TWI787134B - A gpu-accelerated data processing method for rapid noise-suppressed contrast enhancement - Google Patents

Abstract

The present disclosure relates to a data processing method, and more specifically, to a digital image processing method to enable a rapid noise-suppressed contrast enhancement in an optical linear or nonlinear microscopy imaging application. The disclosed method digitally mimics a hardware-based feedback-driven adaptive or controlled illumination technique by means of digitally resembling selective laser-on and laser-off states so as to selectively optimize the signal strength and hence the visibility of the weak-intensity morphologies while mostly preventing saturation of the brightest structures.

Description

A fast noise suppression contrast enhancement Data processing method and image acquisition and processing system including same

The disclosed document relates to a data processing method, in particular to a digital image processing method for rapid noise suppression and contrast enhancement.

Optical neuronal imaging helps researchers study various neurological diseases as well as brain function and dysfunction. Neuronal structures often exhibit dramatic changes in structural texture and signal intensity distribution. For example, when imaging neurons, cell bodies (i.e., soma) are often observed to be brighter than adjacent fibrous structures (i.e., axons and dendrites), which can be thinner than a micron. Indeed, imaging of neurons results in a sufficiently broad distribution of signal intensities. Even with highly advanced acquisition and display systems, due to their limited dynamic range, it is often difficult to digitize and visualize all neuronal details without sacrificing any information. Furthermore, various optical, electrical, and other environmental factors lead to noisy backgrounds that significantly contaminate weak-intensity signals emerging from hyperfine neuronal structures and ultimately deteriorate the signal-to-noise ratio of this modality (Signal- to-Noise Ratio, SNR) and contrast.

Conventional contrast enhancement of weak-intensity structures usually makes the brightest The structure saturates and can cause noise amplification problems, which further degrades SNR.

Adaptive/controlled illumination is a promising technique that allows instant local optimization of signal intensity by adjusting laser excitation power. However, Adaptive Lighting requires a dedicated hardware setup and may sense poor effective bandwidth due to slower electronic response, so irreversible loss of digital resolution (aliasing), especially for extended At high digital resolution in the field of view. Furthermore, adaptive lighting may still suffer from noise amplification while locally enhancing weak intensity structures.

US9639915B1 discloses an image processing method, including based on the different degrees of noise influence on the corresponding pixel, between pixels, other pixels, and pixels adjacent to the corresponding pixel, according to the linear noise model of the image for each image in the image Pixel configuration noise reduction filter. The method further includes performing denoising filtering on each pixel, applying a denoising filter to each pixel to obtain a denoised image.

US8417050B1 employs Robust Filtering at each of multiple scales or resolutions within a signal (such as a still image or video sequence). In certain embodiments, robust filtering includes or includes non-linear neighborhood operations at each scale to produce denoised, sharpened and contrast-enhanced signals as well as correction signals at each scale.

Although quite a few analog/digital signal processing methods have been proposed so far, existing hardware-based analog techniques require dedicated hardware configurations to operate, which increases cost and complexity. Due to electronic limitations, slower response speeds can result in poorer effective bandwidth, which in turn can lead to aliasing. Apart from In addition to hardware-based methods, existing software-based contrast enhancement algorithms often lead to unwanted amplification of noise, resulting in poor signal-to-noise ratios, and often end up saturating the brightest structures in the image.

It is therefore necessary to introduce a digital noise-compensated contrast enhancement technique that can be applied to both optical linear and nonlinear imaging modalities to help improve the visibility of weak signal structures while neither saturating the brightest structures nor Amplifies background noise.

The purpose of the present invention is to provide a digital method that does not require special hardware, which can mimic hardware-based adaptive/controlled lighting techniques, and selectively enhance the contrast of weak intensity structures in optical microscopic images, and at the same time, it is better for those with relatively low intensity. The structure of the bright intensity has little effect. This method employs efficient background noise suppression followed by local intensity enhancement.

