KR101867925B1 - Image processing apparatus for image haze removal and method using that - Google Patents

Abstract

The present application provides an image processing apparatus and method for removing a haze included in a still image, capable of improving image quality by effectively estimating and removing a haze component from the still image including the haze at high speed by approximating a final cost function value by which an image loss amount is minimized while differentially improving a contrast ratio according to the depth information of an image in the image.

Description

[0001] IMAGE PROCESSING APPARATUS FOR IMAGE HAZE REMOVAL AND METHOD USING THAT [0002]

The present application relates to an image processing apparatus and method for eliminating haze included in still images.

Fog is a phenomenon in which atmospheric water vapor floats in a condensed form of water droplets. When the mist is fogged, visibility disturbance of less than 1km appears. When fog occurs, droplet particles are generated in the atmosphere and scattering of light due to these droplet particles occurs. The scattering of light means that the light collides with particles in the air and changes its direction of travel. This depends on the wavelength of light and particle size.

Generally, light scattering is largely divided into Rayleigh scattering and Mie scattering. Rayleigh scattering is applied when the size of the scattering particle is smaller than the wavelength of light, and the scattering energy is inversely proportional to the square of the wavelength. For example, when a sunny day is scattered by air molecules, blue is scattered more than red and the sky is blue. However, the scattering of light in the case where the size of the particle, which is a factor of scattering, is larger than the wavelength of light, is based on the micro scattering theory. In the case of fog, the particle diameter is from several micrometers to several tens of micrometers, which is larger than the wavelength of 400 to 700 nm, which is the wavelength of the visible light ray, and thus it follows the myel scattering theory. According to the Mie scattering theory, when the size of scattering particles such as water drops in the atmosphere is large, the amount of scattering is less influenced by the wavelength, so that almost all the light in the visible light region is scattered almost equally. Therefore, when the fog is caught, the objects appear cloudy. The new light source generated at this time is called air light.

Improvement of image quality through correction of fog distortion can solve visibility problems and make blurred images clear. Also, it has importance as a preprocessing step for recognition by restoring information about text, objects, etc. damaged due to fog.

Conventional fog removal techniques can largely be divided into a non-model method and a model method.

An example of the former is histogram equalization. Histogram Smoothing Histogram smoothing is a method of redistributing the distribution by analyzing the histogram of the image. This histogram smoothing is easy and effective, but it is not suitable for fog images where the depth is not uniform. Also, it is suitable for general image quality improvement, but it does not reflect characteristics of influence of fog on the image, so that there is a disadvantage that the improvement of dark fog image is insignificant.

The latter modeled the effect of light scattering due to fog on the image. First, there is disclosed a technique of correcting distortion caused by fog by comparing a scene of two or more images obtained in different weather environments, estimating a scene depth, and correcting the scene depth. However, this requires two image inputs having different weather conditions, so that real-time implementation requires a video storage space corresponding to the change of weather while sensing the change of weather. Since it is impossible to predict the cycle of the weather change, it becomes difficult to determine the storage cycle. In addition, since the same scene without errors should be photographed, an error may occur when estimating the degree of fog distortion when there is a moving object.

Next, a technique for correcting the distortion caused by mist by estimating and subtracting the degree of change of the pixel value of the image by the fog is specified. However, since it is based on the assumption that the fog is homogeneous, it is applicable to uniform and light fog, but there are more cases of nonuniform fog, and even if the fog is uniform, depending on the distance between the camera and the object, There is a problem in actual application.

In addition, general image contrast ratio enhancement techniques such as histogram smoothing or gamma correction can improve the contrast ratio of the entire image, but it is difficult to improve the contrast ratio of the image in the fog image in which the contrast ratio is lowered differently according to the depth information of the image.

The present application relates to a method and apparatus for estimating a final cost function value by approximating a final cost function value based on a contrast ratio and a video loss amount in an image to improve the contrast ratio in accordance with depth information of the image, The present invention also provides an image processing apparatus and method that can improve the image quality by effectively estimating and removing haze components from still images including haze at a high speed.

