CN1916750A - Digital Imaging Autofocus Method Based on Contourlet Transform - Google Patents

Abstract

The invention discloses a digital imaging automatic focusing method based on Contourlet transformation, which utilizes the multi-resolution and multi-direction characteristics of Contourlet transformation, selects important coefficients of Contourlet domain which can represent the outline sharpness and the detail richness of an image to calculate a focusing evaluation function so as to determine the accurate focusing position of a digital imaging system, and can better utilize the direction information of the image by utilizing the dominant direction of a texture edge to determine a dominant direction sub-band of the Contourlet domain for the image containing a large number of textures with consistent trend, thereby improving the focusing sensitivity and reducing the calculation complexity.

Description

Digital Imaging Autofocus Method Based on Contourlet Transform

technical field

The invention relates to a digital imaging automatic focusing method, in particular to a digital imaging automatic focusing method based on Contourlet transformation.

Background technique

In an imaging system, the lens has an optimal image plane position for imaging an object. Deviation from this position will result in blurred images and reduced imaging quality; therefore, it is very important for an imaging system to be able to focus accurately. The imaging system based on the digital image adopts the automatic focus method, and its key lies in the focus evaluation function. The ideal focus evaluation function curve shows a parabolic shape, its peak value corresponds to the best imaging position, and the function value decreases when leaving the best point. Therefore, the process of autofocus is essentially a process of finding the maximum value of the focus evaluation function.

Usually, most of the energy of an image is concentrated in the low-frequency and middle-frequency bands of the frequency domain, but the sharpness of the image outline and the richness of details depend on the high-frequency components of the image. When the image is clear and rich in details, it shows that the eigenvalues (such as grayscale, color, etc.) of adjacent pixels change greatly in the spatial domain, and it shows that there are many high-frequency components in the frequency spectrum in the frequency domain. Commonly used focus evaluation functions are divided into two types: spatial domain and frequency domain. Several commonly used spatial focus evaluation functions include laplacian operator, sobel operator, prewitt operator, and energy variance operator. The amount of calculation required for the focus evaluation method based on space is relatively small, but its disadvantage is that it is greatly affected by noise, that is, the noise immunity is poor. The frequency-domain focusing evaluation method utilizes the overall characteristics of the image, and its noise resistance is relatively good, but this method usually needs to perform Fourier transform or other transformations on the image first, and then evaluate the sharpness of the image based on the transform coefficients.

The multi-directional and multi-scale characteristics of the Contourlet transform enable it to capture the texture, details and edge information in natural images well, and the process of auto-focusing in digital images is to evaluate the clarity of information such as texture details and find the extreme value. Therefore, the Contourlet transform is a suitable tool for autofocus evaluation. The digital Contourlet transform is divided into two steps. First, the Laplacian Pyramid (LP) is used to perform multi-resolution decomposition of the image to capture singular points, and then the two-dimensional directional filter bank (Directional FilterBank, DFB) gathers the singular points with similar positions into contour segments according to their different orientation characteristics. The Contourlet coefficients of each level have multiple direction subbands, and these subbands are edge images containing high frequency components. After the first-level Contourlet decomposition of the image, the corresponding edge feature image can be obtained. Figure 1(b) is the Contourlet transform coefficient map after the 2-level Contourlet transform in Figure 1(a). But for the imaging system, this is only a discrete edge feature image sequence. To achieve accurate focusing, an objective function must be established to give statistical properties to the discrete image to judge the best focusing position. The process of focusing is a process of image energy change, and generally the image with precise focus has the largest energy.

Contents of the invention

The technical problem to be solved by the present invention is to provide an accurate and effective digital imaging autofocus method based on Contourlet transformation.

The technical solution adopted by the present invention to solve the above-mentioned technical problems is: a digital imaging autofocus method based on Contourlet transformation, which includes the following steps: ① For the digital microscopic imaging system, continuously adjust the focus to obtain images with different resolutions The input signal of ; ② Perform multi-level Contourlet transformation on the two-dimensional input signal f of each image to obtain the set of Contourlet transformation coefficients {C _{j, k} }, where C _{j, k} represents the jth layer of the Contourelet domain (j=1,. .., J, J is the highest resolution layer) coefficient matrix of the kth sub-band; ③Define the focus evaluation function: $f={\mathrm{\Σ}}_j{\mathrm{\Σ}}_k{\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2,$ Among them, ‖E _j,k ^significant ‖ ² represents the energy sum of important coefficients in the kth subband of the jth layer of the Contourelet domain, namely ${\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(C_{j,k,i}^{\mathrm{significant}}\right)}^2,$ C _{j, k, i} ^significant is the important coefficient in the kth subband of the jth layer of the Contourelet domain, namely $C_{j,k,i}^{\mathrm{significant}}\∈C_{j,k},$ N is the number of important coefficients selected in the sub-band, and the value of the focus evaluation function F is calculated for the input signal of each image; ④ When the change trend of the focus evaluation function F value changes after continuous focusing, the focal length is called back until The image input signal corresponding to the maximum value of the focus evaluation function F appears, and then the focusing process ends; ⑤ Take the image corresponding to the maximum value of the focus evaluation function F of the input signal to obtain the image with the best definition and correct focus.

