion function of similarity calculation. The block diagram of the wrinkle removal image processing apparatus which concerns on 2nd Embodiment. The block diagram of an area | region calculation part. The block diagram of the wrinkle removal image processing apparatus which concerns on 3rd Embodiment. The figure which shows an example of the candidate area | region in the wrinkle removal image processing apparatus which concerns on 4th Embodiment.

Hereinafter, embodiments for carrying out the present invention will be described with reference to the drawings.
[First Embodiment]
As shown in FIG. 1, the wrinkle removal image processing apparatus 100 according to the first embodiment of the present invention includes an atmospheric light pixel value calculation unit 200, an atmospheric light correction unit 300, an area calculation unit 400, and a minimum pixel value calculation. Unit 500, transmittance calculation unit 600, and wrinkle removal image generation unit 700. The atmospheric light pixel value calculation unit 200 is connected to the atmospheric light correction unit 300 and the haze removal image generation unit 700. The atmospheric light correction unit 300 is connected to the minimum pixel value calculation unit 500. The area calculation unit 400 is connected to the minimum pixel value calculation unit 500. The minimum pixel value calculation unit 500 is connected to the transmittance calculation unit 600. The transmittance calculation unit 600 is connected to the wrinkle removal image generation unit 700.

  Note that the soot is a phenomenon in which a distant subject cannot be clearly seen due to particles in the atmosphere. Also included is a phenomenon in which the subject is difficult to see due to weather phenomena such as fog, haze, and haze.

  When photographing a subject at a distant location, the subject may be white due to the influence of atmospheric light. Here, atmospheric light refers to light scattered by particles in the atmosphere. Among the pixel values of the image, the component obtained by the scattered light of the particles is set as the atmospheric light pixel value. When the input image is input, the atmospheric light pixel value calculation unit 200 calculates and outputs the atmospheric light pixel value. The calculation method of the atmospheric light pixel value is calculated by the method described in Non-Patent Document 1. In the present embodiment, the atmospheric light pixel value calculation unit 200 calculates one atmospheric light pixel value for the entire input image. In Non-Patent Document 1, the atmospheric light pixel value is calculated using the dark channel of the input image. However, the atmospheric light pixel value is obtained using the dark channel using the area obtained by the area calculating unit 400. Also good.

  Note that the method of obtaining the atmospheric light pixel value is not limited to this method of obtaining from the input image. For example, the user may set the atmospheric light pixel value.

  The atmospheric light correction unit 300 inputs the input image and the atmospheric light pixel value calculated by the atmospheric light pixel value calculation unit 200. An image having a pixel value obtained by dividing each pixel value of the input image by the atmospheric light pixel value is output. Division is performed for each channel. For example, when the image is an RGB image composed of three channels of R, G, and B, in each pixel of the input image, R of each pixel is R of atmospheric light, G of each pixel is G of atmospheric light, and each pixel Are divided by B of atmospheric light. When the input image is a Y image composed of one channel of luminance, Y of each pixel of the input image is divided by Y of atmospheric light. The output image of the atmospheric light correction unit 300 is hereinafter referred to as an atmospheric light correction image.

  The area calculation unit 400 receives an input image, calculates an area for obtaining a dark channel of a target pixel of the input image, and outputs information representing the calculated area. Details of the area calculation will be described later.

  Here, the dark channel of the target pixel refers to the minimum value of all channel values of all the pixels in the local region including the target pixel. In the case of an RGB image composed of three channels of R, G, and B, the dark channel of the target pixel is the minimum value of R, G, and B of all the pixels in the local region including the target pixel. In the case of a Y image composed of one luminance channel, the dark channel of the target pixel is the minimum Y value of all the pixels in the local region including that pixel. Note that the input image may be an image configured by a channel different from that of the RGB image or the Y image.

  The region for obtaining the dark channel is preferably a region where the transmittance is regarded as substantially constant. Therefore, for each pixel, an area where the transmittance can be regarded as almost constant is determined as an appropriate area for obtaining the dark channel. The transmittance will be described later.

  The minimum pixel value calculation unit 500 receives the atmospheric light correction image output from the atmospheric light correction unit 300 and information representing a region for obtaining a dark channel calculated by the region calculation unit 400. Then, for each pixel of the atmospheric light correction image, a dark channel is obtained and output based on the local area indicated by the information calculated by the area calculation unit 400.