An approximation unit for approximating a final cost function value based on a contrast ratio and a video loss amount in the image, and a histogram unit for approximating a final cost function value based on the approximated final cost function, A transmission amount estimating unit estimating a transmission amount and estimating a pixel-by-pixel transmission amount using the block-by-block transmission amount; and an image restoring unit restoring the image using the haze brightness and the pixel- There is provided an image processing apparatus for removing the image.

The image processing apparatus according to an embodiment of the present application approximates the final cost function value using a logarithmic function so that the contrast ratio is improved differentially according to the depth information of the image in the image, The cost function value can be computed at a high speed, and the image quality can be improved by efficiently estimating and removing the haze component from the still image including the haze at a high speed.

1 is a block diagram showing the configuration of an image processing apparatus for removing haze included in a still image according to the present application.
2 is a block diagram showing in detail the configuration of the haze brightness measuring unit in the image processing apparatus according to the present application.
3 is a block diagram showing in detail the structure of the final cost function approximation unit in the image processing apparatus according to the present application.
FIG. 4 is a diagram for explaining a general concept of minimizing a video loss.
5 is a diagram for explaining the concept of discontinuity of the amount of transmission obtained by a general image loss amount.
FIG. 6 is a graph showing the logarithmic function prototype g (x) expressed by Equation (7).
7 is a flowchart illustrating an image processing method according to the present invention.
FIG. 8 is a flowchart illustrating in detail measurement steps of the image processing method according to the present invention.
9 is a flowchart illustrating an approximation step and an estimation step of the image processing method according to the present invention in detail.

The present application relates to an image processing apparatus. An exemplary image processing apparatus of the present application can approximate a final cost function value using a logarithmic function to calculate a final cost function value having a maximum contrast ratio and a minimum image loss at the same time at a high speed, Accordingly, it is possible to effectively estimate and remove the haze component from the still image including the haze at a high speed.

The term " haze " in the present application may be fog in the air, but is not necessarily limited thereto, and may be a fog-like one similar to fog, for example, yellow dust, dust and air suspended in air.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. However, the present application may be embodied in many different forms and is not limited to the embodiments described herein. In order to clearly illustrate the present invention, parts not related to the description are omitted, and similar parts are denoted by like reference characters throughout the specification.

An image processing apparatus according to the present application includes a measuring unit, an approximating unit, an estimating unit, and a restoring unit.

In one example, the image processing apparatus can restore an image including a haze using Equation (1).

[Equation 1]

Here, I _p is (image being affected by haze) brightness value of the captured image at the pixel position p, J _p is (image Im that is being affected by haze) original luminance value of the captured image at the pixel position p, t _p (Depth information, distance between the image and the corresponding point) and A represents the haze brightness value at the pixel position p.

At this time, the image processing apparatus makes a cost function considering the cost (E _c ) considering the contrast ratio and the image loss (E _L ) that can occur when the contrast ratio is improved, and calculates the transfer amount (t) The image including the haze is restored using Equation (1).

Hereinafter, a technique for restoring a still image including a haze using Equation (1) by the image processing apparatus will be described with reference to the drawings.

FIG. 1 is a block diagram illustrating a configuration of an image processing apparatus for removing haze included in a still image according to the present invention, FIG. 2 is a block diagram illustrating a configuration of a haze brightness measuring unit in the image processing apparatus according to the present application, And FIG. 3 is a block diagram showing in detail the structure of the final cost function approximation unit in the image processing apparatus according to the present application.

1, an image processing apparatus 100 for removing haze included in a still image includes a haze brightness measuring unit 110, a final cost function approximating unit 120, a block unit transmission amount estimating unit 130, A transmission amount estimating unit 140, and an image restoring unit 150. The terms " approximate part " and " final cost function approximation part " have the same meaning in the present application.