In a first aspect, the present invention provides a data processing method for fast noise suppression contrast enhancement, comprising configuring a graphics processing unit or a central processing unit to perform the following steps: obtaining an input image, wherein the input image has a first width and a first height, and includes a plurality of pixels or data points having a specified bit depth that allows a maximum pixel value or a maximum pixel intensity raised from 2 to the power of bit depth minus 1 to get. performing a first binning or a first interpolation process on the input image, resizing the input image by a first downscaling factor to generate a first resized image, the first resized image having a first downscaling less than a first width factor, and a second height less than the first height by the first reduction factor. A first low-pass filtering process is performed on the first resized image to obtain a first blurred image. A second interpolation process is performed on the first blurred image to upscale it from a second width and a second height to obtain a second resized image having a first width and a first height. A segmentation process is performed to obtain a first segmented layer image by dividing a first specific number by the second resized image. Threshold processing is performed to truncate the first segmented layer image at a user-defined threshold to obtain a first zoomed-in layer image. A first zoom-in process is performed on the input image to multiply the input image by the first zoom-in layer image to generate a first zoom-in image. performing a second binning or a third interpolation process on the first enlarged image, resizing the first enlarged image by a second reduction factor to generate a third resized image having a width wider than the first a third width less by the second reduction factor, and a third height less than the first height by the second reduction factor. A second low-pass filtering process is performed on the third resized image to obtain a second blurred image. A fourth interpolation process is performed on the second blurred image to enlarge it from the third width and the third height to obtain a fourth resized image having the first width and the first height. A first subtraction process is performed on the fourth resized image and the first enlarged image to subtract the first enlarged image from the fourth resized image to generate a first subtracted image. A third low-pass filtering process is performed on the first subtraction image to obtain a third blurred image. A second subtraction process is performed on the first magnified layer image to subtract a second specific number higher than a user-defined threshold value from the first magnified layer image to obtain a second subtracted image or a second magnified layer image. A second zoom-in process is performed on the third blurred image to multiply the third blurred image by the second zoom-in layer image to generate a second zoom-in image. performing a third subtraction process on the input image and the second magnified image to subtract the second magnified image from the input image to obtain a third subtracted shadow image. A set of arithmetic operations are performed on the first magnification layer image to obtain a third magnification layer image. and performing a third enlargement process on the third subtracted image to multiply the third subtracted image by the third enlarged layer image to generate a third enlarged image or a noise-suppressed contrast-enhanced output image.

In a preferred embodiment, the first interpolation process, the second interpolation process, the third interpolation process and the fourth interpolation process are all bilinear.

In a preferred embodiment, the first downscaling factor in the first binning or first interpolation process is ten.

In a preferred embodiment, the first low-pass filtering process involves performing a Gaussian blur operation of the convolution using a first Gaussian kernel with a kernel size of 29x29.

In a preferred embodiment, an addition process is performed on the first blurred image to add a non-zero number to the first blurred image before performing the second interpolation process.

In a preferred embodiment, the first specified number in this segmentation process is 90% of the maximum pixel intensity.

In the preferred embodiment, the user-defined threshold in this threshold processing ranges from 3.0 to 8.0, allowing floating point numbers.

In a preferred embodiment, the second downscaling factor in the second binning or third interpolation process is three.

In a preferred embodiment, the second low pass filtering process involves performing a Gaussian blur operation of the convolution using a second Gaussian kernel with a kernel size of 29x29.

In a preferred embodiment, the third low-pass filtering process involves using The third kernel is a Gaussian kernel of size 7×7 to perform the convolutional Gaussian blur operation.

In a preferred embodiment, the second specific number in the second subtraction process is 1.25 times the user-defined threshold value.

In a preferred embodiment, the step of performing the set of arithmetic processing on the first magnified layer image to obtain the third magnified layer image includes: firstly, dividing the first magnified layer image by a division factor of 4 to obtain a first magnified layer image Two split layer images. Second, the second segmented layer image is raised to the power of 2 to obtain a modified segmented layer image. Thirdly, a value of 0.9 is added to the modified segmented layer image to obtain the third enlarged layer image.

ADD: Addition layer

AMP1: the first magnified image

AMP2: second magnified image

AMP3: third magnified image

BLR1: The first blurred image

BLR2: second blurred image

BLR3: third blurred image

DIV: first split layer image

INPUT: input image

LAY1: The first enlarged layer image

LAY2: The second enlarged layer image

LAY3: The third magnified layer image

OUTPUT: output image

RES1: First resized image

RES2: Second resized image

RES3: third resized image

RES4: fourth resized image

S01-S03: Processing

SUB1: first subtraction image

SUB2: Second subtraction image

SUB3: The third subtraction image

100: Image acquisition and processing system

110: computer

120: Optical linear or nonlinear microscopy systems

A1-A18: Steps

FIG. 1 is an architectural diagram showing the arrangement of a data processing method according to the present invention.