The transmittance can be calculated from a functional equation with the dark channel. The transmittance calculation unit 600 receives the dark channel calculated for each pixel output from the minimum pixel value calculation unit 500, and obtains the transmittance for each pixel using the following equation (1).
Trn = 1−ω × Drk (1)
Here, Drk is a dark channel, Trn is transmittance, and ω is a parameter given in advance to control the strength of wrinkle removal.

In order to avoid an excessively low transmittance that tends to cause noise and unnaturalness in the image after removal of wrinkles, it can be replaced with Trn0 if the transmittance Trn is smaller than a predetermined threshold value Trn0. That means
Trn = max (Trn, Trn0) (2)
Here, max (a, b) represents the larger of a and b. FIG. 2 is a graph showing the relationship between the transmittance obtained in this way and the dark channel. The transmittance calculation unit 600 outputs the calculated transmittance.

  The wrinkle removal image generation unit 700 receives the input image, the atmospheric light pixel value calculated by the atmospheric light pixel value calculation unit 200, and the transmittance calculated by the transmittance calculation unit 600. By calculating 3), an image from which wrinkles are removed is obtained.

ここで、入力画像の画素値と大気光の画素値は、ｃ個のチャネルから成っているとする。また、Ａｉは大気光のｉ番目のチャネルの値、Ｉｉ（ｐ）は入力画像の座標ｐの画素のｉ番目のチャネルの値、Ｊｉ（ｐ）は霞を除去した画像の座標ｐの画素のｉ番目のチャネルの値とする。

Here, it is assumed that the pixel value of the input image and the pixel value of the atmospheric light are composed of c channels. Ai is the value of the i-th channel of atmospheric light, Ii (p) is the value of the i-th channel of the pixel at the coordinate p of the input image, and Ji (p) is the value of the pixel at the coordinate p of the image from which wrinkles are removed. The value of the i-th channel is assumed.

  Next, a region determination method by the region calculation unit 400 will be described with reference to the flowchart of FIG. In the present embodiment, a method for selecting a region for obtaining a dark channel of a target pixel from among a plurality of candidate regions including the target pixel will be described. In this embodiment, a case where the number of pixels in all candidate areas is not the same will be described.

  In step S301, the number of pixels in the candidate area is set to N (1), N (2),..., N (m) in order from the largest. Here, the ranks 1, 2,..., M of the number of pixels in the candidate area are referred to as pixel ranks. For example, if N (1) = 64, as an example of a candidate area with the number of pixels N (1), if the pixel of interest is an upper right vertex in an 8 × 8 square, the position of the pixel of interest is lower right In the case of a vertex, when the pixel of interest is at the lower left vertex, the pixel of interest is at the lower left vertex, the pixel of interest is at the center, and there are five ways, and a 4 × 16 rectangle of the pixel of interest. When the position is the upper right vertex, when the target pixel position is the lower right vertex, when the target pixel position is the lower left vertex, when the target pixel position is the lower left vertex, or when the target pixel position is the center 10 types of candidate areas can be listed. Here, the shape of the candidate area may be other than a square, and the position of the pixel of interest may be anywhere within the area, and any number of candidate areas having the same number of pixels may be used. Let the pixel number order be i. In step S302, the pixel number order i is set to 1. In step S303, the similarity between all candidate regions of the selected number of pixels N (i) and the target pixel is obtained for each candidate region. The similarity between the candidate region Ω and the target pixel p is calculated as follows.

ここで♯Ωは領域Ωに含まれる画素数、Ｆは予め定められた関数とする。

Here, # Ω is the number of pixels included in the region Ω, and F is a predetermined function.

  This is not necessarily the case, for example

でもよい。ここで、ｗｉは、第ｉチャネルの重みとする。
また、

But you can. Here, wi is a weight of the i-th channel.
Also,

でもよい。
変換関数Ｆとしては、例えば、傾きが０以下の１次関数でもよいし、傾きが０以下の複数の直線で構成される折れ線関数でもよいし、入力がある閾値より小さい場合は１、そうでない場合は０を出力する関数でもよい。図４に（ａ）１次関数の場合、（ｂ）折れ線関数の場合、（ｃ）閾値による関数を示す。また、ガウス関数でもよいし、予め定められたルックアップテーブルでもよい。

But you can.
The conversion function F may be, for example, a linear function with a slope of 0 or less, a polyline function composed of a plurality of straight lines with a slope of 0 or less, or 1 if the input is smaller than a certain threshold value. In this case, a function that outputs 0 may be used. FIG. 4 shows (a) a linear function, (b) a broken line function, and (c) a function based on a threshold value. Further, it may be a Gaussian function or a predetermined lookup table.

