KR100771158B1 - Method AND System For Enhancement Color Image Quality - Google Patents

Abstract

The present invention relates to a method and system for improving the image quality of a color image. Particularly, the present invention relates to a color image obtained in a state in which illumination is uneven using a single scale retinex (SSR) based on JND (Just Noticeable Difference). A method and system for improving the image quality.

In the present invention, while improving the brightness (V; Value) and saturation (S;) while maintaining the hue (H; hue) intact for the H, S, V color image, the brightness (V) first, illumination Based on the model of brightness (V) given by the product of the component and the reflection component, the lighting component is estimated using a JND-based low pass filter and then reflected and illuminated at the original brightness (V) and the estimated lighting component. The logarithm operation is taken for each separation and then the nonuniform illumination is compensated for by the difference between the original brightness and some components of the estimated reflection components. The image of the compensated illumination is automatically adjusted to fit the range of brightness values of the output device using histogram modification and then saturation (S) is adjusted in proportion to the improved proportion of brightness (V). Then, an improved color image is obtained from the improved brightness (V) and saturation (S) in the original color (H).

JND, Retinex, SSR

Description

 Method and system for improving color image quality {Method AND System For Enhancement Color Image Quality}

1 is a diagram illustrating a method of estimating a reflection component by removing an illumination component from a conventional input image.

2 is a view showing an example of application only to the brightness (V) using the color coordinate system of the conventional color (H), saturation (S), V (brightness)

3 is a diagram for improving image quality of a color image using JND-based SSR for an embodiment of the present invention;

4 is a flowchart for improving image quality of a color image using JND-based SSR according to an embodiment of the present invention.

5 is a comparative example of the improved color image 5d and the observed image 5a of the present invention.

6 is a filtering characteristic diagram of a JND based low pass filter according to an exemplary embodiment of the present invention.

The present invention relates to a color image improvement method and system by SSR, and more particularly, to a color image obtained in a state in which illumination is uneven using a single scale retinex (SSR) based on JND (Just Noticeable Difference). It relates to a method and a system to improve the.

When the image is acquired using a portable terminal or a digital camera, the image quality is greatly affected by the lighting conditions. For example, methods of improving the quality of deteriorated images due to poor lighting or uneven lighting include gain and offset correction methods, log or gamma functions, histogram equalization, and images. There is homomorphic filtering based on an image formation model that gives a product of a lighting component and a reflection component. This is described in "AK Jain, Fundamentals of Digital Image Processing , Prentice-Hall, 1989. and RC Gonzalez, RE Woods, Digital Image Processing , Addison-Wesley, 1992." And TK Kim, JK Paik, and BS Kang, "Contrast enhancement system using spatially adaptive histogram equalization with temporal filtering, IEEE Trans. Consumer Electronics , vol. 44, no. 1, pp. 82-86, Jan. 1998 ". Homomorphic filtering takes a logarithmic operation on the observed image and converts it to the sum of the logarithmic calculated illumination components and the reflected components, respectively, to suppress the illumination components having mainly low frequency components, Reflective components with are emphasized, and the exponential function is finally removed to remove the effects of the logarithms, especially for the purpose of improving color images, to compensate for non-uniform lighting components based on homomorphic filtering-like image formation models. Retinex methods include "Z. Rahman, D. Jobson, and GA Woodell," Properties and performance of a center / surround retinex, I EEE Trans. Image Processing: Special Issue on Color Processing , vol. 6 , pp. 451-462, no. 3, Mar. 1997. with DJ Jobson, Z. Rahman, and GA Woodell, "A multi-scale retinex for bridging the gap between color images and the human observation of scenes, IEEE Trans. Image Processing . vol. 6, no.7, July 1997. "The Retinex method first estimates the lighting components using a low pass filter from the observation images. Then, the logarithm calculation is performed on the observation images and the estimated lighting components, respectively. After estimating the difference, the reflection component is estimated, and finally, the brightness range of the estimated reflection component is adjusted according to the brightness range of the display device. The process is applied to the R, G, and B components of the color image, respectively. It can be applied to a color representation that affects only the reflection components without being affected by the lighting, but this method changes colors when R, G, and B are irrelevant, so that when the estimated lighting components are not accurate If the difference between the brightness of each of R, G, and B and the estimated brightness is constant, the color shifts to gray. This shape may be referred to as “R. Kimmel, M. Elad, D. Shaked, R. Keshet, and I. Sobel, A variational framework for retinex, Int. J. Comput. Vision , vol. 52, no. 1, pp. 7-23, Jan. 2003. "

