A multi-layer deviation correction method without positioning points

Abstract

The present invention discloses a multi-layer non-fixed point correction algorithm, which uses different devices to obtain different fluorescence or bright field images of cells, grayscales the bright field layer, and then uses dynamic high contrast retention and then performs adaptive binarization to obtain a bright field external edge enhancement map. After grayscale, the fluorescence layer is also adaptively binarized to obtain an external edge enhancement map of the fluorescence layer. If there is a bright field image in the image to be corrected, the inversion operation is performed and step three is performed again to obtain an internal edge enhancement map. Then, the complete edge enhancement map of the processed image with the highest resolution is used as the standard matching layer, and a sliding window A is defined inside it to be fixed, and a sliding window B is defined to slide within the complete edge enhancement map of other layers; the sliding window A image and all sliding window B images are taken to perform a variance operation, and the minimum value is found after traversing and recording the results. The minimum value is converted to the offset between sliding window B and sliding window A, and finally the corresponding layer is corrected according to the offset.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and more specifically, relates to a multi-layer non-positioning point correction method. Background Art

[0002] With advances in science and technology, fluorescence imaging is gaining increasing attention for the rapid detection of various cancer cells, heavy metals, and trace organisms. This technology typically involves transfecting cells with targeted dyes, sensitizing them to specific fluorescent signals, and then imaging them, enhancing the appearance of previously unseen cells or regions. Fluorescence detection has become a promising method due to its high sensitivity, selectivity, in-situ real-time detection, and affordability.

[0003] Layer deskew is a common operation in image processing, particularly in practical applications. For example, it plays a crucial role in image correction in GIS systems and image stitching in Photoshop. These operations typically involve comparing and matching the feature points of each layer before performing deskew.

[0004] Due to the characteristics of fluorescence imaging, such as slight probe vibration, cell image size distortion caused by different wavelengths, and image offset caused by inconsistent light source center positions, scaling correction and image de-skew must be performed before image processing to achieve uniform image expression. Furthermore, cell staining images suffer from sparse cell edge feature points and similar individual morphologies, making it impossible to use conventional methods such as SUFT feature matching for layer-by-layer de-skew. Therefore, we need to propose a multi-layer, non-locating point de-skew method. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a multi-layer non-positioning point correction method.

[0006] To achieve the above object, the present invention provides the following technical solution: a multi-layer non-positioning point correction method, the method comprising the following steps:

[0007] Step 1: Obtain microscopic images under different fluorescence bands and multi-layer images of the bright field.

[0008] Step 2: Use the Resize function to unify the size of image samples using cubic interpolation.

[0009] Step 3: After grayscale conversion of the brightfield layer, dynamic high contrast retention is used, followed by adaptive binarization to obtain a brightfield outer edge enhancement map. After grayscale conversion of the fluorescence layer, adaptive binarization is also performed to obtain a fluorescence outer edge enhancement map.

[0010] Step 4: Perform the inverse operation on all layers and then perform step 3 again to obtain the internal edge enhancement map.

[0011] Step 5: perform an addition operation on the results of steps 3 and 4 to obtain a complete edge enhancement image;

[0012] Step 6: Use the complete edge-enhanced image of the highest resolution image obtained in step 1 as the standard matching layer, define a fixed sliding window A inside it, and define a sliding window B that slides within the complete edge-enhanced image of the other layer;

[0013] Step 7: Take the sliding window A image and all sliding window B images and perform variance calculation, and record the results.

[0014] Step 8: After traversing the recorded results, find the minimum value and convert the minimum value to the offset between sliding window B and sliding window A.

[0015] Step 9. Correct the corresponding layer according to the offset obtained in step 8.

[0016] The technical effects and advantages of the present invention are as follows: The present invention provides a multi-layer, non-positioning point deflection correction method. Compared with existing technologies, the present invention normalizes cell images obtained under different fluorescent probes in different layers, then automatically sets a sliding window based on the cell-to-image ratio, compares the contents of different layers, and deflects all layers after determining the offset. This results in more uniform image positions, facilitating subsequent cell analysis and identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a logic block diagram of the present invention.

[0018] Figure 2 Schematic diagram of the application scenario and results of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0020] The present invention provides Figure 1 A multi-layer non-positioning point correction method includes the following steps:

[0021] Step 1: Obtain microscopic images under different fluorescence bands and multi-layer images of the bright field.