  In step S304, the maximum similarity Smax is searched to obtain a candidate region having the maximum similarity. In step S305, it is determined whether the pixel number order i has reached the last pixel number order m. When the final pixel number rank m is reached, a candidate region having the maximum similarity is determined as an appropriate region for obtaining the dark channel of the target pixel in step S306. If the final pixel number rank m has not been reached, it is determined in step S307 whether the maximum similarity is smaller than a predetermined threshold value U. If the maximum similarity is not smaller than the threshold value, a candidate region having this maximum similarity is determined as an appropriate region for obtaining the dark channel of the target pixel in step S306. If the maximum similarity is smaller than the threshold value, the pixel number rank i is increased by one in step S308, and the same processing is performed. Hereinafter, such a process is repeated until an appropriate region for determining the dark channel of the target pixel is determined.

Then, after determining the region, the region calculation unit 400 outputs the determined region information.
As described above, the region calculation unit 400 determines an optimal region for obtaining the dark channel for each pixel of interest. Therefore, in the present embodiment, it is possible to estimate the transmittance with fine accuracy from the dark channel without using soft matting.

  If the number of pixels in the area for obtaining the dark channel of the target pixel is too small, there is a high possibility that the dark channel will not correspond to the density of wrinkles. On the other hand, if the number of pixels in the area for obtaining the dark channel is too large, there is a high possibility that the transmittance in the area cannot be regarded as almost constant. In the present embodiment, a search is performed from a candidate area having a large number of pixels, and if it is not sufficiently similar to the target pixel, a search is performed from an area having a smaller number of pixels, so that the area is as large as possible. A region where the transmittance can be regarded as almost constant can be set as a region suitable for obtaining the dark channel.

Further, since an appropriate area is selected from the candidate areas, the amount of calculation is small.
In the present embodiment, the case where the number of pixels in all candidate areas is not the same has been described. However, the number of pixels in all candidate areas may be the same. In this case, the area calculation unit 400 sets the number of all candidate areas to N (1) (step 301). All candidate areas are selected (step S302). The similarity between all the selected candidate areas and the target pixel is obtained from the above-described equation (step S303). The maximum similarity Smax is obtained (step S304). Then, the candidate area having the maximum similarity is determined as an area for obtaining the dark channel of the target pixel (step S306). That is, in the flowchart of FIG. 3, Step S305, Step S307, and Step S308 are omitted.

[Second Embodiment]
FIG. 5 is a configuration diagram of a wrinkle removal image processing apparatus according to the second embodiment of the present invention. The area calculation unit 400 of the first embodiment is changed to an area calculation unit 800, and the other parts have the same configuration as that of the first embodiment. A description of the same parts as those in the first embodiment will be omitted, and the region calculation unit 800 will be described here.

  FIG. 6 is a configuration diagram of the area calculation unit 800. The similarity calculation unit 801 obtains the similarity between the pixel of interest and pixels in a predetermined range around it. Here, the predetermined range is a region larger than the range for obtaining the dark channel. The similarity is obtained, for example, by converting the sum of the absolute values of the differences between the values of the two pixels for each channel using a predetermined function. That means

ここで、Ｓ（ｐ、ｑ）は、座標ｐの注目画素とその周囲の定められた範囲内の座標ｑの画素との類似度としＩ_ｉ（ｐ）は座標ｐの画素の第ｉチャネルの値、ｃはチャネル数、Ｆは予め定められた関数とする。

Here, S (p, q) is the similarity between the pixel of interest at the coordinate p and the pixel at the coordinate q within a predetermined range around it, and I _i (p) is the i-th channel of the pixel at the coordinate p. The value, c is the number of channels, and F is a predetermined function.

  Also, a weighted sum obtained by multiplying the absolute value of the difference between the values of the two pixels for each channel by a predetermined weight for each channel may be used. That means

ここでｗｉは、第ｉチャネルの重みとする。
さらに、二つの画素のチャネル毎の値の差の絶対値の和の代わりに、二つの画素のチャネル毎の値の差の二乗の和を使ってもよい。つまり、

Here, wi is a weight of the i-th channel.
Furthermore, instead of the sum of the absolute values of the differences between the two pixel channels, the sum of the squares of the difference between the two pixel values may be used. That means

変換関数Ｆとしては、例えば、傾きが０以下の１次関数でもよいし、傾きが０以下の複数の直線で構成される折れ線関数でもよいし、入力がある閾値より小さい場合は１、そうでない場合は０を出力する関数でもよい。

The conversion function F may be, for example, a linear function with a slope of 0 or less, a polyline function composed of a plurality of straight lines with a slope of 0 or less, or 1 if the input is smaller than a certain threshold value. In this case, a function that outputs 0 may be used.

