CN113379778A - Image purple boundary detection method based on content self-adaptive threshold - Google Patents

Abstract

The invention discloses an image purple boundary detection method based on a content self-adaptive threshold, which comprises the following steps: subtracting the local minimum value from the local maximum value of the image to obtain a gray level image, and obtaining a mask binary image by a self-adaptive threshold value method; converting the RGB image into a YCbCr image, simultaneously obtaining a first quadrant pixel, performing Gaussian weighted convolution operation on the YCbCr brightness component Yn image, and calculating the gradient size of each point in the Yn image; setting a self-adaptive threshold according to the gradient value and the standard deviation, and selecting a high-contrast pixel value; calculating a norm ratio according to the chromatic value of the high-contrast pixel; selecting a binary area according to the constrained decision domain; and the obtained mask binary image and the binary area are subjected to AND operation to obtain a fine purple fringing detection area.

Description

Image purple boundary detection method based on content self-adaptive threshold

Technical Field

The invention relates to an image detection technology, in particular to an image purple boundary detection method based on a content self-adaptive threshold value.

Background

When a camera takes a picture in a backlight mode, purple edges occur at the edges of objects in the picture, namely, purple stripes are left near the edges with high contrast by some cameras during image acquisition, and in order to detect purple edge areas, a PFA detection method is the most commonly used method for detecting chromatic aberration. The PFA method firstly identifies a high-contrast neighborhood and then applies coloring constraint to detect the purple stripe pixels. At present, the purple fringing detection technology comprises the following methods: the shading constraint is exactly the comparison of the R, G, B channels for a single pixel. Since the position of the purple color is not clearly defined at the time of streak detection, the definition of the purple color is blurred. This may lead to false positive results. These shading constraints can be deployed in any color space, and need not be limited to RGB; to further mitigate the effects of variations in the composition of the light source, a shading constraint is placed in CIExy space. However, this brightness normalization does not work completely; by converting RGB into YCbCr space, complete independence from the luminance space is achieved, where the chrominance channels have exclusive color cast information, opening up the idea of specifying an area in the chrominance space, where the violet concentration is the largest. While angle constraints are utilized to correct for edge aberrations, they do not utilize gradient constraints to detect edge pixels; in the existing PFA detection method, the intermediate threshold for separating PFA regions is absolute. Given the diversity of camera devices and their drawbacks, these approaches are unlikely to cover all forms of purple edges. In summary, the problems of the current purple fringing experiment are as follows: 1. the detection result is accurate, but the detection range is narrow, and a large amount of missed detections exist. 2. The detection result is rough, the detection range is wide, but a large amount of false detection exists. 3. A problem with the RGB space is the drift associated with the intensity variations of the high contrast areas, which causes a shift in the fringe-rendering pattern.

Disclosure of Invention

According to the problems in the prior art, the invention discloses an image purple boundary detection method based on a content self-adaptive threshold, which comprises the following specific steps:

subtracting the local minimum filtering from the local maximum filtering of the image to obtain a gray level image, and obtaining a mask binary image by a self-adaptive threshold method;

converting the RGB image into a YCbCr image, simultaneously obtaining a first quadrant pixel, performing Gaussian weighted convolution operation on the YCbCr brightness component Yn image, and calculating the gradient size of each point in the Yn image;

setting a self-adaptive threshold according to the gradient value and the standard deviation, and selecting a high-contrast pixel value;

calculating a norm ratio according to the chromatic value of the high-contrast pixel;

selecting a binary area according to the constrained decision domain;

and the obtained mask binary image and the binary area are subjected to AND operation to obtain a fine purple fringing detection area.

Further, converting the RGB palette into YCbCr when converting the RGB image into the YCbCr image normalizes the values of luminance components Yn, Cbn and Crn using the following formulas,

wherein R is_n,G_n,B_n∈[0,1]；Y_n∈[0,1],C_bn∈[—0.5,0.5],C_rn∈[-0.5,0.5]High contrast regions are detected using the gradient information, and gaussian-weighted X and Y gradients are calculated at each pixel of Yn.