[0022] Step 2: Use the Resize function to unify the size of image samples using cubic interpolation.

[0023] Step 3: After grayscale conversion of the brightfield layer, dynamic high contrast retention is used, followed by adaptive binarization to obtain a brightfield outer edge enhancement map. After grayscale conversion of the fluorescence layer, adaptive binarization is also performed to obtain a fluorescence outer edge enhancement map.

[0024] Step 4: Perform the inverse operation on all layers and then perform step 3 again to obtain the internal edge enhancement map.

[0025] Step 5: perform an addition operation on the results of steps 3 and 4 to obtain a complete edge enhancement image;

[0026] Step 6: Use the complete edge-enhanced image of the highest resolution image obtained in step 1 as the standard matching layer, define a fixed sliding window A inside it, and define a sliding window B that slides within the complete edge-enhanced image of the other layer;

[0027] Step 7: Take the sliding window A image and all sliding window B images and perform variance calculation, and record the results.

[0028] Step 8: After traversing the recorded results, find the minimum value and convert the minimum value to the offset between sliding window B and sliding window A.

[0029] Step 9. Correct the corresponding layer according to the offset obtained in step 8.

[0030] In the step 1, the multi-layer images are any M fluorescence images and N bright field images. If (M+N)>1 is satisfied, the subsequent steps can be executed.

[0031] The Resize function in step 2 is a function in the Emgucv library. The Resize function is used to resize the source image so that the size of the same batch of cell images at different resolutions is unified. The Resize function can be used to adjust the size of three-channel images (such as RGB images) and single-channel images;

[0032] The cubic interpolation in step 2 can be set to one of adjacent point difference, bilinear difference, bicubic difference, inner difference and triangular difference, preferably bicubic difference, among which bicubic interpolation has a better effect and is widely used in many high-quality image transformations;

[0033] In the grayscale conversion in step 3, the bright field image can be directly obtained by the Gray function of Emgucv. If the fluorescence image is red, green, or blue, it can be directly obtained by channel extraction. Other fluorescent colors can also be obtained by the Gray function of Emgucv.

[0034] The grayscale conversion in step 3 may not be performed. This process is an image dimensionality reduction operation, which can reduce the system overhead of the algorithm.

[0035] The dynamic high contrast retention in step 3 is obtained by subtracting the Gaussian blur image from the input image, wherein the Gaussian kernel for generating the Gaussian blur image is the maximum cell diameter in the image. With image width W or image height H passed Obtain, if G<3, then force G=3, if G>101, then force G=101.

[0036] The inversion operation in step 4 is derived from the Not function in the Emgucv library. This operation addresses the problem of cells having distinct outer or inner edges, depending on cell type, live or dead state, and other factors. Preserving the high contrast between the original image and the inverted image allows for adaptation to any situation, enhancing the algorithm's versatility.

[0037] The addition operation performed in step 5 is obtained by executing the following process: the external edge enhancement image and the internal edge enhancement image are converted from Image format to Mat format, and then executed CvInvoke.add(mat,mat,dst) to obtain the result, where dst is the Mat operation result, which is converted from Mat format to Image format. This step is intended to reduce system overhead during operation execution and improve algorithm efficiency.

[0038] The sliding window size in step 6 is equal to the Gaussian kernel size in step 3. In addition, the default sliding window initial position is (x,y), where x=( ), y=( ).

[0039] The sliding distance of the sliding window in step 6 is G by default, and the step size is 1 pixel by default. The sliding area is to traverse all pixels in the sliding window A.

[0040] The variance calculation in step 7 is calculated by the following formula: , where W and H are the width and height of the sliding window, is the pixel coordinate, is the mean, and all calculated variance values ​​are stored in a one-dimensional array of Double type.

[0041] The step eight traversal record result is obtained by the following process: reset the result to a one-dimensional array, loop to find the minimum value sequence number I, calculate the x-direction offset x= ), y-direction offset y= ).

[0042] The step nine is completed by the WarpAffine function of Emgucv according to the offset correction of the corresponding layer, where the conversion matrix is (f indicates that the computer's operating precision is float floating point type).