Fig. 2 is a flow chart illustrating an amplification layer preparation process.

FIG. 3 is a flowchart illustrating noise suppression processing.

FIG. 4 is a flowchart illustrating local enhancement processing.

5( a )-( d ) are comparison diagrams showing a plurality of input images and the respective processed images.

Figure 6 is a single parameter control chart showing α max.

FIG. 7 is a time complexity diagram showing the average processing time of the disclosed data processing method.

FIG. 8 is a block diagram showing the image acquisition and processing system.

Fig. 9 is a diagram illustrating an image acquisition and processing system according to the present invention Architecture diagram of the detailed configuration of the system.

In order to enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly dictates otherwise. It will be further understood that when used in this specification, the terms "comprising" and/or "comprising" specify the presence of stated features, integers, steps, operations, elements and/or components, but do not exclude the presence or addition of A group of one or more other features, integers, steps, operations, elements, components, and/or the like.

In describing the example embodiments illustrated in the drawings, specific terminology will be employed for the sake of clarity. However, the disclosure is not intended to be limited to the specific terms so selected, and it is to be understood that each specific element includes all technical equivalents that have the same structure, operate in a similar manner, and achieve a similar result.

In the following description, the illustrative embodiments are described with reference to acts and symbolic representations (e.g., in the form of flowcharts) of operations that may be implemented as program modules or functional processes, including routines, programs, objects, components , data structures, etc., which perform specific tasks or implement specific abstract data types, and can use existing network elements or existing hardware on control nodes to fulfill. Such existing hardware may include one or more central processing units (Central Processing Unit, CPU), digital signal processors (Digital Signal Processor, DSP), application-specific integrated circuits (Application-Specific-Integrated-Circuit, ASIC) , Field Programmable Gate Array (Field Programmable Gate Array, FPGA), computer and so on. These terms can often be collectively referred to as a processor.

Unless expressly stated otherwise, or apparent from the discussion, terms such as "processing" or "computing" or "computing" or "determining" or "displaying" refer to the actions and actions of a computer system or similar electronic computing device Processing, which manipulates and converts data represented as physical, electronic quantities in the computer system's temporary registers and memories, and converts them into computer system memory or temporary registers or other such information for storage, transmission or Displays other information similarly represented as physical quantities in a device.

According to the present invention, the input images are of a large Field-of-View (FOV) of brain/neuronal structures at multiple excitation wavelengths using a custom-developed Multiphoton Optical Microscopy (MPM) system. Obtained by two-photon fluorescence imaging satisfying the Nyquis theorem (no aliasing). The method disclosed in the present invention successfully retrieves the weak-intensity hyperfine neuron structure polluted by the strong noise background. The disclosed method can simultaneously improve Signal-to-Noise Ratio (SNR), Signal-to-Background Ratio (SBR) and contrast. Accelerated by the NVIDIA Compute Unified Device Architecture (CUDA) assisted by a Graphics Processing Unit (GPU), the disclosed method can be typically 1000 A 16-bit unsigned image of size ×1000 achieves a time complexity of less than 3 milliseconds.

Data processing method disclosed in the present invention: FIG. 1 illustrates an overview of a data processing method according to an embodiment of the disclosed document. FIG. 2 illustrates a flow chart of a process of obtaining a first enlarged layer image according to an embodiment of the present disclosure. FIG. 3 illustrates a flowchart of noise suppression processing according to an embodiment of the present disclosure. FIG. 4 illustrates a flowchart of local enhancement processing according to an embodiment of the present disclosure.

As shown in FIG. 1 , the input image INPUT is performed by the enlargement layer preparation process S01 to obtain the first enlargement layer image LAY1 . The first enlarged layer image LAY1 is subjected to noise suppression processing S02 to obtain a noise suppressed version SUB3. The noise suppression version SUB3 is performed using local enhancement processing S03 to obtain a processed image as output.

INPUT is a low-contrast 16-bit image contaminated by noise, which can be expressed as f ( r,c ) by R×C pixels, where r and c represent the position of the column and row, respectively.