The haze brightness measuring unit 110 divides an image including the haze into a predetermined number of blocks to obtain representative values for each block, selects a block having the largest value among the representative values, And dividing the selected block into the predetermined number of blocks until a representative value is obtained. The image including the haze may be a still image including haze.

In order to obtain the haze brightness value, it is assumed that "the haze has a very bright value in the image because it is the light in the atmosphere reached from the sun" and "the pixel value influenced by the haze has a value similar to the haze brightness value" . Referring to Equation (1), it can be seen that the larger the amount of transmission (depth information, distance between the image and the object) in the acquired image is, the larger the influence of haze is. Therefore, The standard deviation is small because it is distributed around the value.

Therefore, it can be seen that, by combining the above two assumptions, the haze brightness value has a very bright value in the image and can be obtained in the region most affected by haze. Generally, the haze is selected as the brightest pixel value, which causes a problem because the white object may be selected. Therefore, in order to narrow down the candidate group of the haze value gradually while reducing the block gradually, the detection is divided into blocks. When the candidate group is finally reduced (the block of the set width), all pixel values in the block are mostly influenced by the haze brightness value , It is less likely that an error will occur even if the brightest pixel value is selected as the haze brightness value. Accordingly, the haze brightness measuring unit 110 can select the brightest pixel value as the haze brightness value.

Referring to FIG. 2, the haze brightness measuring unit 110 includes a region dividing module 112, a representative value calculating module 114, and a pixel value selecting module 116.

The area division module 112 divides the image including the haze into a predetermined number of blocks. The predetermined number of blocks may be two or four, that is, it may be two or four.

When a representative value is obtained from a block, the average brightness and standard deviation of the pixel values within the block are used. These values may vary depending on the size of the block. For example, when a representative value is obtained by using a block having a size of 20 x 20, if there is a bright object such as a white building or a white vehicle in a haze-free area, the representative value can be selected to be high. If the size of the block is enlarged, the representative value is selected from the side having a larger haze value. However, since it is difficult to find the haze value in the block, the block is gradually reduced and the image is blocked to narrow the candidate value of the haze value .

The representative value calculation module 114 obtains an average and a standard deviation of pixels for each of the divided blocks in the area division module 112 and obtains a representative value based on a difference between an average and a standard deviation for each block.

That is, the haze brightness value has a very bright value in the image and can be acquired in the region most affected by the haze. Since the majority of the region affected by the haze is distributed around the haze value, . Accordingly, the representative value calculation module 114 obtains a representative value using an average and a standard deviation of each block to obtain a haze brightness value.

The pixel value selection module 116 selects the block having the largest value among the representative values obtained for each block in the representative value calculation module 114. When the width of the selected block becomes equal to or less than a predetermined width, The brightest pixel value in the block is selected as the haze brightness.

That is, when the candidate group is finally reduced (block of the set width), all the pixel values in the block are mostly affected by the haze value. Therefore, even if the brightest pixel value is selected as the haze brightness, the error is less likely to occur. Accordingly, when the width of the block is less than or equal to a predetermined width, the pixel value selection module 116 selects the brightest pixel value as the haze brightness.

The pixel value selection module 116 selects a block having the largest representative value and determines whether the width of the selected block is equal to or less than a predetermined width. Here, the predetermined width may be, for example, 200 (length * vertical length) of the block. As a result of the determination, if the width of the selected block becomes 200 or less, the pixel value selection module 116 selects the brightest pixel value in the selected block and estimates the selected pixel value as the haze brightness.

If it is determined that the width of the selected block is not less than 200, the pixel value selection module 116 divides the selected block into a predetermined number and obtains representative values, selects a block having the largest representative value, If it is determined that the block size is less than or equal to the predetermined width, the brightest pixel value in the selected block is estimated as the haze brightness.

Referring to FIG. 1 again, the final cost function approximation unit 120 divides the image including the haze into a predetermined number of blocks, calculates a video loss amount using a contrast ratio for each block, and Equation (4) Then, the final cost function value based on the contrast ratio and the image loss amount can be approximated by using Equation (10) described later.