delete

In addition, since the observed image changes abruptly at the boundary, but the illumination component estimated through the low pass filter changes slowly at the boundary, the reflection component estimated by the difference between the two images is near the boundary as shown in the example of 500 of FIG. 5C. Dark areas become darker, and bright areas become halo effects. On the other hand, the Retinex method is a low pass filter, and when the filter tap is short, it is estimated as a local lighting component. Therefore, the halo effect is limited to a narrow range, and the detail of the image is emphasized, but the overall image of the image is emphasized. Contrast is not well represented. On the contrary, when a filter having a long length of a filter tap is used, the overall lighting component is estimated because the overall lighting component is estimated, and the overall contrast of the image is well expressed, but the detailed expression is not emphasized. And using only the output of one low pass filter depending on the number of filters used. "Z. Rahman, D. Jobson, and GA Woodell," Properties and performance of a center / surround retinex, I EEE Trans. Image Processing: Special Issue on Color Processing , vol. 6, pp. 451-462, no. 3, Mar. SSR, as described in 1997. "DJ Jobson, Z. Rahman, and GA Woodell," A multi-scale retinex for bridging the gap between color images and the human observation of scenes, IEEE Trans. Image Processing . Vol. 6, no.7, July 1997 "can be divided into MSR (Multi-Scale Retinex) using the output of a number of low-band filter, the conventional SSR as shown in Figure 1 Suppose that each component Ｕ _i (x, y) for the input RGB color image can be expressed by the image generation model as shown in Equation (1) for the color image enhancement.

_I (x, y) = I _i (x, y) -R _i (x, y), i∈ {Ｒ, ∈, Ｂ} .... (1)

Where I _i (x, y) and R _i (x, y) represent the illumination and reflection components, respectively. At this time, the lighting component I _i (x, y) is estimated using the Gaussian function low pass filter as in the following equation (2).

는 추정된 조명성분 *는 콘볼루션(Convolution) 연산자,here

Is the estimated lighting component * is the convolution operator,

Ｇ (x, y) is a low pass filter function expressed as

Where c is the Gaussian Surround Space Constant and K is the normalization constant that satisfies Equation (4):

∬ F (x, y) dxdy = 1 ....... (4)

Then, the reflection components are estimated by taking the logarithms in [Equation 1] and [Equation 2] as follows and calculating the difference.

는 완만하게 변화되기 때문에 두 영상의 차에 의해 추정된 반사성분은 에지 부근에서 관측 영상이 어두운 곳은 더 어둡게 되고, 밝은 곳은 더 밝게 되는 후광 효과가 발생한다. 마지막으로 다음과 (6)식과 같은 게인 및 옵 셋 조정을 통하여 반사성분의 밝기 범위를 디스플레이 장치의 밝기범위에 맞게 조정하여 출력 영상을 얻게 된다.In this case, the input observed image Ｕ _i (x, y) changes abruptly at the edges but is estimated by the low pass filter of Gaussian function.

Since the gradual change is caused, the reflection component estimated by the difference between the two images causes a halo effect in which the observed image becomes darker in the dark area and brighter in the bright area near the edge. Finally, through the gain and offset adjustment as shown in the following equation (6), the brightness range of the reflection component is adjusted according to the brightness range of the display device to obtain the output image.

-필터를 이용하는데, 이 경우 필터 창의 내부의 중심에 위치하는 밝기 값과 일정 크기 이상 차이가 나는 곳의 밝기 값이 중심의 밝기 값으로 대체하는 저역 통과 필터의 특성을 갖게 되는 데, 그러나 이 필터의 경우도 클리핑을 위한 이론적 근거가 부족하며, 밝기 값의 크기가 역전될 수 있는 현상이 발생되는 단점이 있다. Where O _i (x, y) represents the output image, and a and b represent constants related to the gain and offset adjustment. In Equation 6, the output image is greatly influenced by the values of constants a and b that are determined manually. MSR achieves the effect of adding the image of the short length and the image of the long length, and expresses the overall contrast of the image well and simultaneously satisfies the detailed expression emphasis, and also the halo generated by using the low pass filter The effect can be somewhat mitigated, but fundamentally it does not completely suppress the halo effect. As a method for suppressing such a halo effect, "M. Ogata, T. Tsuchiya, T. Kubozono and K. Ueda, Dynamic range compression based on illumination compensation, IEEE Trans. Consumer Electronics , vol. 47, no. 3, Aug. Proposed in 2001. "

In this case, the filter has the characteristics of a low pass filter in which the brightness value at the center of the inside of the filter window is different from the brightness value of a certain size by replacing it with the brightness value at the center. In case of, the theoretical basis for clipping is insufficient, and the phenomenon that the magnitude of the brightness value is reversed may occur.

