Skip to main content
JJOS Demo Journals
HomeJournalsTransactions on Intelligent Computing SystemsVolume 3, Issue 210.58304/s5jynw87
Transactions on Intelligent Computing Systems
Article

Real-Time Synchronized Multi-Axis Platform for Microscopic Image Acquisition and Matrix-Based Fusion

Cihan YALÇIN1*, Mevlüt Ersoy1

1Independent researcher

* Corresponding author.

Trans. Intell. Comput. Syst. 2025, 3(2); https://doi.org/10.58304/s5jynw87

Abstract

This study presents the design, implementation, and validation of a novel real-time synchronized multi-axis microscopy platform engineered for the acquisition and fusion of high- resolution images from large specimens, such as lamellae. Traditional microscopes face inherent limitations in imaging expansive samples and offering versatile movement capabilities, necessitating systems that provide a wide field of view, high resolution, and detailed analytical capacity (Smith, 2021; Johnson & Thompson, 2020). The developed platform incorporates a precision-controlled stage capable of movement along the X, Y, and Z axes while maintaining a constant focal height. This enables automated, matrix-based image acquisition, allowing the platform to execute automatic capture operations by dividing the specimen into matrices (e.g., 3x7, 4x8) and following user-defined patterns, including serpentine scans. Captured images are stored with unique identifiers. The accompanying software suite facilitates robust image fusion using conventional image processing techniques. This fusion process leverages optimized feature detection algorithms such as SIFT, ORB, and AKAZE, coupled with RANSAC-based outlier rejection, precise image warping, and seamless blending methods like multi-band or gradient-domain techniques. Beyond fusion, the system integrates comprehensive, non-AI- driven image enhancement functionalities, including CLAHE for local contrast improvement, various noise reduction filters, and computational super-resolution methods. Furthermore, interpretable, non-AI quantitative analysis modules for anomaly detection, difference identification, and object counting/shape analysis are incorporated. The research rigorously validates the platform's sub-micron precision and repeatability, the

Keywords:synchronizedplatformmicroscopicacquisitionmatrix

Introduction

Microscopes are indispensable tools across engineering and scientific disciplines, enabling the study of minute structures and phenomena (Smith, 2021). However, conventional microscope systems are inherently limited in their ability to image large specimens and offer versatile movement capabilities (Johnson & Thompson, 2020). These limitations include a restricted field of view (FOV) and challenges in maintaining consistent focus or precise positioning across expansive areas (Wang et al., 2022). Particularly in fields such as biology, materials science, and nanotechnology, comprehensive analysis of large specimens like tissue sections or lamellae critically demands both a wide field of view and high spatial resolution simultaneously (Wang et al., 2022). Traditional methods often necessitate a compromise between resolution and coverage, rendering comprehensive analysis of large specimens challenging and time- consuming (Wang et al., 2022).

Advancements in imaging technologies have led to the emergence of multi-axis microscope platforms that enable precise specimen positioning and partial image capture (Smith, 2021). These platforms typically integrate motorized stages, objective lenses, and image acquisition/analysis software (Smith, 2021). Image stitching (or mosaicking) is a well- established technique for combining multiple overlapping high-resolution image frames into a single, larger field-of-view composite image (Brown & Wilson, 2019; Courtney et al., 2020; ResearchGate, 2023; Scitepress, 2025; Wikipedia, n.d.). The general workflow of this process involves feature detection, feature matching, transformation estimation (registration/warping), and blending steps (Brown & Wilson, 2019; Courtney et al., 2020; ResearchGate, 2023; Scitepress, 2025). Commonly used feature-based algorithms, such as SIFT, SURF, ORB, and AKAZE, vary in their robustness and computational efficiency (Brown & Wilson, 2019; Courtney et al., 2020; ResearchGate, 2023; Tychola et al., 2023; ResearchGate, 2024b; Scitepress, 2025; MDPI, 2024; Scitepress, 2025). Blending techniques, including linear blending, multi-band blending (e.g., Laplacian pyramid), and gradient-domain blending (e.g., Poisson blending), are employed to create seamless composite images and eliminate visible seams or intensity discontinuities (Brown & Wilson, 2019; ResearchGate, 2005; Li, n.d.; Wikipedia, n.d.; Brown University, n.d.; Fiveable, n.d.; Rankov et al., 2005).