In order to reduce the computational complexity and achieve the purpose of fast focusing, the ‖E _j,k ^{sgnificant ‖2} in the above-mentioned focusing evaluation function F can also be defined as ${\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|C_{j,k,i}^{\mathrm{significant}}\right|.$

The two-dimensional input signal f in the above method may be the whole area or a partial area or a combination of local areas or a down-sampled signal of the digital microscopic imaging image.

In the selection of the direction sub-bands involved in the calculation of the focus evaluation function, several important directions of the image to be focused can be determined according to the energy of each direction sub-band, so that the sub-bands containing important direction information are dominant in the focus evaluation function Components, the direction sub-bands containing less direction information are given less shares, and the different importance of different direction sub-bands is achieved by weighting, that is, the focus evaluation function can also be defined as:

$f={\mathrm{\Σ}}_j\underset{k=1}{\overset{K_j}{\mathrm{\Σ}}}{w}_{j,k}\×{\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2,$ Among them, K _j is the number of sub-bands in the jth layer, and w _{j, k} are the weights.

In the selection of the direction sub-bands involved in the calculation of the focus evaluation function, the dominant direction of the image to be focused can also be determined according to the energy of each direction sub-band, and only the important coefficients of several sub-bands determined by the dominant direction are used for focusing. Computation of the evaluation function.

Compared with the existing classical focus evaluation method, the digital imaging autofocus method based on Contourlet transform of the present invention makes good use of the flexibility of Contourlet transform to capture image direction information and possible direction numbers, and can effectively extract image The high-frequency information of the image can be used to determine the image with the largest energy in the image sequence, so as to determine the precise focus position of the digital imaging system.

Description of drawings

Figure 1(a) is the original image of peppers;

Figure 1(b) is a 2-level Contourlet transform domain coefficient map of peppers;

Figure 2(a) is a microscope picture of a section of ciliated epithelial cells with blurred focus;

Fig. 2 (b) is a microscope image of a clearly focused ciliated epithelial cell section;

Fig. 3 is the numbering figure of each direction sub-band of the resolution highest level of the 2-level Contourlet transform;

Figure 4 is a comparison of the autofocus methods based on Contourlet transform with different subband coefficients;

Figure 5 is a comparison between the auto-focus method based on Contourlet transform and the classic focus method.

Detailed ways

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

A digital imaging autofocus method based on Contourlet transform, which includes the following steps: ①For a digital microscopic imaging system, continuously adjust the focus to obtain input signals of images with different resolutions; ②Two-dimensional input for each image The signal f is subjected to multi-level Contourlet transformation to obtain a set of Contourlet transformation coefficients {C _{j, k} }, where C _{j, k} represents the jth layer of the Contourelet domain (j=1,..., J, J is the layer with the highest resolution) The coefficient matrix of the k-th subband; ③Contourlet transform is sparse, that is, only a small number of transform domain coefficients can be used for inverse transformation to effectively approximate the original image. As far as the high-frequency coefficients of Contourlet transform are concerned, most of their amplitudes are gathered around zero, and only a small number of coefficients have large amplitudes. These coefficients with large amplitudes are called "significant coefficients" (significant) , which describes the detailed texture information of the image. Therefore, a certain number of important coefficients can be selected for focus evaluation in different direction subbands at different resolutions, that is, the focus evaluation function is defined: $f={\mathrm{\Σ}}_j{\mathrm{\Σ}}_k{\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2,$ Among them, ‖E _j,k ^significant ‖ ² represents the energy sum of important coefficients in the kth subband of the jth layer of the Contourelet domain, namely ${\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\left(C_{j,k,i}^{\mathrm{significant}}\right)}^2,$ C _{j, k, i} ^significant is the important coefficient in the kth subband of the jth layer of the Contourelet domain, namely $C_{j,k,i}^{\mathrm{significant}}\∈C_{j,k},$ N is the number of important coefficients selected in the sub-band, and the value of the focus evaluation function F is calculated for the input signal of each image; ④ when the focus is continuously adjusted, the change trend of the value of the focus evaluation function F (from increasing to (decrease or change from decrease to increase), call back the focal length until the image input signal corresponding to the maximum value of the focus evaluation function F appears, and then end the focusing process; ⑤ take the image corresponding to the maximum value of the focus evaluation function F of the input signal That is, the sharpest image with the correct focus is obtained.