3, the final cost function approximation unit 120 includes an image segmentation module 122, a cost function calculation module 124, and a final cost function calculation module 126. The final cost function approximation unit 120 includes a final cost function approximation unit 120, do.

The image division module 122 divides the image including the haze into a predetermined number of blocks. In this case, when the block size is too small, it is difficult to estimate the transmission amount. Therefore, the minimum size of the block is set to 16 * 16 or more, and the number of blocks can be arbitrarily set. Here, the area division is performed to compensate the haze component in consideration of the unevenness of the haze, and the number of blocks for the area division can be appropriately selected in consideration of the hardware complexity.

In order to restore the image, Equation 1 is used. Since only the haze brightness value obtained by the haze brightness measuring unit 110 is known, the transmission amount t _p and the original brightness value J _p must be calculated for every pixel. There is an under-determined problem. Therefore, if we assume that the transfer quantities (t _p ) are all the same in the block, the number of unknowns can be greatly reduced to (B + 1) unknowns for B (the number of pixels in the block). Therefore, if the same amount of transmission (t) is assumed for the entire image without dividing the image, the image degradation according to the depth information of the image can not be adaptively compensated.

The control unit 120 calculates the contrast ratio and the image loss amount while changing the transmission amount of the predetermined range for each of the divided blocks in the image division module 122 by a predetermined ratio.

(1), the loss of the output image determined by the amount of transmission (t) should be minimized. (2) When the transmission rate t is to be calculated, Should be satisfied.

Therefore, the cost function calculation module 124 creates a cost function that considers the cost (E _c ) considering the contrast ratio and the image loss (E _L ) that can occur when the contrast ratio is improved, and calculates the cost ratio .

Therefore, the cost function calculation module 124 includes a contrast ratio calculation module 124a for calculating a contrast ratio, and a video loss calculation module 124b for obtaining a video loss amount.

The contrast ratio calculation module 124a calculates the contrast ratio E _c using the variance (δ ² ) using the pixel values of each pixel as shown in Equation ( ² ).

&Quot; (2) "

는 원래 객체가 가지고 있는 밝기의 평균을 의미한다.In Equation (2), N is the number of pixels, J _p is the brightness of the original object, p is the position of each pixel,

Means the average of the brightness of the original object.

Equation (2) uses RMS contrast, which is often used as a method of calculating the contrast in an image, and RMS contrast measures the contrast ratio using the variance of the image.

In addition, since the control ratio of the output image determined by the amount of transmission t must be large, the objective is to find the amount of transmission t at which the cost function is minimized (optimum value) Since the value increases as the contrast ratio increases, the contrast function is marked with a minus sign to set the cost function to decrease as the amount of transmission decreases.

That is, the contrast ratio calculation module 124a multiplies the variance using the pixel value of each pixel for each block by -1 to obtain a contrast ratio. The second cost function is a function for finding the amount of transmission that maximizes the contrast ratio of the restored image. The second cost function is a function for finding the block-by-block transmission amount maximizing the contrast ratio Always converges to '0'. When the block unit transmission amount converges to '0', the output pixel value for the input pixel value is determined as a value exceeding the pixel value [0 255] that the computer can represent.

That is, when the block unit transmission amount converges to '0', an output pixel value of '0' or less or '255' or more exists as a video loss, and it is necessary to minimize the video loss. The video loss amount refers to an amount of the output pixel value of '0' or less or '255' or more, which is cut to 0 and 255.

FIG. 5 is a diagram for explaining the concept of discontinuity of the amount of transmission obtained by a general amount of image loss, and FIG. 6 is a diagram illustrating a logarithmic function circularity g (x)).

Meanwhile, in the conventional image processing apparatus, the video loss calculation module 124b calculates a video loss amount for each block by using a second cost function according to Equation (3) to minimize a video loss.