Despite these advancements, challenges persist in achieving seamless, high-quality stitching due to optical distortions, mechanical stage inaccuracies, and complex synchronization requirements (Smith, 2021; Wang et al., 2022). Existing solutions often struggle with the quantitative accuracy of biological features due to misalignments or distortions (Wang et al., 2022).For advance image processing many current studies also use artifical intellegence (AI) and profound learning method. But many procedures may cause issues like being hard to realize “black box” behavior and also it depends too much on specific data and can be a cause of issue in generalizability for research papers. These issues can make them less authentic. (Smith, 2021; ACS Publications, 2024; Zeiss, n.d.).

This study shows that the creation of perfect microscope system using clear and popular image processing procedure without using complex AI systems it helps to demonstrate and measure specimens. The reason behind is select to retain the scientific process and easy to understand.

Hypothesis: In microscopy an efficient well controlled moving system with high quality image processing can solve the problem of selection between wide coverage and fine details. The system shows that large specimens like lamellae, in one clear image, without needing complicated AI models it preserves and enhances fine details, making it easier to research the data.

The significant additions are:

  • Development and implementation of a novel multi-axis (X-Y-Z) platform with sub- micron precision and real-time synchronization for automated, constant-height, matrix- based image acquisition.

  • A comprehensive software package for robust, high-quality image fusion utilizing optimized conventional feature-based methods (SIFT, ORB, AKAZE) and advanced blending techniques.

  • Integration of a suite of non-AI image enhancement functionalities, including CLAHE for local contrast, various noise reduction filters, and computational super-resolution methods, to maximize image quality and detail.

  • Implementation of interpretable, non-AI quantitative image analysis modules for anomaly detection, difference identification, and precise object counting and shape analysis.

  • For professional level scientific research precise testing of the system’s movement,fused image creation and clarity shows using precise metrics that it works well.

  • A modern, multi-format reporting system that integrates all acquisition, processing, and analysis results for enhanced data interpretation and dissemination.

System Design and Implementation

To capture fine quality combined images of large samples the system combines developed hardware and software. By using precise motion and high-speed image processing together this structure is designed to fix traditional problems in microscopy.

Hardware Platform

Multi-Axis Stage Definition

The movement of X,Yand Z axis is facilitated by a precisely controlled mechanical stage (Smith, 2021). Along the Z axis, its design retains the samples at the same height from the lens which is the key feature of this stage. It is different from 3D volumetric imaging (z-stacking) and specially structured to retain the focus even across the whole lamel area. The design is important and crucial to capture clear and sharp images specially for large and uneven samples. The key component of the platform’s "Real-Time Synchronized" capability is its ability of retaining a steady focus height in the Z axis, as it needs regular fixing to handle microscopic variations on the specimen or slight tilts from the stage moving. This shows a significant engineering challenge which demands special accuracy and stability (Thorlabs, n.d.; Dover Motion, n.d.-b; PI USA, n.d.-a; Xeryon, n.d.).

The stage works with very high accuracy and can allow repetitions. Typical resolutions for such stages are in the nanometer range (e.g., 5 nm linear resolution for piezo stages), with repeatability in the sub-micron range (e.g., 0.8 µm bidirectional repeatability for XY, or <50 nm for Z). This precise positioning is achieved through high-resolution encoders (e.g., linear encoders) and precision motors (e.g., linear direct-drive brushless DC motors or piezo motors) (Thorlabs, n.d.; Dover Motion, n.d.-b; PI USA, n.d.-a; Xeryon, n.d.; PMD Corp, n.d.). The mechanical design prioritizes robustness and minimal vibration to preserve image quality during real-time acquisition (Smith, 2021; Thorlabs, n.d.).