Further, when selecting the high contrast pixel value: assuming that the GREY-SCALE image is represented by a function Yn (x, y), the results of the convolution with horizontal and vertical Gaussian-weighted gradient kernels, respectively, are as follows:

Y_h(x,y)＝Y_n(x,y)*G_x(x,y)

Y_v(x,y)＝Y_n(x,y)*G_v(x,y),

Y_v(x,y),Y_h(X, y) are horizontal and vertical Gaussian weighted gradient kernel convolutions, respectively, at one point (X)_i，Y_i) The magnitude of the upper estimated gradient is calculated from the following equation:

calculating the average gradient size (x) of all pixel points_i,y_i)(μ_grad) And standard deviation (σ)_grad),α∈[0,3]For setting a threshold parameter for selecting the maximum gradient, the global gradient threshold amplitude being set

T_grad＝μ_grad+ασ

Alpha is greater than or equal to 0.5, high contrast pixel positions are selected, and by calculating smooth discrete derivatives, the Gaussian weighting process ensures that striations and texture patterns do not occur in the gradient selection, where (x)_i,y_i) Each pixel in a location satisfies the following equation:

M_Y(x_i,y_i)＞T_grad

are considered to be high contrast pixels.

Further, the high contrast pixel is represented as (x)_h(i),y_h(i))；i＝1，2…N_hIn which N is_hIs the total number of detected high contrast pixels, a w × w window is constructed around each high contrast pixel, w ═ 5, for the position (x)_h(i),y_h(i)) Scanning all pixels in the window, the pixels being in coordinates (x)_h(i)+p,y_h(i)+ q) denotes p, q ∈ 0, ± (w-1)/2 · all these w²The chromaticity value at a position is calculated as,

C_b(p,q)＝C_bn(x_h(i)+p,y_h(i)+q)

C_r(p,q)＝C_rn(x_h(i)+p,y_h(i)+q)

form w composed of these chrominance components²Vector sequence

For each vector X_p,qCalculating the norm ratio:

wherein the content of the first and second substances,

decision domain of constraint: rho_p,q＜T_PFANamely selecting the purple boundary area.

Due to the adoption of the technical scheme, the method for detecting the purple fringing of the image based on the content self-adaptive threshold is different from the traditional detection learning method, and the provided method can break through the traditional experience threshold assumption, namely the statistical data of the image edge does not need to be acquired, so that the method has the advantages of more accurate detection result, capability of avoiding the omission and the false detection of the purple fringing and capability of accurately identifying the purple fringing region.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative efforts.

FIG. 1 is a flow chart of the method disclosed herein.

Detailed Description

In order to make the technical solutions and advantages of the present invention clearer, the following describes the technical solutions in the embodiments of the present invention clearly and completely with reference to the drawings in the embodiments of the present invention:

as shown in fig. 1, the method for detecting purple boundary of an image based on a content adaptive threshold specifically includes the following steps:

subtracting the local minimum value from the local maximum value of the image to obtain a gray level image, and obtaining a mask binary image by a self-adaptive threshold value method;

converting the RGB image into a YCbCr image, simultaneously obtaining a first quadrant pixel, performing Gaussian weighted convolution operation on the YCbCr brightness component Yn image, and calculating the gradient size of each point in the Yn image;

setting a self-adaptive threshold according to the gradient value and the standard deviation, and selecting a high-contrast pixel value;

calculating a norm ratio according to the chromatic value of the high-contrast pixel;

selecting a binary area according to the constrained decision domain;

and the obtained mask binary image and the binary area are subjected to AND operation to obtain a fine purple fringing detection area.

Further, converting the RGB palette into YCbCr when converting the RGB image into the YCbCr image normalizes the values of luminance components Yn, Cbn and Crn using the following formulas,

wherein R is_n,G_n,B_n∈[0,1]；Y_n∈[0,1],C_bn∈[-0.5,0.5],C_rn∈[-0.5,0.5]Detecting a high contrast region by using gradient information, and calculating a Gaussian-weighted X and Y gradient on each pixel of Yn;

further, assuming that the green-SCALE image is represented by a function Yn (x, y), the result of convolution with horizontal and vertical gaussian-weighted gradient kernels, respectively, is as follows:

Y_h(x,y)＝Y_n(x,y)*G_x(x,y)

Y_v(x,y)＝Y_n(x,y)*G_v(x,y),

Y_v(x,y),Y_h(X, y) are horizontal and vertical Gaussian weighted gradient kernel convolutions, respectively, at one point (X)_i，Y_i) The magnitude of the upper estimated gradient is calculated from the following equation:

calculating the average gradient size (x) of all pixel points_i,y_i)(μ_grad) And standard deviation (σ)_grad),α∈[0,3]For setting a threshold parameter for selecting the maximum gradient, the global gradient threshold amplitude being set

T_grad＝μ_grad+ασ

Alpha is greater than or equal to 0.5, high contrast pixel positions are selected, and by calculating smooth discrete derivatives, the Gaussian weighting process ensures that striations and texture patterns do not occur in the gradient selection, where (x)_i,y_i) Each pixel in a location satisfies the following equation:

M_Y(x_i,y_i)＞T_grad

are considered to be high contrast pixels.

