CN107909592A - A kind of line drawing drawing generating method for mural painting image - Google Patents

Abstract

The present invention relates to a kind of line drawing drawing generating method for mural painting image, comprise the following steps：The artwork of input is smoothly pre-processed using global variation smoothing filter, obtains smoothed image I；Smoothed image I is converted into gray level image I₀；Based on Gaussian Blur to I₀Carry out High frequency filter lifting, handling result I₂Represent；The tangential flow field in edge is constructed, recycles improved FDoG filtered methods to extract the lines figure with enhancing effect, output result is represented with H'：The characteristics of on the basis of the tangential flow field in edge is constructed, improving FDoG methods, strengthening mural painting using the lines of high boost extraction, the gray level image I of original input₀It is substituted for it and High frequency filter lifting result I₂The result I obtained after multiplication_m, define new FDoG wave filters；Processing is filtered to image H ' using median filtering technology.

Description

A Line Drawing Generation Method for Mural Images

technical field

It involves line drawing generation technology for mural images, especially smooth denoising and contour extraction technology for color mural images.

Background technique

Murals are one of my country's precious cultural heritages, and line drawing, which originated from Dunhuang murals, is not only a unique painting method of traditional Chinese painting, but also a special means of mural protection, which has important research value. Traditional line drawing mainly relies on manual operation. Due to the particularity of murals, painters can only rely on local lighting to draw; with the promotion of multimedia technology, they instead take pictures or record cave murals, and then compare them to digital murals for line drawing. The process of relying on manual line drawing is very complicated and time-consuming. Compared with the spread of threats faced by murals, the purpose of protecting murals is very small. Using computer and image processing technology, the line drawing generation process of murals , will not cause damage to cultural relics, but also has the characteristics of high efficiency and reusability that traditional methods do not have. Therefore, it is necessary to carry out research on line drawing generation technology for digital murals.

The key technology of line drawing generation involves edge extraction. Typical edge extraction operators include Canny[1] and so on. However, when these operators are directly applied to the extraction of line drawing, they will respond to both sides of the stroke, so that the same stroke corresponds to two edges. The global edge probability detection operator (Glabalized Probability of a Boundary, gPb) [2] has a high accuracy rate, and only generates one edge for some thinner strokes, but its width is single pixel and the continuity is poor .If skeleton extraction is not performed, although the result of gPb operator can reflect the width of strokes, it cannot extract relatively dense strokes. The flow-based difference of Gaussian (FDoG) algorithm [3] uses a flow-based anisotropic filter to replace the traditional edge detection operator, and uses a nonlinear kernel to filter the vector field to construct an edge tangent flow, so that the salient edge retains its original With direction, the direction of the weak edge is consistent with the direction of the prominent edge in its neighborhood, which is an effective line representation method. However, under the influence of natural or human factors, mural images generally face defects, cracks, oxidation and other phenomena, so there are a lot of noise in the image, or many unnecessary details, so smoothing or denoising technology must be considered for preprocessing operations on mural images.

references:

[1] Zhuang Jun, Li Bicheng, Chen Gang. An effective method for binarizing text and images. Microcomputer Information, 2005(8): 56-58.

[2] Chen Dan, Zhang Feng, He Guiming. An Improved Text Image Binarization Algorithm. Computer Engineering, 2003(13): 85-86.

[3] J.Bernsen.Dynamic Thresholding of Gray level.Internal Conference on Pattern Recognition,1986:1251-1255.

Contents of the invention

The invention proposes a method for generating a line drawing for a mural image, which can automatically convert a color mural image into a corresponding line drawing, and has the characteristics of strong anti-interference ability and wide application range. The technical solution is as follows:

A method for generating line drawings for mural images, comprising the following steps:

(1) Use a global variational smoothing filter to perform smoothing preprocessing on the input original image to obtain a smoothed image I;

(2) Convert the smooth image I to a grayscale image I ₀ , and its red, green, and blue three-channel images are represented by I _R , I _G and I _B respectively;

(3) Carry out high-frequency filtering and upgrading to I ₀ based on Gaussian blur, and the processing result is represented by I ₂ ;

(4) Construct the edge tangential flow field, and then use the improved FDoG filter method to extract the enhanced line drawing, the output result is represented by H', and the following process is used for contour extraction:

Step 1: Construct the edge tangent flow field from the smooth image I, t(x) represents the local edge direction, and has the maximum contrast in its vertical direction, that is, the gradient direction;

Step 2: On the basis of constructing the edge tangential flow field, improve the FDoG method, use the lines extracted by high-frequency lifting to strengthen the characteristics of the mural, and replace the original input grayscale image I ₀ with it and the high-frequency filter lifting result The result I _m obtained after multiplying I ₂ defines a new FDoG filter H(x);

(5) Use the median filtering technique to filter the image H'.