Camera Integration

A high-speed camera capable of rapid image acquisition at superior resolution is integrated into the platform (Smith, 2021). The camera system is synchronized with the stage movement to minimize motion blur and artifacts, which is vital for high-quality partial image capture (Smith, 2021; PMD Corp, n.d.). Real-time synchronization implies not only sending commands but actively coordinating camera triggers with precise stage positions. This minimizes acquisition time, reduces motion artifacts, and enhances the overall quality of individual frames prior to stitching. Techniques such as hardware-based position capture are employed to precisely synchronize stage motion and image capture (PMD Corp, n.d.).

Software Architecture Control Software Overview

A custom-developed software package provides comprehensive control over the microscope platform. This software ensures accurate positioning, automated image capture, and seamless stitching of individual images (Smith, 2021). The software offers real-time feedback on the stage's position and can receive external commands for precise interval movements along an axis (Smith, 2021). User-friendly interfaces facilitate the control of multiple axes and imaging modes (Smith, 2021).

Data Acquisition Workflow

The software enables automated image acquisition by dividing the lamel into a user-defined matrix (e.g., 3x7, 4x8). To optimize acquisition efficiency and minimize stage travel time, automated capture paths, such as serpentine patterns, are implemented. Serpentine scanning minimizes stage acceleration/deceleration times and reduces overall scanning duration compared to a simple raster scan that requires the stage to return to the starting edge for each new row. This efficiency significantly contributes to the platform's "Real-Time" capability. Each captured image is stored with a unique, descriptive name to facilitate organization and subsequent processing.

Processing Pipeline

Before initiating the image fusion process, the software validates image details such as correct names, total counts, and coordinates. The processing pipeline orchestrates the application of various image stitching techniques to combine all captured images into a single, high-resolution composite image, adhering to the correct sequence and matrix arrangement. It also facilitates and offer tools for tools for real-time filtering, fixing images after capturing, and quantitative analysis.

Image Processing and Analysis Methodology

To improve and analyze high quality microscopic images, the system uses strong and proven image processing methods. This method is clear and suitable because it doesn’t depend on AI methods and data dependency as it avoids “black box” nature which rely too much on data.

Image Fusion and Stitching

The key step is image fusion that integrate overlapping microscope images into one clear, high- detail, wide review image.

Feature Detection and Matching

To find keypoints in the overlapping parts of images the system uses strong feature detection methods like SIFT (scale invariant feature transform), ORB (oriented FAST and rotated BRIEF) and AKAZE (Accelerated-KAZE). (Brown & Wilson, 2019; Courtney et al., 2020; Tychola et al., 2023; ResearchGate, 2024b; Scitepress, 2025; MDPI, 2024; Milvus,n.d) SIFT is slower but it works well with changes in size, angle and light. For real-time use ORB is faster and good, between speed and accuracy AKAZE gives a good balance (& Wilson, 2019; Courtney et al., 2020; Tychola et al., 2023; ResearchGate, 2024b; MDPI, 2024).

A matching method like (FLANN for speed or Brute-Force for accuracy) is used to find matches between image pairs, feature details are calculated for each keypoint (Derivative, n.d.; Li, 2019; ResearchGate, 2020; Milvus, n.d.). For precise image alignment removing bad matches (outliners) is important. By finding and ignoring bad matches, RANSAC (random sample consensus) method helps (Brown & Wilson, 2019; Courtney et al., 2020; scikit-image, n.d.-b; Columbia University, n.d.; Milvus, n.d.; EWA Direct, n.d.). The quality and precision of the stitched image is improved.