  Next, the center-of-gravity coordinate calculation unit 802 calculates the center-of-gravity coordinate r with the similarity calculated by the similarity calculation unit 801 as a weight as follows.

ここで、Ωは座標ｐの周囲の定められた範囲を示す。
共分散行列算出部８０３は類似度を重みにした重み付け共分散行列Ｍを次のようにして求める。

Here, Ω represents a predetermined range around the coordinate p.
The covariance matrix calculation unit 803 obtains a weighted covariance matrix M with the similarity as a weight as follows.

ここでｑｘ、ｒｘは、座標ｑと座標ｒのｘ座標、ｑｙ、ｒｙは、座標ｑと座標ｒのｙ座標とする。

Here, qx and rx are the x coordinates of the coordinates q and r, and qy and ry are the y coordinates of the coordinates q and r.

  Using the centroid coordinate r obtained by the centroid coordinate calculation unit 802 and the covariance matrix M obtained by the covariance matrix calculation unit 803, an appropriate region Ψ for obtaining the dark channel of the pixel at the coordinate p is determined as follows. .

この式は、重心座標ｒとのマハラノビス距離の二乗がＴ未満となる領域Ω内の座標を、座標ｐの画素のダークチャネルを求めるのに適切な領域Ψとするということを意味している。ここでＴは、予め与えられた閾値、Ｍ^−１は、共分散行列Ｍの逆行列とする。

This equation means that the coordinates in the region Ω where the square of the Mahalanobis distance with the barycentric coordinate r is less than T is set as a region Ψ appropriate for obtaining the dark channel of the pixel at the coordinate p. Here, T is a threshold value given in advance, and M ⁻¹ is an inverse matrix of the covariance matrix M.

  Thus, the barycentric coordinate and the covariance matrix M are information representing the area determined to be appropriate for obtaining the dark channel of the pixel at the coordinate p. The matrix M is output as information representing the determined area. The minimum pixel value calculation unit 500 receives the center-of-gravity coordinate r and the covariance matrix M, obtains the region Ψ from the equation (12), and calculates the minimum pixel value.

By determining the region in this way, it is possible to find a region in which pixels having a high degree of similarity are distributed to the pixel at the coordinate p. Such a region may be regarded as having almost constant transmittance, so that it becomes an appropriate region for obtaining the dark channel, and an optimum region for determining the dark channel is determined for each pixel, so that the transmittance with high accuracy is obtained. Can be calculated. In addition, since the transmittance with high accuracy is calculated without performing soft matting, it is possible to perform processing with a reduced amount of calculation.
[Third Embodiment]
FIG. 7 is a configuration diagram of a wrinkle removal image device according to the third embodiment. The difference from the first embodiment is that a transmittance noise removing unit 900 is added. Other parts are the same as those in the first embodiment. The description overlapping with the first embodiment will be omitted, and here, the transmittance noise removing unit 900 will be described.

  The transmittance noise removing unit 900 receives the transmittance output from the transmittance calculating unit 600 and outputs the transmittance from which noise has been removed. Various methods of noise removal are conceivable, for example, a median filter.

  Instead of using soft matting, determine the appropriate area for obtaining the dark channel for each pixel, and the transmission noise generated when estimating the transmittance from the dark channel obtained using that area, By removing by the transmittance noise removing means, the image quality of the image from which wrinkles are removed can be improved.

Needless to say, the transmittance noise removing unit 900 of the present embodiment is also applicable to the second embodiment.
[Fourth Embodiment]
The wrinkle removal image apparatus of this embodiment has the same configuration as that of the first embodiment. Therefore, the description of the parts that overlap with those in the first embodiment is omitted, and here, the position of the target pixel in the candidate area will be described.

  In the first embodiment, the number of pixels in the candidate area, the shape, and the position of the target pixel can be various. However, in this embodiment, the number of pixels and the shape of the candidate area are constant, and only the position of the target pixel is different. Suppose.

  Here, for simplification, the number of pixels and the shape of the candidate area will be described as, for example, a 7 × 7 square.