Accordingly, an object of the present invention is to provide an image enhancement method using a JSR-based low pass filter for suppressing a halo effect based on an SSR using an existing single filter.

      Another object of the present invention is to provide a color image enhancement method using JSR based SSR.

      It is still another object of the present invention to provide an image enhancement method based on an SSR using an existing filter but using a JND based low band filter to suppress a halo effect.

It is still another object of the present invention to provide a method and system for improving brightness and saturation while maintaining the color intact under the assumption that the H S V color image is a white light illumination state such as sunlight.

It is still another object of the present invention to provide a method and system for emphasizing a detailed representation of an image and displaying a dark part brightly.

The present invention for carrying out the above object is a setting process for setting a white lighting state equal to sunlight for color image enhancement, and receives the input image while maintaining the color (H) as it is, saturation (S) and brightness (V) RGB / HSV conversion process of converting RGB coordinate system to HSV coordinate system to improve the performance, and illumination estimation process of estimating illumination component to estimate reflection components using less change in brightness (V). The reflection component estimation process reduces the ratio of highlighting unnatural detail by the reflection component by removing the influence of the illumination of image quality degraded by the illumination and removing only a part of the estimated illumination component, and outputting a range of brightness values of the image. The histogram correction process for adjusting the brightness value of the device, and the saturation in proportion to the ratio of the brightness of the output device from the histogram correction value. Consists of a saturation enhancement control process and, HSV / RGB conversion for converting the HSV coordinate system to the RGB coordinate system to the same color coordinate system of the output image and the output device for group.

The illumination estimation process includes a process of not using a place where the difference in brightness value from the center pixel is greater than a predetermined ratio in the window in the low pass filter for the center pixel based on JND. By repeating the filter, the filter tap is repeatedly applied to the filter tap so that the interval of the filter tap is widened by a predetermined multiple. The JND-based low pass filter is configured to include a process of preserving a large change in the brightness value, such as a boundary, to the original brightness value and preventing a halo effect from the estimated reflection component using the estimated illumination component. do.

The reflection component estimating process is a product of a reflection component and an illumination component of an input image, separating the illumination and the reflection component among the original brightness of the input image and the brightness through the JND based low pass filter, and the JND based low pass. It includes a process of removing some of the estimated illumination components generated while passing through the filter, the separation of the illumination and reflection components is configured to be achieved by a log operation.

According to another aspect of the present invention, there is provided an illumination component process for estimating an illumination component using a JND based low pass filter, and a reflection component estimation for estimating a reflection component using a retinex method. In order to compensate for the unnatural appearance of the reproduced image by the estimated reflection component due to the incorrectly estimated illumination component from the reflection component estimation value, only a part of the estimated lighting component is removed from the reproduced image to produce uneven illumination. An illumination component removal process of obtaining a compensated image, a histogram correction process of correcting a histogram to match a range of brightness values from the image of which the non-uniform illumination is compensated to a range of brightness values of an output device, and the output from the histogram correction value Improve saturation in proportion to the ratio of the brightness value of the device Saturation consists of a control process for improving.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, specific details such as specific components are shown, which are provided to help a more general understanding of the present invention, and it is understood that these specific details may be modified or changed within the scope of the present invention. It will be obvious to those of ordinary skill in the field. Hereinafter, the same reference numerals in the accompanying drawings will be found to have similar or identical functions.

3 is a diagram for improving image quality of a color image using JND-based SSR for an embodiment of the present invention.