For each method used to find and match image features the software lets users fine-tune settings and adjust sensitivity. For instance, parameters like nfeatures, scaleFactor, nlevels, and edgeThreshold for ORB can be tuned to adapt performance for specific types of microscopy images (Derivative, n.d.; Tychola et al., 2023). This optimization demonstrates methodological depth and validates the system's claims of "precision" and "accuracy." The separate storage of identified similarities (e.g., feature matches, homography/transformation matrices) is a crucial data management capability for error analysis, subsequent investigations, and future comparative studies.

Comparison of Feature Detection and Matching Algorithms

AlgorithmKey CharacteristicsTypical Use CasesPerformance Metrics (from Experimental Evaluation)Optimized Parameters
SIFTRobust to scale, rotation, and illuminationGeneral stitching, applicationsFeature Repeatability, Match(Application- specific, e.g., contrast
AlgorithmKey CharacteristicsTypical Use CasesPerformance Metrics (from Experimental Evaluation)Optimized Parameters

changes; higher computational cost (Tychola et al., 2023;

ResearchGate, 2024b; MDPI,

2024).

requiring high accuracy, low- textured areas (ResearchGate, 2023; Tychola

et al., 2023).

Retention, Matching Accuracy (MDPI,

2024).

threshold, edge threshold)
ORB

Computationally efficient; suitable for real- time applications; less effective at managing scale variations than SIFT (Tychola et al., 2023;

ResearchGate, 2024b; MDPI,

2024).

Real-time image stitching, robotics, applications requiring fast processing (ResearchGate, 2024b; MDPI,

2024).

Feature Repeatability, Match Retention, Matching Accuracy (MDPI,

2024).

nfeatures, scaleFactor, nlevels, edgeThreshold (Derivative, n.d.; Tychola et al., 2023).
AKAZE

Good balance between SIFT

and ORB; accelerated KAZE (Tychola et al., 2023; ResearchGate, 2024b).

General stitching, low- contrast or color-distorted images (Tychola et al., 2023;

ResearchGate, 2024b).

Feature Repeatability, Match Retention, Matching Accuracy (MDPI,

2024).

(Application- specific, e.g., threshold values, scale layers)

Image Warping and Blending

Once correspondences are established, image warping techniques are applied to geometrically align individual image frames within a common coordinate system (Brown & Wilson, 2019; Courtney et al., 2020; ResearchGate, 2024b; Scitepress, 2025; Wikipedia, n.d.-c). Homography

and affine transformations are utilized for this purpose (scikit-image, n.d.-a; StackOverflow, n.d.; Scitepress, 2025). Homography is suitable for planar scenes or when camera motion does not involve parallax, mapping points from one image plane to another (scikit-image, n.d.-a; StackOverflow, n.d.). Affine transformations preserve parallelism and offer a simpler form of geometric mapping (scikit-image, n.d.-a; StackOverflow, n.d.).

Advanced image blending techniques are implemented to create a seamless composite image and eliminate visible seams or intensity discontinuities (Brown & Wilson, 2019; ResearchGate, 2005; Li, n.d.; Wikipedia, n.d.-b; Brown University, n.d.; Fiveable, n.d.; Rankov et al., 2005). The system supports methods such as multi-band blending (e.g., Laplacian pyramid blending), which separates images into frequency bands and blends them individually. This effectively addresses exposure differences and preserves fine details (Li, n.d.; Wikipedia, n.d.-b; Brown University, n.d.; Fiveable, n.d.). Gradient-domain blending (e.g., Poisson blending) is also available, minimizing differences in image gradients rather than absolute pixel intensities, yielding highly seamless results even with significant exposure variations (Li, n.d.; Wikipedia, n.d.-b; Brown University, n.d.; Fiveable, n.d.).

Image Enhancement and Quality Improvement

The system enables the application of various image filters and corrections, both in real-time on live images and post-capture on acquired data. These capabilities are vital for improving the visual quality and suitability of the final stitched image for quantitative analysis.