According to process S01, a 10-fold downscaling is first applied to f ( r,c ), and a first resized image fD ⁽ r ^/ ,c ^/ ) is obtained with a reduced number of pixels R ^/ ×C ^/ . See equation (1) and the first resized image RES1 in FIG. 2 . For all r ^/ xc ^/ pixel images, r ^/ and c ^/ represent column and row positions, respectively. f ^D ( r ^/ , c ^/ ) is low-pass filtered by a 29×29 kernel Gaussian blur to generate the first blurred image. Please refer to the first blurred image BLR1 in FIG. 2 . Each pixel of the first blurred image BLR1 is incremented by 1.0 to avoid division by zero in subsequent steps. See l ( r ^/ , c ^/ ) in Eq. (2) and the additive layer ADD in Fig. 2. l ( r ^/ ,c ^/ ) is rescaled back to R×C pixels by bilinear interpolation, thereby generating a second resized image. See ^lU ( r ,c ) in equation (3) and the second resized image RES2 in FIG. 2 . The reciprocal of each l ^U ( r,c ) pixel value is multiplied by 90% of the maximum allowable intensity, which is 0.9×(2 ¹⁶ -1) of the 16-bit image, to generate the first segmented layer image. Please refer to d ( r,c ) in formula (4) and the first split layer image DIV in FIG. 2 . Pixel values in d ( r,c ) higher than the user-defined threshold α _max will be truncated to α _max to generate the first zoom layer image. Please refer to α ( r,c ) in formula (5) and the first enlarged layer image LAY1 in FIG. 2 . The first enlarged layer image LAY1 will be used for the following noise suppression processing S02. In an embodiment, the user-defined threshold α _max ranges from 3.0 to 8.0, allowing floating point numbers.

According to the process S02, INPUT or f ( r,c ) is firstly multiplied pixel by pixel with the first enlarged layer image LAY1 or α ( r,c ) to generate the first enlarged image. Please refer to the first enlarged image AMP1 in FIG. 3 and g ( r,c ) in formula (6). g ( r,c ) is subjected to a 3× downscaling to produce a third resized image. Please refer to the third resized image RES3 in Fig. 3 and g ^D ( r ^// , c ^// ) in Equation (7), where r ^// and c ^// represent the positions of columns and rows, respectively. g ^D ( r ^// ,c ^// ) means a reduction in the number of pixels R ^// ×C ^// . g ^D ( r ^// , c ^// ) uses a 29×29 core Gaussian blur to perform low-pass filtering to generate the second blurred image BLR2. Please refer to the second blurred image BLR2 in FIG. 3 and L ( r ^// , c ^// ) in formula (8). L ( r ^// ,c ^// ) is resized back to R×C pixels by bilinear interpolation to generate a fourth resized image. Please refer to the fourth resized image RES4 in FIG. 3 and ^LU ( r,c ) in equation (9). ^Subtract g (r, c ) from LU ( r,c ) to generate a first subtraction image SUB1. Please refer to the first subtraction image SUB1 in FIG. 3 . The first subtraction image SUB1 uses a 7×7 kernel Gaussian blur to perform low-pass filtering to generate a third blurred image BLR3. Please refer to the third blurred image BLR3 in FIG. 3 and L ^/ ( r,c ) in formula (10). α ( r, c ) uses subtraction, that is, [1.25× α _max - α ( r, c )] to obtain the second subtraction image SUB2 or the second magnification layer image LAY2. Please refer to the second subtraction image SUB2 or the second magnification layer image LAY2 in FIG. 3 . Then, L ^/ ( r,c ) or the third blurred image BLR3 is multiplied pixel by pixel with the second subtraction image SUB2 or the second magnification layer image LAY2 to generate the second magnification image AMP2. Please refer to the second enlarged image AMP2 in FIG. 3 . Subtract the second enlarged image AMP2 from f ( r,c ) or INPUT to obtain the third subtracted image SUB3. Please refer to the third subtraction image SUB3 in FIG. 3 and S ( r,c ) in formula (11). The third subtraction image SUB3 or S ( r,c ) is a noise-suppressed version of INPUT or f ( r,c ), which will be used in the subsequent local enhancement processing S03.

g(r,c)=f(r,c)×α(r,c) formula (6)

S ( r,c )= f ( r,c )- L ^/ ( r,c )×[1.25× α _max - α ( r,c )] formula (11)