When the RGB color image is input to the RGB / HSV converter 201 through the image input terminal 101, the color image is converted into a color image of hue (H), saturation (S), and brightness (V), and then the color (H) is assumed to be a white light illumination state. ) And the brightness (V) and saturation (S) are improved. In the improvement of the brightness V, the JND-based low pass filter 301 is used to estimate the illumination component by weak low pass filtering as the value having a value larger than the value of the center pixel in the filter window. Later, when estimating the reflection component, the halo effect generated by the reflection component can be reduced. Here, the JND-based low pass filter 301 applies the one-dimensional filter in the vertical direction to the image to which the filtering is applied in the horizontal direction as the one-dimensional filter. The filter tap is designed to filter the low band at high speed by repeatedly applying a filter in which the interval between the filter taps is doubled every time it is repeated using a short filter. The brightness (V) of the output of the RGB / HSV converter 201 takes log (303, 305) operations to separate the illumination and reflection components of the estimated output through the JND based low pass filter (301), and the log (305). The amount of illumination component that is input to the illumination component removal stage 307 of the estimated illumination component generated while passing through the JND-based low pass filter 301 among the estimated illumination and reflection components of the calculated output is input to the illumination component removal stage 307. Removed in accordance with (a), the adder 309 eventually finds a difference from each value (+,-). The difference generated here is the estimated reflection component of the present invention. The histogram correction section 311 is then used to adjust the range of brightness values at the output device. On the other hand, the saturation (S) of the output of the RGB / HSV converter 201 is improved by the saturation enhancement control unit 315 in proportion to the ratio of the brightness (V) past the brightness gain control unit 313, the last The improved values and hue (H) of brightness (V) and saturation (S) are converted from the HSV color image to the RGB color image by the HSV / RGB converter 203 to use the same color coordinate system as the output device. do.
The example of FIG. 3 will be described in detail with reference to the flowchart of FIG. 4.

delete

First, as shown in the example of 401, the white light state is set as an initial setting condition such as sunlight, and the input image signal of step 403 inputted to the image input terminal 101 is as follows.

Ｕ _i (x, y), i∈ {Ｒ, Ｇ, Ｂ}
i = icomponent of the input image

In the RGB / HSV converter 201 of FIG. 3, the RGB color coordinate system is converted into the HSV color coordinate system. The reason for the conversion is white illumination, so the saturation (S; saturation) is maintained as it is. And brightness (V; value) components only, and details of this method are disclosed in "RWG Hunt. Measuring Color. Ellis Horwood Series in Applied Science and Industrial Technology, Halsted Press, New York, NY, 1989." Doing.

A portion 400 shown in FIG. 4 is a JND-based low pass filtering portion for estimating an illumination component. The brightness {Ｕv (x, y)} and the saturation {, s (x, y)} of the RGB / HSV converter 201 are shown. To improve the brightness (V) and the saturation (S) in the color {Ｕh (x, y)}. In step 407, step 413, since the lighting component has little change at the original brightness, the lighting component is estimated through the JND-based low pass filter 301. In the next step, the lighting component is reflected. Estimate the component. To this end, the low-band filter is performed at high speed. In this case, the low-band filter should not be used where the difference in brightness value from the center pixel is higher than a predetermined ratio within the low-band filter window for the center pixel. The coefficients of the JND based low pass filter 301 are performed separately in the horizontal and vertical directions as e (m, n) = e (m) .e (n), and e (m) and e (n) High-speed low-band filtering is performed by repeatedly applying a filter that the interval between filter taps is doubled every time, using a short tap filter that is all (1/4, 2/4, 1/4). Be sure to And the thresholds used for filtering and the values that emphasize the detail expressions are experimentally up to some extent to JND.
The value used for the image, the discarding value of the JND minimum or less, and the removal value of some lighting components generated during the JND filtering are determined according to the desired image quality. To this end, in step 407, the variable k that is stored to know the repeated number of filters is initialized. In step 409, when the predetermined number of times is exceeded, the reflection component estimation process of the next step is performed. According to Weber's law, the minimum JND of the local brightness change that can detect the change of the brightness value in the uniform image is known to be approximated as shown in [Equation 7].

Where I represents the uniform brightness of the image, JND (I) represents the JND value for brightness I, and c and d are constants. If I is somewhat bright, c is ignored, and the ratio of I and JND is expressed as a constant value in [Equation 8].

에서의 밝기 Ｕ_V(x,y)와

에서

만큼 떨어진 곳의 밝기 Ｕ_V(x-m, y-n)의 차 △Ｕ_V(m,n)이 중심 화소의 JND 값보다 큰 경우에는 밝기 값의 차를 감지할 수 있다. 본 발명에서는 이러한 JND 특성을 이용하여 중심 화소에 대한 저역 통과 필터 창 내에서 중심 화소와의 밝기 값의 차를 감지할 수 있는 화소는 저역 통과 필터링에 사용하지 않도록 하여 원래의 밝기 값을 보존하는 필터링을 수행한다. 이러한 조건부 필터링을 위하여 다음과 같은 중심 화소

의 JND 값 JND(Ｕ_v(x,y))와 밝기의 차 △Ｕ_V(m,n)의 비 J(m,n)을 정의한다. Using the JND model of Equation 7, the center pixel in the filter window to which the low band filter is applied

Brightness at Ｕ _V (x, y) and

in

When the difference ΔＵ _V (m, n) of the brightness Ｕ _V (xm, yn) at a distance is greater than the JND value of the center pixel, the difference of the brightness value may be detected. In the present invention, the pixel that can detect the difference between the brightness value and the center pixel in the low pass filter window for the center pixel by using the JND characteristic is not used for low pass filtering to preserve the original brightness value. Do this. For this conditional filtering, the following center pixel

A defines the JND value _{JND (U v (x, y} )) and the ratio J (m, n) of the difference △ U _V (m, n) of the brightness.