[0043] In summary, the present invention first uses different devices to acquire different fluorescence or brightfield images of cells. Then, using the Resize function and cubic interpolation, the multi-layer image samples are scaled down. The brightfield layer is then grayscaled, dynamically retaining high contrast, and then adaptively binarized to obtain a brightfield external edge enhancement map. The fluorescence layer is grayscaled and similarly adaptively binarized to obtain a fluorescence external edge enhancement map. If a brightfield image exists in the image to be de-skewed, a negation operation is performed and step three is repeated to obtain an internal edge enhancement map. The complete edge enhancement map of the highest-resolution processed image is then used as the standard matching layer. A fixed sliding window A is defined within it, while a sliding window B is defined within the complete edge enhancement maps of the other layers. A variance calculation is performed on the image of sliding window A and all images of sliding window B. The results are recorded and the minimum value is found. This minimum value is converted to the offset between sliding window B and sliding window A. Finally, the corresponding layer is de-skewed according to the offset.

[0044] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A multi-layer non-positioning point correction method, characterized by: The steps include: Step 1: Acquire microscopic images at different fluorescence wavelengths and multi-layer bright field images; Step 2: Use the Resize function to unify the size of image samples using cubic interpolation. Step 3: After graying the bright field layer, adaptive binarization is performed using dynamic high contrast retention and then obtained as a bright field outer edge enhancement map; after graying the fluorescent layer, adaptive binarization is also performed to obtain an outer edge enhancement map of the fluorescent layer; Step 4: Perform the inversion operation on all layers and then repeat step 3 to obtain the internal edge enhancement map; Step 5: perform an addition operation on the results of steps 3 and 4 to obtain a complete edge enhancement image; Step 6: Use the complete edge-enhanced image of the highest resolution image obtained in step 1 as the standard matching layer, define a fixed sliding window A inside it, and define a sliding window B that slides within the complete edge-enhanced image of the other layer; Step 7: Take the sliding window A image and all sliding window B images and perform variance calculation, and record the results; Step 8: After traversing the recorded results, find the minimum value and convert the offset between sliding window B and sliding window A corresponding to the minimum value; Step 9. Correct the corresponding layer according to the offset obtained in step 8.

2. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The Resize function in step 2 is a function in the Emgucv library. The Resize function is used to resize the source image so that it accurately matches the target image. The Resize function can be used to adjust the size of a three-channel image and a single-channel image.

3. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The cubic interpolation in step 2 may be set to one of adjacent point difference, bilinear difference, bicubic difference, inner difference and trigonometric difference.

4. The multi-layer non-positioning point correction method according to claim 1, characterized in that: In the grayscale conversion in step 3, the bright field image can be directly obtained by the Gray function of Emgucv. If the fluorescence image is red, green, or blue, it can be directly obtained by channel extraction. Other fluorescent colors can also be obtained by the Gray function of Emgucv.

5. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The dynamic high contrast retention in step 3 is obtained by subtracting the Gaussian blur image from the input image, wherein the Gaussian kernel for generating the Gaussian blur image is the maximum cell diameter in the image. With the image width W through get.

6. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The negation operation in step 4 is obtained from the Not function in the Emgucv library.

7. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The addition operation performed in step 5 is obtained by executing the following process: the external edge enhancement map and the internal edge enhancement map are converted from Image format to Mat format, and CvInvoke.add (mat, mat, dst) is executed to obtain them, where dst is the Mat operation result, which is obtained after converting the Mat format to Image format.

8. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The sliding window size in step six is ​​equal to the Gaussian kernel size in step three.

9. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The starting position of the sliding window B in step 6 during the inner sliding of other layers is obtained based on prior knowledge. The default starting point is the value of the Location of the sliding window A, and the default distance is to traverse all pixels of the sliding window A.

10. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The variance calculation in step 7 is calculated by the following formula: , where W and H are the width and height of the sliding window, is the pixel coordinate, is the mean.

11. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The result of the traversal record in step eight is obtained by the following process: Convert the image to Mat format, reset the result to a one-dimensional array, and loop to find the minimum value.

12. The multi-layer non-positioning point correction method according to claim 1, characterized in that: The step nine is completed by the WarpAffine function of Emgucv according to the offset correction of the corresponding layer, where the conversion matrix is , f means the computer's operating precision is float floating point type.

Images