Further, the high contrast pixel is represented as (x)_h(i),y_h(i))；i＝1，2…N_hIn which N is_hIs the total number of detected high contrast pixels, a w × w window is constructed around each high contrast pixel, w ═ 5, for the position (x)_h(i),y_h(i)) Scanning all pixels in the window, the pixels being in coordinates (x)_h(i)+p,y_h(i)+ q) denotes p, q ∈ 0, ± (w-1)/2 · all these w²The chromaticity value at a position is calculated as,

C_b(p,q)＝C_bn(x_h(i)+p,y_h(i)+q)

C_r(p,q)＝C_rn(x_h(i)+p,y_h(i)+q)

form w composed of these chrominance components²Vector sequence

For each vector X_p,qCalculating the norm ratio:

wherein the content of the first and second substances,

decision domain of constraint: rho_p,q＜T_PFANamely selecting the purple boundary area.

The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art should be considered to be within the technical scope of the present invention, and the technical solutions and the inventive concepts thereof according to the present invention should be equivalent or changed within the scope of the present invention.

Claims (4)

1. An image purple boundary detection method based on a content self-adaptive threshold value is characterized by comprising the following steps:

subtracting the local minimum filtering from the local maximum filtering of the image to obtain a gray level image, and obtaining a mask binary image by a self-adaptive threshold method;

converting the RGB image into a YCbCr image, simultaneously obtaining a first quadrant pixel, performing Gaussian weighted convolution operation on the YCbCr brightness component Yn image, and calculating the gradient size of each point in the Yn image;

setting a self-adaptive threshold according to the gradient value and the standard deviation, and selecting a high-contrast pixel value;

calculating a norm ratio according to the chromatic value of the high-contrast pixel;

selecting a binary area according to the constrained decision domain;

and the obtained mask binary image and the binary area are subjected to AND operation to obtain a fine purple fringing detection area.

2. The method of claim 1, wherein: converting the RGB palette to YCbCr when converting the RGB image to the YCbCr image normalizes the values of luminance components Yn, Cbn and Crn using the following formulas,

wherein R is_n,G_n,B_n∈[0,1]；Y_n∈[0,1],C_bn∈[-0.5,0.5],C_rn∈[-0.5,0.5]Detection of high contrast regions using gradient informationA gaussian weighted X and Y gradient is calculated over each pixel of Yn.

3. The method of claim 1, wherein: when selecting the high contrast pixel value: assuming that the GREY-SCALE image is represented by a function Yn (x, y), the results of the convolution with horizontal and vertical Gaussian-weighted gradient kernels, respectively, are as follows:

Y_h(x,y)＝Y_n(x,y)*G_x(x,y)

Y_v(x,y)＝Y_n(x,y)*G_v(x,y),

Y_v(x,y),Y_h(X, y) are horizontal and vertical Gaussian weighted gradient kernel convolutions, respectively, at one point (X)_i，Y_i) The magnitude of the upper estimated gradient is calculated from the following equation:

calculating the average gradient size (x) of all pixel points_i,y_i)(μ_grad) And standard deviation (σ)_grad),α∈[0,3]For setting a threshold parameter for selecting the maximum gradient, the global gradient threshold amplitude being set

T_grad＝μ_grad+ασ

Alpha is greater than or equal to 0.5, high contrast pixel positions are selected, and by calculating smooth discrete derivatives, the Gaussian weighting process ensures that striations and texture patterns do not occur in the gradient selection, where (x)_i,y_i) Each pixel in a location satisfies the following equation:

M_Y(x_i,y_i)＞T_grad

are considered to be high contrast pixels.

4. The method of claim 1, wherein: high contrast pixels are denoted as (x)_h(i),y_h(i))；i＝1，2…N_hIn which N is_hIs the total number of high contrast pixels detected, at each highA w × w window is constructed around the contrast pixel, w ═ 5, for the position (x)_h(i),y_h(i)) Scanning all pixels in the window, the pixels being in coordinates (x)_h(i)+p,y_h(i)+ q) denotes p, q ∈ 0, ± (w-1)/2 · all these w²The chromaticity value at a position is calculated as,

C_b(p,q)＝C_bn(x_h(i)+p,y_h(i)+q)

C_r(p,q)＝C_rn(x_h(i)+p,y_h(i)+q)

form w composed of these chrominance components²Vector sequence

For each vector X_p,qCalculating the norm ratio:

wherein the content of the first and second substances,

decision domain of constraint: rho_p,q＜T_PFANamely selecting the purple boundary area.

Images