와 비교한다. 즉 중심 화소

에 대한 저대역 필터 창 내에서 J(m,n)이

보다 큰 화소는 중심 화소

의 저대역 필터링에 사용하지 않는다. 그리고 J(m,n)이

보다 작은 화소는 J(m,n)값에 따라 저역 통과 필터링의 강약을 조절하여 필터링된다. J (m, n): the ratio of the JND value JND (U _V (x, y) and the brightness difference △ U _V (m, n) of the center pixel (x, y)
In this case, when J (m, n) is greater than 1, it becomes a brightness difference that ΔV _V (m, n) can detect. In the present invention, J (m, n) is not 1 to improve the filter performance.

Compare with Center pixel

J (m, n) within the low pass filter window for

The larger pixel is the center pixel

Do not use for low-band filtering of. And J (m, n)

Smaller pixels are filtered by adjusting the strength of the low pass filtering in accordance with the value of J (m, n).

delete

The illumination component estimation of brightness using the JND based low band filter of the present invention is expressed as follows.

: 추정된 밝기의 조명 성분

: Lighting component of estimated brightness

e (m, n): low band filter coefficients

W: filter window

는 추정된 밝기의 조명 성분, e(m,n)은 저대역 필터 계수, W는 필터 창을 나타내고, λ(m,n)은 J(m,n)의 값에 따라 저대역 필터링의 강약 조절을 위한 함수로 다음(11식)과 같이 표현된다. 즉, 도 6의 예와 같이 밝기의 화상중 필터창 내에서 화소

에 위치한 중심 화소 보다

만큼 떨어진 화소에 위치한 밝기값과 중심화소의 밝기 값과의 비[ J(m,n)]와 저역 통과 필터의 강약 조절을 위한 함수 값[λ(m,n)]의 관계로부터 상기 중심화소

에서

만큼 떨어진 화소의 밝기값 과 중심 화소의 밝기 값의 비 [J(m,n)]가 높으면 높을수록 저 대역 강약조절을 약하게 하여 선형(601)이 아닌 비선형(603)의 특성으로 나타나게 하는 필터링 방법에 의해 조명성분을 추정한다. 이특성은 중심화소의 값보다 큰값을 가질수록 약한 저대역를 사용하기 위한 것이며,이때 중심 화소의 값보다 큰 값을 가지는 화소의 밝기 값은 더 잘 유지할 수 있으며, 다음 반사 성분을 추정할 때 반사 성분에서 발생하는 후광 현상은 줄일 수 있다.here

Is the lighting component of the estimated brightness, e (m, n) is the low-pass filter coefficient, W is the filter window, and λ (m, n) is the intensity control of low-band filtering according to the value of J (m, n). The function for is expressed as That is, the pixel in the filter window of the image of the brightness as shown in the example of FIG.

Than located at the center pixel

The center pixel from the relationship between the ratio of the brightness value located at the pixel spaced apart from the brightness value of the center pixel [J (m, n)] and the function value [λ (m, n)] for the strength adjustment of the low pass filter.

in

The higher the ratio [J (m, n)] between the brightness value of the pixel and the brightness value of the center pixel, the higher the weaker the low-band intensity control, the filtering method that appears as a nonlinear 603 rather than a linear 601. The lighting component is estimated by This characteristic is to use the weaker band when the value is larger than the value of the center pixel, and the brightness value of the pixel having a value larger than the value of the center pixel can be maintained better. The halo phenomenon that occurs can be reduced.