&Quot; (3) "

The concept of Equation (3) can be expressed as an area of area A as shown in FIG. The A region is a pixel value of '0' or less or '255' or more, and refers to a video loss amount.

5, since the pixel value that can be represented by the computer is [0, 255], the actual function of the second cost function is such that the part of the area A is '0' or '255' . Accordingly, the actual function of the second cost function appears as a function having a discontinuous value at the point C where the values of the x-axis are? And?. In other words, the actual function of the second cost function appears as a function having a partially linear property.

In general, it is difficult to obtain a solution of a discontinuous function. Therefore, in the case of a final cost function based on a first cost function and a second cost function having a discontinuous value, it's difficult.

On the other hand, in the present application, since the image loss is approximated using the first conversion function and the second conversion function that are obtained by converting the continuous log function, and the final cost function value is obtained based on the approximated image loss amount, The device can calculate the final cost function value that maximizes the contrast ratio while simultaneously minimizing the video loss.

In one example, the video loss calculation module 124b may approximate the video loss using a first transform function and a second transform function. For example, the difference between the first conversion function and the second conversion function can be used to approximate the video loss amount.

Also, the video loss calculation module 124b approximates the video loss amount for each block using the following equation (4).

&Quot; (4) "

는 제1 변환 함수를 나타내며,

는 제2 변환 함수를 나타낸다.In Equation (4)

Represents a first transform function,

Represents a second transform function.

In one example, the first conversion function is a function obtained by converting a logarithmic function prototype (g (x)) described later, and can be expressed by a function of the following equation (5).

&Quot; (5) "

In Equation (5), A represents the brightness of the haze, and t represents the amount of transmission.

In Equation (4), the second conversion function is a function obtained by converting the function of Equation (13), which will be described later, and can be expressed by a function of Equation (6).

&Quot; (6) "

In Equation (6), A represents the brightness of the haze, and t represents the amount of transmission.

The image processing apparatus of the present application approximates a video loss amount by using a first conversion function and a second conversion function converted to a continuous log function instead of the second cost function having a conventionally partially linear property, Can be obtained accurately and at a high speed.

In one embodiment, the image loss amount calculation module 124b can approximate the image loss amount using the log function shown in FIG. 6, and more specifically, the log function can be approximated by a logarithmic function circular form g (x)).

&Quot; (7) "

는 함수의 x축 방향 이동을 나타낸다.In Equation (7), L represents a magnitude of a y-axis direction function which is an integer within a range of 0 to 255, k represents a slope,

Represents the x-axis movement of the function.

In one example, the video loss calculation module 124b can approximate the video loss amount by converting the log function prototype g (x) expressed by Equation 7 into the first conversion function described above.

Specifically, the first transformation function is a transformation function approximated by a function relationship expressed by the following equation (8) in the logarithmic function prototype g (x).

&Quot; (8) "

는 x가 '0'일 때 x에 대해 미분한 로그 함수 원형을 나타내고,

는 x가 '0'일 때 x에 대해 미분한 제2 변환함수를 나타낸다.In Equation (8)

Represents a differentiated logarithmic function prototype for x when x is '0'

Represents a second transform function that is differentiated with respect to x when x is '0'.

In one example, the logarithmic function prototype g (x) in Equation (8) may have an L value of 255, and the slope approximated by the function relationship of Equation (8) Can be expressed.

&Quot; (9) "

In Equation (9), t represents the amount of transmission.

As described above, the first conversion function may be a conversion function when the logarithmic function prototype g (x) has the slope of Equation (9).

When the control ratio and the video loss amount are obtained through Equations (2) and (4), the final cost function calculating module 126 obtains the final cost function value based on the calculated contrast ratio and the video loss amount.

That is, when the transmission amount t is to be obtained, the output ratio of the output image determined by the amount of transmission must be large (1), and the loss of the output image determined by the amount of transmission must be minimized (2). In order to find the amount of transmission satisfying the above two conditions, it is necessary to make the two conditions as a function called the cost function and find the amount of transmission that minimizes the amount. In other words, when the cost function is minimized, the contrast ratio of the output image is large and the loss of the image is small.