The beneficial effects of the invention are as follows: 1) it has strong anti-noise interference ability; 2) the generated line drawing can retain the main structure and artistic style of the mural.

Description of drawings

Figure 1 Block diagram of the proposed method

Figure 2 Example diagram of the processing process of the proposed method

(a) Input image (b) Smoothing result (c) Grayscale image

(d) Grayscale incremental image (e) High frequency lifting result (f) Line drawing extraction and denoising result

Figure 3 Part of the experimental results (a) Three input mural images (b) Three mural line drawing generation results

Detailed ways

The mural image-oriented line drawing extraction and generation method proposed by the present invention includes steps such as smoothing, high-frequency filtering, edge tangent flow construction, contour extraction, and median filtering. First, global variational smoothing is performed on the original mural image to remove unnecessary details, and then grayscale processing is performed, and then high-frequency filtering based on Gaussian blur is performed to simplify the distribution of background pixel grayscale values, and then the edge is constructed Tangent flow (ETF), and use the improved FDoG to extract the outline to get the abstract line drawing. Finally, the median filtering method is used to process the result.

It mainly includes: smoothing preprocessing, grayscale, high-frequency filter enhancement, flow Gaussian difference contour extraction and median filter and other steps. Figure 1 presents the block diagram of the proposed method.

1. Smooth preprocessing

Due to natural or man-made factors causing mural damage, oxidation and other problems, we must first smooth the mural image to eliminate noise interference as much as possible. The proposed method uses a global variational smoothing filter. The specific process is as follows:

Algorithm 1: Global Variational Smoothing

Step 1: input signal g, output signal f, and the objective function is defined as formula (1):

λ is a weight that controls the proportion between the two. In fact, it is a smoothing parameter. When its value is larger, it becomes smoother. The number of non-zero gradients in the image has a monotonically increasing relationship with λ.

Step 2: Calculate the L0 norm c(f) of the image gradient, that is, the constraints are as in formula (2).

c(f)＝#{p||f _p -f _p-1 |≠0} (2)

Where p and p-1 are adjacent elements in the image, |f _p -f _p-1 | is the image gradient, #{} represents the count, so this expression represents the non-zero number of output or the gradient is not zero The number is equal to k. The image processing results are represented by I.

2. Grayscale

Convert the smoothed image obtained in the previous step to a grayscale image. The smooth image is represented by I, and the red, green and blue three-channel images are represented by I _R , I _G and I _B respectively. Use the formula (3) to get the grayscale image, and express it with I ₀ , that is:

3. High-frequency filter enhancement

Generally, the stroke pixels and background pixels in the mural image have a large difference in gray value in the local range, but the gray value range overlaps in the global range. In order to make the stroke easier to be extracted, while maintaining the distance between the stroke and the background Contrast, this paper performs high-frequency filtering on I ₀ based on Gaussian blur, and the processing result is represented by I ₂ . The specific process is as follows:

Algorithm 2: High frequency boost

Step 1: Perform Gaussian blurring on the grayscale image I ₀ , that is, Gaussian low-pass filtering. The definition of the filter is shown in formula (4), and the processing result is represented by I ₁ .

Among them, σ is the standard deviation of the Gaussian distribution, which can represent the degree of expansion of the distribution curve.

Step 2: Use formula (5) to calculate the increment of the grayscale image, and the result is represented by ΔI ₀ , that is,

Step 3: Do a high frequency boost. The result is represented by I ₂ , that is,

I ₂ (x,y)=min(255,I ₀ (x,y)+ΔI ₀ (x,y)) (6)

4. Contour extraction

The edge tangential flow field is constructed, and then the improved FDoG filter method is used to extract line drawings with enhanced effects. The output result is represented by H'. The present invention uses the following process for contour extraction:

Algorithm 3: Flow Gaussian Difference Filtering

Step 1: Use the classic ETF (Edge Tangent Flow) method to construct the edge tangent flow t(x) from the smooth image I. t(x) represents the local edge direction, which means that there is maximum contrast in its vertical direction, that is, the gradient direction.