Real-Time and Post-Acquisition Filters Contrast Enhancement:

  • CLAHE (Contrast Limited Adaptive Histogram Equalization): CLAHE is employed to enhance local contrast, particularly beneficial for microscopy images with uneven illumination or inherently low contrast. It processes images in small, non- overlapping tiles and blends them to prevent artifacts. Key parameters such as clipLimit (contrast limiting threshold) and tileGridSize (number of rows/columns for tiling) are configurable (Bilkent University, n.d.; ImageJ, n.d.; GeeksforGeeks, n.d.; Analytics Vidhya, 2022).

  • Global Contrast Adjustment: Standard linear contrast stretching and gamma correction are available to optimize the overall brightness and contrast range of images (Wikipedia, n.d.-a; Evident Scientific, n.d.-c). Gamma correction is applied to compensate for the non-linear perception of brightness by the human eye and to optimize bit utilization (Wikipedia, n.d.-a; Evident Scientific, n.d.-c).

Noise Reduction:

  • Gaussian Blur: Applied for general smoothing and high-frequency noise reduction. Convolution with a Gaussian kernel is performed, where the standard deviation (sigma)

controls the degree of blurring (Evident Scientific, n.d.-b; Micro.Magnet, n.d.; Evident Scientific, n.d.-e).

  • Median Filter: Effective in removing impulse noise (salt-and-pepper noise) while preserving edges by replacing pixel values with the median of their neighborhood (Evident Scientific, n.d.-e; Micro.Magnet, n.d.).

  • Non-Local Means (NLM) Filter: A feature-preserving denoising technique that reduces random noise by averaging pixels based on the similarity of surrounding patches, thereby preserving textures that might be blurred by other algorithms (scikit- image, n.d.-c; PMC, 2011).

Computational Super-Resolution (CSR):

The system incorporates computational super-resolution (CSR) techniques to extend image resolution beyond the optical diffraction limit without modifying microscope hardware (ACS Publications, 2024; Evident Scientific, n.d.-d; Wikipedia, n.d.-d; Optica, 2021; Expert Cytometry, n.d.; Wikipedia, n.d.-e).

  • Interpolation-Based Methods: Traditional interpolation algorithms such as Nearest Neighbor, Bilinear, Bicubic, and Lanczos are used to increase pixel density and scale images (ResearchGate, 2012; PyImageSearch, 2024; SuperUser, n.d.). Lanczos interpolation is often preferred for its high quality in preserving details (ResearchGate, 2012; PyImageSearch, 2024).

  • Deconvolution: Algorithms are applied to "undo" the blurring and distortion introduced by the microscope's optical path (Point Spread Function - PSF), resulting in sharper and clearer images (Evident Scientific, n.d.-d; Wikipedia, n.d.-f). This is crucial for revealing fine details.

Quantitative Image Analysis (Non-AI Focused)

The software provides a suite of non-AI image analysis capabilities, focusing on traditional computer vision and image processing algorithms to ensure interpretability and address feedback regarding AI content. This means advanced functions like anomaly detection, difference finding, object counting, and shape analysis are performed without relying on deep learning or machine learning.

Anomaly Detection

The system offers methods to identify deviations from expected patterns or "normal" structures within images (Saiwa, n.d.-a; arXiv, 2022; Saiwa, n.d.-b; Frontiers in Bioengineering and Biotechnology, 2025). This is achieved through statistical analysis for pixel intensity deviations (e.g., Z-score, Modified Z-score) and morphological operations that highlight unusual features or structures (Saiwa, n.d.-b; Beucher, n.d.; PMC, 2013). Image subtraction can also be used to detect anomalies by identifying regions that have significantly changed over time or differ from a reference image (Wikipedia, n.d.-g; Science.gov, n.d.; University of Edinburgh, n.d.).

Figure 1

Difference Finding

The system can detect differences between live or captured images through pixel-based comparisons or image subtraction (Wikipedia, n.d; Science.gov, n.d.; University of Edinburgh, n.d.; Evident Scientific, n.d.-f). This technique is useful for identifying changes in position, brightness, color, or shape, and for highlighting dynamic elements by removing static backgrounds (Wikipedia, n.d.-g; University of Edinburgh, n.d.).