Therefore, since both of the above two conditions must be included in the final cost function, when calculating the block unit transfer amount, a final cost function such as Equation (4) including all of them is created and a final cost function value is obtained using the final cost function.

&Quot; (10) "

In Equation (10), E denotes a final cost function value, E _c denotes a control ratio value, E _L denotes a video loss amount, and _L denotes a video loss amount coefficient.

The block-based transfer amount detection module 128 detects an amount of transfer which minimizes the final cost function value obtained by the final cost function calculation module 126 as a block-by-block transfer amount. That is, since the block-based transmission amount is one of values corresponding to a range from 0 to 1, the block-based transmission amount detection module 138 changes the values (numbers) between 0 and 1 to the block- Value, and detects the amount of transmission at which the final cost function value becomes the minimum as the block-by-block transmission amount. Here, a value (number) between 0 and 1 means a value such as 0.1, 0.2, etc., and the value can be changed at an arbitrary interval or ratio.

1, the block-based transmission-amount estimating unit 130 and the pixel-based transmission-amount estimating unit 140 may be combined into one unit. In this case, the block-based transmission-amount estimating unit 130 and the pixel- May be constituted by a transmission amount estimating unit.

Referring to FIG. 3, the block-based transfer amount estimating unit 130 may include a block-based transfer amount detecting module 132, and the detecting module may calculate a block- And detects the amount of transmission that minimizes the value as a block-by-block transmission amount. That is, since the block-by-block transmission amount is one of values corresponding to a range from 0 to 1, the block-by-block transmission amount detection module 128 changes values (numbers) between 0 and 1 to block- Value, and detects the amount of transmission at which the final cost function value becomes the minimum as the block-by-block transmission amount.

In the present application, the term " value (number) between 0 and 1 " means a value such as 0.1, 0.2, etc., and the value may be changed at an arbitrary interval or ratio.

Referring again to FIG. 1, the pixel-unit-amount-of-transmission-amount estimating unit 140 applies the block-unit transmission amount detected by the block-by-block transmission-amount estimating unit 130 to an edge-preserving filter to convert it into a pixel-by-pixel transmission amount. The edge preserving filter generates an approximate amount of pixel-by-pixel transmission using the pixel values of the input image.

Therefore, the pixel-unit-transmission-amount estimating unit 140 obtains an approximate transmission amount per pixel using Equation (5).

&Quot; (11) "

는 화소 단위의 근사 전달량, I_p는 카메라에서 획득한 밝기를 의미하고, a, b는 에지 보존 필터를 적용하여 구해진 선형함수의 변수를 의미한다.In Equation (11) above,

I _p is the brightness obtained by the camera, and a and b are the parameters of the linear function obtained by applying the edge preserving filter.

The input image has a value between 0 and 255, where t has a value between 0 and 1. Accordingly, the expression (11) is a formula for reducing the value between 0 and 255 between 0 and 1 through a and b in order to express t using the input image, and adjusting the size (average value) to be similar.

Accordingly, the pixel-unit-amount-of-transmission-amount estimation unit 140 can obtain the pixel-by-pixel transmission amount by obtaining a and b that minimize the difference between the block-based transmission amount and the approximate transmission amount. Compute a and b using a window of size W and calculate a and b by moving to the next pixel. A and b are averaged in the overlapping region.

In this case, Equation 12 is defined as Equation 12 to minimize the difference between the approximate transmission amount and the block unit transmission amount.

&Quot; (12) "

In Equation (12), p denotes a pixel position, and B denotes a block.

In order to find a and b which minimizes Equation (12), the pixel-unit-transmission-amount estimating unit (130) uses a least squares method used in linear algebra.

Accordingly, the pixel-based transmission amount estimation unit 140 detects the approximate transmission amount by the a and b variables obtained by the least squares method as the pixel-by-pixel transmission amount.