According to process S03, the first magnification layer image LAY1 or α ( r,c ) is performed through a set of arithmetic operations, given as [ X +{ α ( r,c )/ Y } ⁿ ], to generate the third magnification layer Image LAY3. Please refer to the third enlarged layer image LAY3 in Figure 4. In one embodiment, the values of X, Y and n are selected as 0.9, 4.0 and 2.0, respectively. The third subtraction image SUB3 (ie S ( r, c )), that is, the third subtraction image SUB3 is multiplied pixel by pixel by the third amplification layer image LAY3 to generate a third amplification image AMP3 or a noise-suppressed contrast-enhanced output image . See OUTPUT in Figure 4 and F ( r,c ) in Equation (12).

F ( r,c )= S ( r,c )×[0.9+{ α ( r,c )/4.0} ^2.0 ] formula (12)

A region of high intensity in INPUT will cause the value of the corresponding region in the third magnification layer image LAY3 to be close to 1, thereby preventing saturation in OUTPUT when the third subtraction image SUB3 is multiplied by the third magnification layer image LAY3. And the low-intensity area in INPUT will cause the value of the corresponding area in the third magnification layer image LAY3 to be greater than 1, so when the third subtraction image SUB3 is multiplied by the third magnification layer image LAY3, the low-intensity area in OUTPUT is locally enhanced.

The application of the disclosed data processing method uses two-photon excitation fluorescence (TPEF) images of Nav1.8-tdTomato positive mouse dorsal root ganglion (Dorsal-Root-Ganglion, DRG) parts and images from Thy1 - as evidenced by coronal sections of cortical regions of GFP-positive mice. DRG partially organized by bright body The cortical part consists of axons, dendrites, and dendritic spines. For Nav1.8-tdTomato and Thy1-GFP positive specimens, TPEF imaging was performed at the central excitation wavelengths of 1070nm and 919nm (70MHz, <60fs, <40mW average excitation power), respectively.

Figures 5(a) and 5(c) depict two TPEF images of Nav1.8-tdTomato and Thy1-GFP specimens, respectively, each with a scale bar of 150 μm. Both Figure 5(a) and 5(c) show poor SNR, SBR and contrast. Applying the data processing method disclosed in the present invention, the image in Fig. 5(a) has α _max =8.0, the image in Fig. 5(c) has α _max =6.0, and the processed images are shown in Fig. 5(b) and 5(d). The disclosed data processing methods produce significant image quality improvements in each case.

Figure 6 provides two regions of interest (Regions-of-Interest, ROI) of the Nav1.8-tdTomato image and two regions of interest of the Thy1-GFP image, respectively taken from Figure 5(a) and 5(c) , aiming to quantitatively visualize the effect of α _max . The first column (i.e., INP in Figure 6) depicts the unprocessed region of interest. The next column depicts the processed region of interest for different values of αmax _sequentially varying from 3.0 to 8.0.

Figure 7 plots the average processing time (in milliseconds) of the disclosed data processing method against the input image size (16-bit unsigned format). The average processing time consists of uploading the raw input image from the host to the GPU, processing the uploaded image in the GPU, and downloading the processed output image from the GPU to the host. The first curve above (the one with the squares) plots the average processing time through a traditional CPU i7-9800X, consuming a maximum of about 2000 ms for a 10,000 x 10,000 pixel 16-bit input image. The second and third curves (the one with the triangle and the one with the sphere) plot the average processing time across two CUDA-capable GPUs (Quadro P1000 and Quadro RTX 8000, respectively), each showing the difference in processing speed. Significantly improved. The Quadro RTX 8000 consumes about 111 milliseconds on the same 10,000×10,000 pixel 16-bit input image, pointing to an 18x performance improvement over the i7-980X.

In one embodiment, an image acquisition and processing system for fast noise-suppressed contrast enhancement accelerated by a graphics processing unit with digitally implemented hardware-based automation in optical linear or nonlinear microscopy applications is provided. Adaptive/controlled lighting effects while ensuring a time complexity of less than 3 milliseconds for a 1000×1000 pixel 16-bit input image.

FIG. 8 is a schematic diagram showing the image acquisition and processing system 100 . The image acquisition and processing system 100 includes a computer 110 with a display card supporting Compute Unified Device Architecture (CUDA), and an optical linear or nonlinear microscope system configured to acquire images of biological specimens with high digital resolution 120.