In order to reduce the computational complexity and achieve the purpose of fast focusing, the ‖E _j,k ^{significant ‖2} in the above-mentioned focus evaluation function F can also be defined as ${\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}\left|C_{j,k,i}^{\mathrm{significant}}\right|.$

The two-dimensional input signal f in the above method may be the whole area or a partial area or a combination of local areas or a down-sampled signal of the digital microscopic imaging image.

Compared with frequency domain evaluation methods such as wavelet transform and Fourier transform, the biggest advantage of Contourlet transform lies in the flexibility of capturing the direction information of the image and the number of possible directions. The number of direction subbands of wavelet transform is constant, and there are only three directions: horizontal, vertical and diagonal. In the Contourlet transform domain, the number of direction sub-bands is more, and the capture of direction information is more flexible. In order to make better use of the multi-directional characteristics of the Contourlet transform, several important directions of the image to be focused can be determined according to the energy size analysis of the sub-bands in each direction, so that the sub-bands containing important direction information are dominant in the focus evaluation function Components, the direction sub-bands containing less direction information are given less shares, and the different importance of different direction sub-bands can be realized by weighting. In addition, due to the sparsity of the Contourlet transform, the measurement of the energy of the above-mentioned direction subbands can be realized by only selecting some "important" coefficients. That is, the focus evaluation function can also be defined as: $f={\mathrm{\Σ}}_j\underset{k=1}{\overset{K_j}{\mathrm{\Σ}}}{w}_{j,k}\×{\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2,$ Among them, K _j is the number of sub-bands in the jth layer, and w _{j, k} are the weights.

In addition, in the selection of the direction sub-bands involved in the calculation of the focus evaluation function, the dominant direction of the image to be focused can also be determined according to the energy size analysis of each direction sub-band, and only the important coefficients of several sub-bands determined by the dominant direction are used to determine the Calculate the focus evaluation function.

By moving the focusing lens equidistantly and taking one image for each step, in this embodiment, 18 microscopic images of "ciliated epithelial cells" sections were collected, and the size of the images was 512×512. The image quality has gone through a process from blurry to clear and then to blurry, and the focus evaluation function used should show a changing law from small to large and then reduced. Figure 2(a) and Figure 2(b) show two images with different degrees of focus accuracy, that is, the microscopic observation results of the blurred and clear slices of "ciliated epithelial cells". The texture edge information contained in this series of images is mainly divided into two categories: one is a large number of large or small circular areas and dots. Since the direction information of the circle is evenly distributed in each direction, there is no specific dominant direction; the second is the stripe-like veins, the direction of these veins is roughly the same, close to the 45.0° direction, which is expressed in the Contourlet transform domain, which is The energy is relatively concentrated on the 5th direction subband corresponding to the highest resolution layer of the 2-level Contourlet transform as shown in Fig. 3. Figure 4 shows five focusing evaluation function curves based on Contourlet transform for the image sequence shown in Figure 2. For the convenience of comparison, normalization processing is carried out. These five methods are:

M1: the energy sum of all direction subbands on the highest layer of Contourlet domain resolution;

M2: That is, the energy sum of the important coefficients of all direction subbands on the highest layer of the Contourlet domain resolution, where the 1% coefficient with the largest amplitude on each direction subband is taken;

M3: That is, the focus evaluation function is obtained by weighting the sub-bands in different directions on the highest layer of the Contourlet domain resolution according to their importance. For this embodiment, the following focus evaluation function is used: $f=\underset{k=1}{\overset{8}{\mathrm{\Σ}}}{w}_k\×{\left|\right|{\mathrm{E.}}_{J,k}\left|\right|}^2,$ Among them, k is the serial number of the direction sub-band, w _k is the weight value, and the No. 5 sub-band corresponding to the dominant direction has a larger weight;

M4: Only the energy of the No. 5 sub-band corresponding to the dominant direction is used to evaluate the focusing situation, that is, the focusing evaluation function is: $f={\left|\right|{\mathrm{E.}}_{J,5}\left|\right|}^2;$

M5: Only select the important coefficients in the No. 5 sub-band corresponding to the dominant direction to calculate the energy sum, that is, the focus evaluation function is: ${f=\left|\right|{\mathrm{E.}}_{J,5}^{\mathrm{significant}}\left|\right|}^2,$ Here, the 1% coefficient with the largest amplitude in the No. 5 subband is selected.