T _max : J (m, n) maximum threshold

에 위치한 중심 화소보다

만큼 떨어진 화소에 위치한 밝기 값과 중심 화소의 밝기 값과의 비T _min : J (m, n) minimum threshold
J (m, n): Pixels in the filter window among images of brightness

Than the center pixel located in

The ratio of the brightness value located at pixels apart by the brightness value of the center pixel

When the desired number of times of repetition in step 409 is exceeded, the JND based low pass filtering is completed, and the reflection estimation process of step 415 is performed, and the reflection component of the illumination image estimated from the input image is estimated. In order to improve the image quality of the image by removing the influence of light on the image of image quality degraded by uneven illumination, the image composed of only the reflection component becomes unnatural with only the detailed expression emphasized. To compensate for this, some of the estimated lighting components are removed to reduce the emphasis on detail. For this purpose, the illumination and reflection of the original brightness are separated by log 303 association, and the estimated illumination and reflection through the JND based low pass filter 301 are separated through log 305 operation and subjected to JND filtering. For some of the lighting components generated as a result, the removal value A is modeled as a product in the skeleton 308. In other words, logarithmic operations 303 and 305 are performed to separate reflection and illumination components from the brightness of the input image and the brightness through the JND-based low pass filter, and then removed from the skeleton 308 for some illumination components. By obtaining the difference in the adder 309, the reflection component is estimated for the enhanced brightness image as shown in Equation (12).

: 향상된 밝기 영상

: Improved brightness image

는 향상된 밝기 영상을 나타내고, a는 조명 성분의 일부만을 제거하기 위한 상수이다.a: constant to remove only a couple of lighting components
U: original video
I: Image applied through low pass filter
Q: reflection image
v: brightness
here

Represents an improved brightness image, and a is a constant for removing only a part of the lighting component.

delete

의 히스토그램의 분포를 자동으로 변화시켜 출력 장치의 밝기 값의 범위에 맞게 자동으로 조절하기 위해 히스토그램의 수정을 하게 된다. 상기 히스토그램 수정법에서 변환 함수 z(k)는 영상의 변환된 누적분포함수로서 조절 계수

의 값에 따라 다음(13)식과 같이 달라지게 된다.Finally, in step 417, the histogram correction unit 311 of FIG. 3 is performed as described in "Reference: AK Jain, Fundamentals of Digital Image Processing , Prentice-Hall, 1989."

The histogram is modified to automatically adjust the distribution of the histogram to match the range of brightness values of the output device. In the histogram correction method, a transform function z (k) is a control coefficient as a transformed cumulative distribution function of an image.

It depends on the value of (13).

영상의 변환된 누적 분포 함수.z (k): v

Transformed cumulative distribution function of the image.

: k번째 밝기 레벨의 밝기 값.

: The brightness value of the kth brightness level.

L-1: Maximum level of brightness value.

의 범위를 가짐.x: by the control coefficient

Has a range of

If x = 1: Same effect as applying histogram equalization.

의 히스토그램에 선형 히스토그램 스트레칭을 적용한 것과 같은 효과.
P: 화소가 존재할 확률
여기서

는 k번째 밝기 레벨의 밝기 값을 나타내고, L-1은 밝기 값의 최대 레벨을 나타낸다. 이때 x 는

의 범위에서 실험에 근거하여 x=1인 경우 히스토그램 등활화를 적용한 것과 같은 효과가 나타나고, x=0인 경우

의 히스토그램에 선형 히스토그램 스트레칭을 적용한 것과 같은 효과가 나타난다. 일반적으로 채도 향상은 높은 채도를 가지는 영상은 낮은 채도를 가지는 영상에 비하여 더 선명하게 보이므로 채도를 향상하여 영상의 선명도를 향상시키게 된다. 이때 대부분의 영상에서 밝기 값이 어두운 부분의 채도 값이 작다고 가정하여 밝기의 비율에 의존하여 채도를 향상시키면 더욱 선명한 영상을 획득하는데,(419)과정에서 밝기 이득 조절부(313)에서 먼저 다음 (14)식과 같은 밝기의 향상된 비율 g(x,y)를 계산한다.If x = 0:

Equivalent to applying linear histogram stretching to a histogram.
P: probability of existence of a pixel
here

Denotes the brightness value of the k-th brightness level, and L-1 denotes the maximum level of the brightness value. Where x is

In the range of, based on the experiment, x = 1 shows the same effect as applying histogram equalization, and x = 0

The effect is the same as applying linear histogram stretching to the histogram of. In general, the saturation enhancement is more sharp than the image having a high saturation, so that the saturation is improved to improve the saturation of the image. At this time, assuming that the saturation value of the dark part of the brightness value is small in most images, if the saturation is improved depending on the ratio of brightness, a clearer image is obtained. In step 419, the brightness gain control unit 313 first performs the following ( Calculate the improved ratio g (x, y) of brightness as

delete

g (x, y): Improved ratio of brightness.
O: output video
U: input video
v: brightness

Thereafter, as shown in Equation 15, the saturation enhancement controller 315 improves the saturation in proportion to the improved ratio of the brightness as follows.