FIG. 9 is a block diagram showing the configuration of the image acquisition and processing system 100 . Images acquired by the optical linear or nonlinear microscopy system 120 are provided as input and processed with a data processing method dedicated to fast noise suppression contrast enhancement that digitally mimics hardware-based adaptive/responsive control lighting technology. The data processing method includes configuring a CUDA-supported graphics processing unit of the computer 110 to perform the following steps.

A1: Obtaining a noise-affected low-contrast input image from an optical linear or nonlinear microscopy system, wherein the input image has a first width and a first height, and includes a plurality of pixels or data points having a specific bit depth that allows a maximum pixel value or maximum pixel intensity obtained by raising 2 to the power of the bit depth minus 1 .

The amplification layer preparation process S01 includes steps A2 to A7.

A2: Perform a first bilinear interpolation process on the input image to resize the input image by a first downscaling factor of 10 to generate a first resized image having a width less than the first width A second width of the first reduction factor, and a second height less than the first height by the first reduction factor.

A3: Performing a first Gaussian blurring process on the first resized image by performing a convolution operation on the first resized image using a Gaussian kernel with a first kernel size of 29×29 to obtain a first blurred image .

A4: Perform an addition process to add a non-zero number (ie 1.0) to the first blurred image to obtain a non-zero image.

A5: Performing a second bilinear interpolation process on the non-zero image to upscale it from a second width and a second height to obtain a second resized image having the first width and the first height.

A6: A segmentation process is performed, dividing 90% of the maximum pixel intensity by the second resized image to obtain the first segmented layer image.

A7: Perform threshold processing to truncate the first segmented layer image at a user-defined threshold value ranging from 3.0 to 8.0 to obtain the first enlarged layer image.

The noise suppression processing S02 includes steps A8-A16.

A8: Perform the first enlargement process on the input image to convert the input The image is multiplied by the first magnified layer image to generate a first magnified image.

A9: Perform a third bilinear interpolation process on the first enlarged image to resize the first enlarged image by a second downscaling factor of 3 to generate a third resized image having a ratio The first width is less than a third width by the second reduction factor, and the third height is less than the first height by the second reduction factor.

A10: Performing a second Gaussian blur on the third resized image by convolving the third resized image with a second Gaussian kernel of size 29×29 to obtain a second blurred image .

A11: performing a fourth bilinear interpolation process on the second blurred image, and enlarging it from the third width and the third height to obtain a fourth resized image having the first width and the first height.

A12: performing a first subtraction process on the fourth resized image and the first enlarged image, so as to subtract the first enlarged image from the fourth resized image to generate a first subtracted image.

A13: Convolving the first subtraction image with a third Gaussian kernel with a size of 7×7 to perform a third Gaussian blurring process on the first subtraction image to obtain a third blurred image.

A14: performing a second subtraction process on the first magnified layer image to subtract the first magnified layer image from a user-defined threshold of 1.25 times to obtain a second subtracted image or a second magnified layer image.

A15: performing a second zoom-in process on the third blurred image, so as to multiply the third blurred image by the second zoom-in layer image to generate a second zoom-in image.

A16: performing a third subtraction process on the input image and the second enlarged image, so as to subtract the second enlarged image from the input image to obtain a third subtracted image.

The local enhancement process S03 includes steps A17 and A18.

A17: Perform a set of arithmetic processing on the first magnification layer image. First, divide the first enlarged layer image by a division factor of 4 to obtain a second divided layer image. Second, the second segmented layer image is raised to the power of 2 to obtain the modified segmented layer image. And thirdly, a value of 0.9 is added to the modified segmented layer image to obtain a third enlarged layer image.

A18: Performing a third enlargement process on the third subtraction image, so as to multiply the third subtraction image by the third enlargement layer image, and generate a third enlargement image or an output image with enhanced contrast.

In short, instead of hardware-based solutions, this disclosure reports here a dedicated digital method for fast noise-suppression contrast enhancement that digitally mimics optical feedback-driven in linear or nonlinear microscopy applications Adaptive/controlled lighting technology.

All documents mentioned in this application are herein incorporated by reference to the same extent as if each individual document was specifically and individually indicated to be incorporated by reference. In addition, it should be understood that those skilled in the art can make various modifications and changes to the present invention after reading the teachings of the present invention, and these equivalents also fall within the scope defined in the claims.