In other words, the data of the image changes every pixel, and since the calculated amount of calculation is calculated in units of blocks, when the image is restored by using the amount of transmission, block deterioration occurs due to the difference in the amount of transmission at the boundary of the block. The block deterioration phenomenon refers to a phenomenon in which a difference in pixel value greatly spreads at a boundary portion of a block. Therefore, in order to eliminate the block deterioration phenomenon, it is necessary to convert the computed transmission amount into the pixel-based transmission amount.

The image reconstructing unit 140 removes haze from the image using the haze brightness obtained by the haze brightness measuring unit 110 and the pixel-by-pixel transmission amount obtained by the pixel-based transmission amount estimating unit 140.

For example, the image reconstructing unit 150 generates an image with the haze removed using Equation (13).

&Quot; (13) "

In Equation (13), Ip is the brightness acquired by the camera, Jp is the brightness of the original object, p is the position of each pixel, A is the haze brightness, and tp is the pixel-unit transmission amount.

The present application relates to an image processing method for removing haze contained in still images. The above method can be performed using the image processing apparatus, and therefore, the description of the parts overlapping with those described in the image processing apparatus will be omitted. An exemplary image processing method includes: a measurement step of measuring haze brightness in an image including haze; an approximation step of approximating a final cost function based on the contrast ratio and the image loss amount in the image; Estimating a block-by-block transmission amount, estimating a pixel-by-pixel transmission amount using the block-by-block transmission amount, and restoring the image using the haze brightness and the pixel-by-pixel transmission amount.

Hereinafter, the image processing method of the present application will be described with reference to the accompanying drawings, and the attached drawings are illustrative, and the scope of the image processing method of the present application is not limited by the accompanying drawings.

7 is a flowchart illustrating an image processing method according to the present invention.

Referring to FIG. 7, in the image processing method according to the present application, when an image including haze is input (S702), the haze brightness is measured on the image (S704). The haze brightness measurement (S704) will be described in detail with reference to FIG.

In step S704, the image processing apparatus divides the image including the haze into a predetermined number of blocks and estimates the amount of block-by-block transmission in step S706. The details of the block-by-block transmission amount estimation (S706) will be described with reference to Fig.

In the image processing method according to the present application, if the step S706 is performed, the block unit transmission amount is applied to the edge preservation filter and converted into the pixel unit transmission amount (S708). Then, the haze-free image is generated using Equation (13) using the above-described haze brightness and the pixel-unit transmission amount.

FIG. 8 is a flowchart illustrating in detail measurement steps of the image processing method according to the present invention.

Referring to FIG. 8, the measuring step may be performed according to a series of processes. In step S802, for example, the image including the haze is divided into two or four blocks (S802), and the average and standard deviation of the pixel values of the respective blocks are obtained (S804).

Then, a representative value based on the difference between the average and the standard deviation of each block is obtained for each block (S806). Then, a block having the largest value among the representative values obtained for each block is selected (S808), and it is determined whether or not the width of the selected block is 200 or less (S810).

If it is determined in step S810 that the size of the selected block is 200 or less, the brightest pixel value in the selected block is selected and defined as the haze brightness (S812).

If it is determined in step S810 that the size of the selected block is less than or equal to 200, step S802 is performed.

9 is a flowchart illustrating an approximation step and an estimation step of the image processing method according to the present invention in detail.

The approximating and estimating steps may be performed according to a series of steps as shown in FIG. For example, the process divides the image including the haze into blocks of size B (S902). After the execution of step S902, the transmission ratio and the image loss amount are determined while varying the predetermined amount of transmission for each block by a predetermined ratio (S904). In step S904, the video loss amount is obtained by using, for example, Equations 4 to 9 described above.

After performing step S904, the image processing apparatus obtains a final cost function value based on the contrast ratio and the image loss amount for each block (S906), and selects an amount of transmission which minimizes the obtained final cost function value (S908) . In step S906, the final cost function value is obtained, for example, by using the above-described equation (10).