Figure 4 shows that the focus evaluation function curves of these five evaluation methods are all parabolic single-peak curves, and the peak points of the functions all appear at the ninth image, which is the precise focus point of the series of microscopic images. The difference between the function curves lies in the sharpness of the curves. Among the five focus evaluation curves, the curve of M5 is the steepest, and the peak of the curve is relatively sharp, while the curves of M1 and M2 are relatively close, and the peaks are relatively flat. M3 and M4 The sharpness of is in the middle. The sharpness of the five curves is as follows: M1≈M2<M3<M4<M5. The normalized curves of the five methods show that the autofocus method based on Contourlet transform is effective and feasible, especially for images containing more texture edge information in specific directions, using the coefficient of the dominant direction subband can not only improve The sensitivity of autofocus, and because only some coefficients are used, the computational complexity is reduced.

Figure 5 shows the comparison between the autofocus method based on Contourlet transform and some classic focus evaluation methods. The focus evaluation algorithms used include: Laplacian operator, Sobel operator, Prewitt operator, energy variance operator (standard) and wavelet transform ( wavelet) focus evaluation algorithm. It can be seen from Figure 5 that the normalization curves are all in the shape of a single-peak parabola, but the peaks of the Sobel operator and the Prewitt operator appear in the 10th image, which is inconsistent with other methods, that is, they cannot focus accurately. The rest of the classic focusing methods can achieve the purpose of precise focusing, and the methods M1 and M2 based on Contourlet transform have the steepest curves.

In summary, due to the multi-resolution and multi-directional characteristics of the Contourlet transform, the focus evaluation function based on the Contourlet transform can effectively extract the high-frequency information of the image, judge the image with the largest energy in the image sequence, and determine the accuracy of the digital imaging system. focus position. For images containing a large number of textures with relatively consistent directions, the method of determining the dominant direction subband of the Contourlet domain by using the dominant direction of the texture edge can make better use of the direction information of the image, thereby improving the focus sensitivity and reducing the computational complexity.

The invention is not limited to the specific details and examples shown and described herein without departing from the spirit and scope of the general concept defined by the claims and equivalents.

Claims (5)

1, a kind of digital imaging autofocus method based on Contourlet transformation, it comprises the following steps: 1. for the digital microscope imaging system, continuous focusing obtains the input signal of the image that clarity is different; Multi-level Contourlet transform is performed on the dimensional input signal f to obtain a set of Contourlet transform coefficients {C _{j, k} }, where C _{j, k} represents the jth layer of the Contourelet domain (j=1,..., J, J is the highest resolution layer) the coefficient matrix of the kth subband; ③ define the focus evaluation function: $f={\mathrm{\&Sigma;}}_j{\mathrm{\&Sigma;}}_k{\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2,$ in, ${\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2$ Indicates the energy sum of the important coefficients in the kth subband of the jth layer of the Contourelet domain, namely ${\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\left(C_{j,k,i}^{\mathrm{significant}}\right)}^2,$ C _{j, k, i} ^significant is the important coefficient in the kth subband of the jth layer of the Contourelet domain, namely $C_{j,k,i}^{\mathrm{significant}}\&Element;C_{j,k},$ N is the number of important coefficients selected in the sub-band, and the value of the focus evaluation function F is calculated for the input signal of each image; ④ When the change trend of the focus evaluation function F value changes after continuous focusing, the focal length is called back until The image input signal corresponding to the maximum value of the focus evaluation function F appears, and then the focusing process ends; ⑤ Take the image corresponding to the maximum value of the focus evaluation function F of the input signal to obtain the image with the best definition and correct focus. 2, a kind of digital imaging autofocus method based on Contourlet transformation as described in claim 1, it is characterized in that in the focus evaluation function F ${\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}\left|C_{j,k,i}^{\mathrm{significant}}\right|.$ 3. A digital imaging autofocus method based on Contourlet transform as claimed in claim 1 or 2, characterized in that the two-dimensional input signal f is the entire area or partial area or a combination of local areas of the digital microscopic imaging image or the following sample signal. 4. A digital imaging auto-focusing method based on Contourlet transform as claimed in claim 1 or 2, characterized in that first, according to the energy size analysis of each direction sub-band, the important direction of the image to be focused is determined, so that important direction information is included The sub-bands of are the dominant components in the focus evaluation function, and the sub-bands containing less directional information account for a smaller share, and the importance of sub-bands in different directions is determined by weighting, that is, the focus evaluation function is defined as: $f={\mathrm{\&Sigma;}}_j\underset{k=1}{\overset{K_j}{\mathrm{\&Sigma;}}}{w}_{j,k}\&times;{\left|\right|{\mathrm{E.}}_{j,k}^{\mathrm{significant}}\left|\right|}^2,$ Among them, K _j is the number of sub-bands in the jth layer, and w _{j, k} are the weights. 5. A digital imaging auto-focusing method based on Contourlet transform as claimed in claim 1 or 2, characterized in that the dominant direction of the image to be focused is determined according to the energy size analysis of each direction sub-band, and only the dominant direction is used to determine The important coefficients of the subbands are used to calculate the focus evaluation function.

Images