S01-S03: Processing

LAY1: The first enlarged layer image

SUB3: The third subtraction image

Claims (13)

A data processing method for fast noise suppression contrast enhancement, comprising configuring a graphics processing unit to perform the following steps: obtaining an input image, wherein the input image has a first width and a first height and includes a plurality of pixels or data points having a specific bit that allows a maximum pixel value or a maximum pixel intensity meta depth; performing a first binning or a first interpolation process on the input image, resizing the input image by a first downscaling factor to generate a first resized image having a ratio a second width less the first width by the first reduction factor, and a second height less the first height by the first reduction factor; performing a first low-pass filtering process on the first resized image to obtain a first blurred image; performing a second interpolation process on the first blurred image, upscaling it from the second width and the second height to obtain a second resized image having the first width and the first height; performing a segmentation process of dividing a first specified number by the second resized image to obtain a first segmented layer image; performing a threshold value processing to truncate the first segmented layer image at a user-defined threshold value to obtain a first enlarged layer image; performing a first magnification process on the input image to multiply the input image by the first magnification layer image to generate a first magnification image; performing a second binning or a third interpolation process on the first enlarged image, and resizing the first enlarged image by a second reduction factor to generate a third resized image, the third resized the size image has a third width less than the first width by the second downscaling factor, and a third height less than the first height by the second downscaling factor; performing a second low-pass filtering process on the third resized image to obtain a second blurred image; performing a fourth interpolation process on the second blurred image, upscaling it from the third width and the third height to obtain a fourth resized image having the first width and the first height; performing a first subtraction process on the fourth resized image and the first enlarged image to subtract the first enlarged image from the fourth resized image to generate a first subtracted image; performing a third low-pass filtering process on the first subtracted image to obtain a third blurred image; performing a second subtraction process on the first magnified layer image to subtract a second specific number higher than the user-defined threshold value from the first magnified layer image to obtain a second subtracted image or a first Two enlarged layers of images; performing a second zoom-in process on the third blurred image to multiply the third blurred image by the second zoom-in layer image to generate a second zoom-in image; performing a third subtraction process on the input image and the second enlarged image to subtract the second enlarged image from the input image to obtain a third subtracted image; performing a set of arithmetic processing on the first magnification layer image to obtain a third magnification layer image; and performing a third magnification process on the third subtraction image to multiply the third subtraction image by the first Three upscaled images to generate a third upscaled image or a contrast-optimized output image. The method as claimed in claim 1, wherein the first interpolation process, the second interpolation process, the third interpolation process and the fourth interpolation process are all bilinear. The method according to claim 1, wherein the first reduction factor in the first binning or the first interpolation process is 10. The method of claim 1, wherein the first low pass filtering process involves performing a Gaussian blur operation of convolution using a Gaussian kernel with a first kernel size of 29×29. The method of claim 1, wherein before performing the second interpolation process, an addition process is performed on the first blurred image to add a non-zero number to the first blurred image. The method of claim 1, wherein the first specified number in the segmentation process is 90% of the maximum pixel intensity. The method as recited in claim 1, wherein the user-defined threshold in the threshold processing ranges from 3.0 to 8.0, allowing floating point numbers. The method as claimed in claim 1, wherein the second downscaling factor in the second pixel binning or the third interpolation process is 3. The method of claim 1, wherein the second low pass filtering process involves using a Gaussian kernel with a second kernel size of 29×29 to perform a Gaussian blur operation of convolution. The method of claim 1, wherein the third low-pass filtering process involves using a Gaussian kernel with a third kernel size of 7×7 to perform a Gaussian blur operation of convolution. The method according to claim 1, wherein the second specific number in the second subtraction process is 1.25 times the user-defined threshold. The method according to claim 1, wherein the step of performing the set of arithmetic processing on the first magnified layer image to obtain the third magnified layer image comprises: First, dividing the first enlarged layer image by a division factor of 4 to obtain a second divided layer image; Second, raising the second segmented layer image to a power of 2 to obtain a modified segmented layer image; and Thirdly, a value of 0.9 is added to the modified segmented layer image to obtain the third enlarged layer image. An image acquisition and processing system, comprising a device for performing the method described in any one of Claims 1-11.

Images