_S (x, y): Improved saturation image

δ: constant that determines the amount of saturation enhancement
Us: Saturation (s) value of input image
g: Ratio of brightness (Ov / Hv; Ov: brightness of output image, Hv: brightness of color)
Where _S (x, y) represents the enhanced saturation image, and δ is a constant that determines the magnitude of the saturation enhancement. The saturation (s) and the brightness (V) are improved values such as hue (H) in the HSV / RGB converter 203 converts the HSV color coordinate system to the RGB color coordinate system in step 421, the reason is the output image And to use the same color coordinate system as the output device, similar to the example disclosed in RWG Hunt. Measuring Color.Ellis Horwood Series in Applied Science and Industrial Technology, Halsted Press, New York, NY, 1989. The output image form output to the image output terminal 112 is output in an RGB (Red, Green, Blue) color coordinate system.
Therefore, the present invention shows an image having a strong edge in the center of Figure 5a and a dark and weak edge on both sides and the resultant images. In the result image of the histogram equalization method of FIG. Although the overall contrast is well expressed, it can be seen that the dark parts of the observed image do not look bright and do not emphasize the detailed representation of the tree. As shown in the resultant image by the SSR of FIG. 5C, the darker part becomes brighter than that of FIG. 5B, and the detailed expression of the tree is also emphasized. However, the color of the sky and the part of the soil turns gray, and a wide halo effect occurs around the tower. However, in the image of the result of the method of the present invention of FIG. 5D, the overall contrast and the detailed expression of the image are emphasized, and unlike the image of FIG. 5C, the color of the image does not change and the halo effect does not appear.

delete

delete

As described above, the color image enhancement method using the JND-based SSR of the present invention is applied to the HSV color image to maintain the color and improve the brightness based on the JND-based low-band filter in proportion to the improved ratio of the saturation brightness. It has the advantage of enhancing the detailed expression of the image and showing the dark areas brightly, and shows the vivid color with improved saturation without changing the color with little halo effect near the edges.

The above-described embodiments are merely examples of implementing the present invention. The invention is disclosed by the embodiments described above, but is not limited to the embodiments. Rather, one of ordinary skill in the art would understand many improvements and alternatives applied to particular components that make up a network architecture that can be adapted to the present invention without departing from the scope of the following claims, which define the invention. There will be.

Claims (28)

 In the image quality improvement method of a color image,  The process of estimating the lighting component using the lighting and reflection components of the brightness of the color image, Compensating for a non-uniform image of brightness and the estimated illumination component of the color image; Adjusting a range of brightness values of the compensated image to a range of brightness values of an output device; And improving the saturation in proportion to the ratio of the brightness values of the output device. The method of claim 1, wherein the illumination component estimation is performed by JND-based low pass filtering using a weak low pass filter as the illumination component estimate is larger than a value of a center pixel of the filter window. The method of claim 2, wherein the JND based low pass filtering is performed by the following Equation (11).

T _max : J (m, n) maximum threshold T _min : J (m, n) minimum threshold J (m, n): Center pixel within the filter window of images with brightness

see

Is the ratio of the brightness value of the pixel delete The method of claim 1, wherein the compensation of the non-uniform image comprises taking a log of each of the original brightness (V) and the estimated illumination component; And estimating a reflection component by obtaining a difference between the logarithmic values. The method of claim 5, wherein the non-uniform image is compensated by Equation (12).

: Improved brightness image a: constant to remove only a couple of lighting components U: original video I: Low Pass Filter (LPF) v: brightness delete The method of claim 1, wherein a range of brightness values of the output device is adjusted according to the brightness according to Equation (13).

z (k): v

Transformed cumulative distribution function of the image

: brightness value of the kth brightness level L-1: Maximum level of brightness value x: by the control coefficient

Has a range of If x = 1: same effect with histogram equalization If x = 0:

Equivalent to applying linear stretching to the histogram of P: Probability of Pixel The method of claim 1, wherein the ratio of the brightness value of the output device is achieved by the following Equation (14).

g (x, y): improved proportion of brightness O: output video U: input video v: brightness The method of claim 1, wherein the saturation (s) is proportional to a ratio of brightness values of the output device in accordance with Equation (15).