  In this case, as shown in FIG. 8, a candidate area having the highest similarity with the target pixel is obtained from the following five area candidates as a candidate area for the target pixel 811.

  Here, the coordinates of the pixel of interest 811 are assumed to be (x, y). A pixel of interest, a region 821 in FIG. 8A having (x-6, y-6), (x-6, y), and (x, y-6) as vertices, a pixel of interest 811, and (x, y-6), (x + 6, y-6), and (x + 6, y) as vertices, the region 822 in FIG. 8B, the pixel of interest 811 and (x-6, y), (x-6, y + 6) ), (X, y + 6) as a vertex in the region 823 and the target pixel 811 in FIG. 8C, and (x, y + 6), (x + 6, y), (x + 6, y + 6) as the vertex in FIG. ) Region 824, (x−3, y−3), (x−3, y + 3), (x + 3, y−3), and (x + 3, y + 3) as a vertex of the region 825 in FIG. Five areas are candidate areas.

  The region calculation unit 400 obtains a candidate region having the highest degree of similarity with the target pixel from among these five region candidates. The calculation is obtained by the method described above. The similarity between each candidate area and the target pixel is calculated to obtain the maximum similarity. A candidate region having the maximum similarity is determined as a region for obtaining a dark channel.

  As described above, the region calculation unit 400 determines the optimal region for obtaining the dark channel for each pixel, so that high-accuracy transmittance can be calculated.

  In addition, since the number of pixels and the shape of the candidate area are determined, the similarity can be examined by a relatively easy calculation.

  As described above with reference to the embodiment, according to the present invention, since an optimum region for obtaining a dark channel is determined for each pixel of interest, accuracy can be improved without using soft matting with a large amount of calculation. High transmittance can be calculated, and soot can be removed from the input image at high speed using highly accurate transmittance.

  Further, according to the present invention, it is possible to calculate the similarity between the target pixel and the plurality of candidate areas, and select the candidate area having the highest similarity as an appropriate area. Therefore, the candidate region having the highest similarity with the target pixel can be selected as the optimum region for obtaining the dark channel, and thus the transmittance with high accuracy can be calculated.

  Further, according to the present invention, it is possible to select an appropriate region for obtaining a dark channel from a plurality of candidate regions having different numbers of pixels. As a result, a more appropriate region can be set for each target pixel as compared with a case where a candidate region having a fixed number of pixels is selected.

  In addition, according to the present invention, it is possible to select an appropriate region for obtaining a dark channel from a plurality of candidate regions that have a fixed number of pixels and a different shape and differ only in the position of the target pixel. Thereby, it is possible to set an appropriate area for each target pixel with relatively easy calculation.

  In the present invention, the candidate area is a square having the target pixel as the lower right vertex, a square having the target pixel as the upper right vertex, a square having the target pixel as the lower left vertex, and a square having the target pixel as the upper left vertex. By using a square centered on the target pixel, an appropriate area can be set for each target pixel with a simple calculation.

  In the present invention, when the absolute value of the difference between the average pixel value and the pixel value of the target pixel is small, the similarity between the candidate region and the target pixel is high. Thereby, it is possible to easily calculate the similarity between the target pixel and the candidate region.

  Further, in the present invention, instead of using soft matting, an appropriate region for determining a dark channel is determined for each pixel, and the transmittance is estimated from the dark channel obtained using the region. By removing noise from the transmittance and removing wrinkles from the image using the transmittance after noise removal, the image quality of the image from which wrinkles have been removed can be improved.

  The present invention has been described above based on the embodiments. However, the present invention is not limited to the above-described embodiments, and various modifications and applications are possible within the scope of the gist of the present invention. . For example, the above functions can be realized by supplying a software program for realizing the functions of the above embodiments to a computer and executing the programs by the computer.

  Further, various inventions can be extracted from the above embodiment. For example, even if several constituent requirements are deleted from all the constituent requirements shown in the embodiment, a configuration in which this constituent requirement is deleted can be extracted as an invention if an effect can be obtained.