Step 2: On the basis of constructing the edge tangential flow field, improve the classic FDoG (Flow-Based Difference Of Gaussian) method, use the lines extracted by high-frequency enhancement to strengthen the characteristics of murals, and convert the original input grayscale image I ₀ is replaced by the result I _m obtained after multiplying it with the high-frequency boost result I ₂ . Use H(x) to define a new FDoG filter, which is defined as follows:

In the formula, G _σ is a Gaussian function with unilateral quantity and variance σ ² , see formula (4). σ _m determines the length of the flow core, assigning a fixed value to σ _m determines the size of the gradient direction S. F(s) is defined as follows:

I _m (x, y) = I ₀ (x, y)·I _m (x, y)/255(9)

l _s (t) indicates the point at t on the straight line l _s , t is the arc length parameter, and its value is between [-T, T]. I _m (l _s (t)) represents the value of the input image _Im at l _s (t). ρ controls the detected noise level and generally varies in the [0.97,1] region.

Step 3: Use the threshold to extract the black and white line image, and the result is represented by H'(x).

Among them, tanh(H(x)) is the hyperbolic tangent function, and τ is the threshold in the [0,1] range, which controls the sensitivity of edge detection.

In the above algorithm, ρ=0.998, τ=0.5, σ _s =1.6σ _c , and the value of σ _c can be adjusted by setting by the user.

5. Median filtering

Considering that the results generated by the above method still contain some noise, the image H' is filtered by median filtering technology to reduce the influence of noise. G represents the image processed by bilateral filtering.

Visual C++ 2010 under Windows 7 system is used as the experimental simulation platform. The scanned document images collected by ourselves are selected as the test set, with a total of 120 images. The horizontal/vertical resolution is 300dpi, and the number of pixels is 800×680. The method proposed by the invention is used to process the test image, and a good processing effect is obtained. The average processing speed is 10s, and the processing speed can meet the requirements.

Figure 2 shows an example of the processing process of the proposed method. Figure 3 shows more processing results, where (a) is the input mural image, and (b) is the processing result of the method proposed in the present invention.

The present invention can be summarized as following method step;

Step 1: Using Algorithm 1, combined with formula (1) and formula (2), the input color mural image is smoothed and preprocessed, and the processing result is represented by I.

Step 2: Use formula (3) to perform grayscale processing on I, and the processing result is represented by I ₀ .

Step 3: Use Algorithm 2, combined with formula (4), formula (5) and formula (6), to carry out high-frequency filtering and upgrading on I ₀ , and the processing result is represented by I ₂ .

Step 4: Use Algorithm 3 and combine Equation (7) to Equation (10) to construct edge tangent flow, perform flow-based Gaussian difference line extraction on I _m , and denote the processing result by H'.

Step 5: Carry out median filtering on H' to remove noise, and the processing result is represented by G.

Claims (2)

1. a kind of line drawing drawing generating method for mural painting image, comprises the following steps：

(1) artwork of input is smoothly pre-processed using global variation smoothing filter.Obtain smoothed image I；

(2) smoothed image I is converted into gray level image I₀, its red, green, blue triple channel image uses I respectively_R、I_GAnd I_BRepresent；

(3) based on Gaussian Blur to I₀Carry out High frequency filter lifting, handling result I₂Represent；

(4) the tangential flow field in edge is constructed, recycles improved FDoG filtered methods to extract the lines with enhancing effect Figure, output result are represented with H', and contours extract is carried out using following process：

1st step：Edge tangent flow field is built from smoothed image I, t (x) represents local edge direction, has most in its vertical direction Big contrast, i.e. gradient direction；

2nd step：On the basis of the tangential flow field in edge is constructed, FDoG methods are improved, the lines extracted using high boost are strong The characteristics of changing mural painting, the gray level image I of original input₀It is substituted for it and High frequency filter lifting result I₂The result obtained after multiplication I_m, define new FDoG wave filter H (x)；

3rd step：The lines image of black and white is extracted using threshold method, as a result with H'(x) represent；

(5) processing is filtered to image H ' using median filtering technology.

2. according to the method described in claim 1, it is characterized in that, the formula of the lines image of black and white is extracted using threshold method For：

Wherein, τ is the threshold value of [0,1] scope, controls the sensitivity of edge detection.