_S (x, y): Improved saturation image δ: constant that determines the amount of saturation enhancement Us: Saturation (s) value of input image g: Ratio of brightness (Ov / Hv; Ov: Output video brightness, Hv; Color brightness) In the image quality improvement method of color image of hue (h), brightness (v), saturation (s), An illumination component process of estimating an illumination component using a JND-based low band filter using weak low pass filtering as the value having a value larger than the center pixel of the filter window is improved to improve brightness; A reflection component estimating process of estimating a reflection component by a retinex method to the illumination component estimation value; In order to compensate for the unnatural appearance of the reproduced image by the estimated reflection component due to the incorrectly estimated illumination component from the reflection component estimation value, the illumination is obtained by removing the estimated illumination component from the reproduced image to obtain an image in which non-uniform illumination is compensated. Component removal process,  A histogram correction process of correcting the histogram to adjust a range of brightness values of the image from the image compensated for the non-uniform illumination to a range of brightness values of the output device; And a saturation enhancement control process for improving saturation in proportion to the improved ratio of the brightness from the histogram correction value. In the color image quality improvement method, A setting process of setting a white lighting state such as sunlight for improving a color image, An RGB / HSV conversion process of converting an RGB coordinate system into an HSV coordinate system in order to improve saturation (S) and brightness (V) while maintaining the color (H) while receiving an input image; An illumination estimation process of estimating an illumination component in order to estimate a reflection component in an illumination component having a small change by a low pass filtering using a weak low pass filter as the brightness V has a value greater than a center pixel of the filter window; A reflection component estimating process that removes the influence of illumination of image quality deteriorated by uneven illumination and reduces the ratio of emphasis of unnaturally detailed expressions by reflection components from estimated illumination components; A histogram correction process for adjusting the brightness value range of the image to the brightness value range of the output device; A saturation enhancement process for improving saturation in proportion to the improved ratio of the brightness from the histogram correction value; A method for improving the image quality of a color image, comprising: an HSV / RGB conversion process of converting an HSV coordinate system into an RGB coordinate system in order to use the same color coordinate system as the output device. The method of claim 12, The illumination estimation process improves the image quality of a color image, characterized in that the difference in brightness value with the center pixel is not used in the window in the JND-based low pass filter for the center pixel based on JND. Way. The method of claim 13, The JND-based low pass filter is configured to perform fast filtering by repeatedly applying a filter in which the interval between the filter taps is increased by n times each time using a short length filter. Way. The method of claim 13, The JND-based low pass filter has a larger value than that of the center pixel, and thus, a portion having a large change in brightness value, such as a boundary, by the low pass filtering using a weak low pass filter preserves the original brightness value to obtain an estimated illumination component. And a halo effect does not occur in the reflection component estimated by using the method. The method of claim 13, The reflection component estimating process is performed by multiplying a reflection component and an illumination component of an input image by performing a logarithm operation on the input image and the estimated lighting component, respectively; And estimating a reflection component from the difference between the logarithmic values and eventually removing the estimated lighting component. An RGB / HSV converter that converts the R, G, B coordinate system into a hue (H), saturation (S), and brightness (V) coordinate system, Filtering by using a weak low pass filter as the RGB / HSV converter has a value greater than the value of the center pixel of the hue (H), saturation (S), and brightness (V) of the output (V). JND based low pass filter for estimating the lighting First and second logarithm calculation units configured to separate the illumination and reflection components of the output of the JND based low pass filter and the output of the RGB / HSV converter; Illumination that compensates for non-uniform illumination by removing the estimated illumination component from the estimated reflection and illumination component separated by the second logarithm calculation unit and calculating a difference between the separated illumination and the reflection component value from the brightness of the original image of the first logarithm calculation unit. Compensation Department, A histogram correction unit for adjusting an output of the lighting compensator to a range of brightness values in an output device; A brightness gain control unit for adjusting a gain in proportion to the saturation S of the output of the RGB / HSV converter in proportion to an improved ratio of the brightness V of the output of the histogram correction unit; A saturation enhancement control unit for improving saturation according to an improved ratio of brightness of the brightness gain control unit; And a HSV / RGB converting unit for converting the brightness of the histogram correcting unit and the improved saturation of the chroma enhancement control unit and the output of the original color (H) of the RGB / HSV converter to use the same color coordinate system as the image of the output device. A system for improving image quality of color images. In the image quality improvement method of color image of hue (h), brightness (v), saturation (s), The center pixel of the image of the brightness of the non-uniform color image of the hue (h), brightness (v), saturation (s) in

The center pixel from the relationship between the ratio of the brightness value of the pixel to the brightness of the center pixel [J (m, n)] and the function [λ (m, n)] for low / low intensity control

Estimating the lighting component by filtering using a weak low pass filter, Compensating for the non-uniform image by the original brightness and the estimated reflection component; Matching the range of brightness values to the range of brightness values of an output device; And improving the saturation (s) in proportion to the ratio of the enhanced brightness. delete delete delete delete delete delete delete delete delete delete

Images