  DESCRIPTION OF SYMBOLS 100 ... Image processing apparatus, 200 ... Atmospheric light pixel value calculation part, 300 ... Atmospheric light correction part, 400 ... Area | region calculation part, 500 ... Minimum pixel value calculation part, 600 ... Transmittance calculation part, 700 ... Haze removal image generation part 800 ... Area calculation unit 801 ... Similarity calculation unit 802 ... Minimum pixel value calculation unit 803 ... Area calculation unit 900 ... Transmittance noise removal unit

Claims (13)

An atmospheric light correction unit for correcting the input image with the pixel value of the atmospheric light;
The pixel of interest of the input image, and the area calculation unit that the pixel of interest unrealized transmittance to calculate the local region that considered to be constant,
A minimum pixel value selection unit for selecting a minimum pixel value in the local region in the input image corrected by the pixel value of the atmospheric light;
A transmittance calculating unit that calculates the transmittance of the target pixel from the minimum pixel value;
A wrinkle removal image generating unit that removes a wrinkle component from the input image using the transmittance;
An image processing apparatus comprising:   The image processing according to claim 1, wherein the region calculation unit calculates a similarity between the target pixel and a plurality of candidate regions, and selects a candidate region having the highest similarity as the local region. apparatus.   The area calculation unit calculates a maximum similarity from the similarity of areas having a maximum number of pixels of candidate areas having different numbers of pixels, and the candidate area having the maximum similarity if the maximum similarity is equal to or greater than a predetermined threshold If it is less than the predetermined threshold, the maximum similarity is calculated from the similarity of the area having the next largest number of pixels, and the process of comparing with the threshold is repeated until the area is selected. Item 3. The image processing apparatus according to Item 2.   The image processing apparatus according to claim 2, wherein the region calculation unit selects the local region from a plurality of candidate regions that have a fixed number of pixels and a different shape, and differ only in the position of the target pixel.   The candidate area is a square with the target pixel as a lower right vertex, a square with the target pixel as an upper right vertex, a square with the target pixel as a lower left vertex, a square with the target pixel as an upper left vertex, The image processing apparatus according to claim 4, wherein the image processing apparatus is a square centered on the target pixel.   The region calculation unit is configured such that when the absolute value of the difference between the average value of the pixel values in the candidate region and the pixel value of the target pixel is small, the similarity between the candidate region and the target pixel is high. The image processing apparatus according to claim 2. A transmittance noise removing unit that performs noise removal on the transmittance calculated by the transmittance calculating unit;
The image processing apparatus according to claim 1, wherein the wrinkle removal image generation unit removes wrinkle components from the input image using the transmittance after noise removal. An atmospheric light correction unit for correcting the input image with the pixel value of the atmospheric light;
Calculating a similarity between peripheral pixels of the target pixel and the pixel of interest of the input image, unrealized transmittance the pixel of interest by using the similarity degree and the area calculation unit for calculating a local region that considered to be constant ,
A minimum pixel value selection unit for selecting a minimum pixel value in the local region in the input image corrected by the pixel value of the atmospheric light;
A transmittance calculating unit that calculates the transmittance of the target pixel from the minimum pixel value;
A wrinkle removal image generating unit that removes a wrinkle component from the input image using the transmittance;
An image processing apparatus comprising:   The area calculation unit obtains a centroid coordinate with the similarity as a weight, inputs the similarity of the peripheral pixels and the centroid coordinate, calculates a weighted covariance matrix with the similarity as a weight, The image processing apparatus according to claim 8, wherein an appropriate area for obtaining a minimum pixel value is obtained from the weighted covariance matrix and a predetermined threshold value. Correcting the input image with atmospheric light pixel values;
The pixel of interest of the input image, the steps of the pixel of interest unrealized transmittance to calculate the local region that considered to be constant,
Selecting a minimum pixel value in the local region in the input image corrected by the pixel value of the atmospheric light;
Calculating the transmittance of the pixel of interest from the minimum pixel value;
Removing wrinkle components from the input image using the transmittance;
An image processing method comprising: Further comprising the step of removing noise from the transmittance calculated by the step of calculating the transmittance;
The image processing method according to claim 10, wherein a wrinkle component is removed from the input image using the transmittance after noise removal. Correcting the input image with atmospheric light pixel values;
The pixel of interest of the input image, the steps of the pixel of interest unrealized transmittance to calculate the local region that considered to be constant,
Selecting a minimum pixel value in the local region in the input image corrected by the pixel value of the atmospheric light;
Calculating the transmittance of the pixel of interest from the minimum pixel value;
Removing wrinkle components from the input image using the transmittance;
An image processing program for causing a computer to exhibit the above. Further comprising the step of removing noise from the transmittance calculated by the step of calculating the transmittance;
The image processing program according to claim 12, wherein a wrinkle component is removed from the input image using the transmittance after noise removal.

Images