Then, the image processing apparatus detects the selected transmission amount as a block-by-block transmission amount (S910).

In the image processing method of the present application, for example, the processes of S904 and S906 refer to an approximation step, and the process of S908 refers to an estimation step.

The present invention can be used to improve the contrast ratio of images in various fields such as flight, airport, transportation, shipping, underwater surveillance, medical imaging (removal of cloudy substances generated by internal gas, etc.), agricultural surveillance, .

The method for removing the haze on the still image can be programmed, and the codes and code segments constituting the program can be easily deduced by programmers in the field.

Thus, those skilled in the art will appreciate that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics thereof. It is therefore to be understood that the embodiments described above are to be considered in all respects only as illustrative and not restrictive. The scope of the present invention is defined by the appended claims rather than the detailed description and all changes or modifications derived from the meaning and scope of the claims and their equivalents are to be construed as being included within the scope of the present invention do.

100: image processing device
110: Haze brightness measuring unit
120: final cost function approximation part
130: Block-by-block transmission amount estimating unit
140: Per pixel unit transmission amount estimating unit
150:

Claims (12)

A haze brightness measuring unit for measuring haze brightness in an image including haze;
An approximation unit for approximating a final cost function value by the contrast ratio and the image loss amount in the image;
A transfer amount estimating unit estimating a block unit transfer amount based on the approximated final cost function value and estimating a pixel unit transfer amount using the block unit transfer amount; And
And an image reconstruction unit for reconstructing the image using the haze brightness and the pixel-
Wherein the approximating unit includes an image processing unit for removing haze included in a still image approximating a video loss amount using Equation (4)
&Quot; (4) "

In Equation (4)

Represents a first transform function,

Represents a second transformation function. delete delete The image processing apparatus according to claim 1, wherein the first transform function is represented by Equation (5)
&Quot; (5) "

In Equation (5), A represents the brightness of the haze, and t represents the amount of transmission. The image processing apparatus according to claim 1, wherein the second transformation function is represented by Equation (6)
&Quot; (6) "

In Equation (6), A represents the brightness of the haze, and t represents the amount of transmission. 2. The image processing apparatus according to claim 1, wherein the approximation unit includes an image processing unit for removing haze included in a still image approximating a video loss amount using a logarithmic function prototype g (x)
&Quot; (7) "

In Equation (7), L represents a magnitude of a y-axis direction function which is an integer within a range of 0 to 255, k represents a slope,

Represents the x-axis movement of the function. 7. The image processing apparatus according to claim 6, wherein the slope (k) is approximated using a function relationship represented by the following equation (8): < EMI ID =
&Quot; (8) "

In Equation (8)

Represents a differentiated logarithmic function prototype for x when x is '0'

Represents a second transform function that is differentiated with respect to x when x is '0'. 7. The image processing apparatus according to claim 6, wherein the slope (k) is represented by the following equation (9) when the L value is 255:
&Quot; (9) "

In Equation (9), t represents the amount of transmission. A measuring step of measuring a haze brightness in an image containing haze;
An approximation step of approximating a final cost function based on the contrast ratio and the image loss amount in the image;
Estimating a block unit transmission amount based on the approximated final cost function value and estimating a pixel unit transmission amount using the block unit transmission amount; And
And restoring the image using the haze brightness and the pixel-unit transmission amount,
Wherein the approximating step includes a step of removing the haze included in the still image approximating the image loss using Equation (4)
&Quot; (4) "

In Equation (4)

Represents a first transform function,

Represents a second transform function. delete The image processing method according to claim 9, wherein the first transform function is represented by Equation (5)
&Quot; (5) "

In Equation (5), A represents the brightness of the haze, and t represents the amount of transmission. The image processing method according to claim 9, wherein the second transformation function is represented by Equation (6)
&Quot; (6) "

In Equation (6), A represents the brightness of the haze, and t represents the amount of transmission.

Images