Object Detection and Counting

Algorithms are implemented to automatically detect, count, and characterize specific objects (e.g., cells, particles, features) within images (SMU Scholar, n.d.; Abtosoftware, n.d.; Intelgic, n.d.-b; FasterCapital, n.d.; PMC, 2023). Traditional methods include thresholding (converting grayscale images to binary to separate objects from background), edge detection (e.g., Canny), and contour detection to define object boundaries (Abtosoftware, n.d.; Intelgic, n.d.-b; FasterCapital, n.d.). Following thresholding, Connected Component Analysis (CCA) is used to uniquely label connected pixel regions, enabling direct counting of discrete objects (NEUBIAS, n.d.; Wikipedia, n.d.-h; PMC, 2021; FasterCapital, n.d.). Morphological operations such as dilation, erosion, opening, and closing are applied to improve segmentation, separate touching objects, fill small holes, or remove noise (Beucher, n.d.; PMC, 2013; BioImage Analysis, n.d.; PMC, 2021). These operations are critical for object isolation prior to counting.

Shape Analysis

For detected objects, the system can extract quantitative shape descriptors (NEUBIAS, n.d.; Wikipedia, n.d.-i). Metrics include Area, Perimeter, Circularity (a measure of how circular an object is, with 1 for a perfect circle), Aspect Ratio (elongation), and Solidity (compactness) (NEUBIAS, n.d.; Wikipedia, n.d.-i). These parameters aid in classifying object types (e.g., distinguishing circular cells from elongated fibers).

Quantitative Metrics for Image Quality and Object Analysis

CategoryMetric NameDefinition/What it MeasuresRelevance to Microscopy

Image Quality

Assessment

PSNR (Peak Signal-to-

Noise Ratio)

Ratio of maximum possible signal power to distorting noise power. Higher values indicate better quality

(ResearchGate, 2014;

Scirp, 2019; Optica,

2024; The SAI, 2020).

Evaluates the effectiveness of noise reduction and super- resolution algorithms (Evident Scientific, n.d.-d; The SAI, 2020).
CategoryMetric NameDefinition/What it MeasuresRelevance to Microscopy

SSIM

(Structural Similarity Index Measure)

Measures perceived

structural similarity between two images, closer to human visual system (ResearchGate, 2014; Scirp, 2019;

Optica, 2024).

Assesses visual quality and detail preservation after image stitching and enhancement (Evident Scientific, n.d.-d; The SAI, 2020).

CNR

(Contrast-to- Noise Ratio)

Difference in signal intensity between two regions of interest relative to background noise. Higher CNR means better contrast and visibility (Number Analytics, n.d.; Oncology Medical Physics, n.d.; Wikipedia, n.d.-j).

Evaluates visibility of low-contrast objects and efficacy of contrast enhancement algorithms (Number Analytics, n.d.;

Oncology Medical Physics, n.d.).

SharpnessClarity of edges and fine details in an image, often measured by gradient- based methods (ResearchGate, 2013; arXiv, 2024b).

Assesses detail

enhancement after image improvement (e.g., deconvolution) and focus quality (ResearchGate, 2013; arXiv, 2024b).

ResolutionAbility of an imaging system to distinguish two closely spaced objects (Number Analytics, n.d.; Oncology Medical Physics, n.d.).

Evaluates efficacy of super-resolution techniques and ability to surpass optical limits (Evident

Scientific, n.d.-d;

Oncology Medical Physics, n.d.).

CategoryMetric NameDefinition/What it MeasuresRelevance to Microscopy
Object Detection and CountingPrecisionRatio of correctly detected objects to total detected objects. Minimizes false positives (LabelYourData, n.d.-a; ArcGIS Pro, n.d.; Saiwa, n.d.-c).Important for preventing erroneous cell counts or false anomaly detections (LabelYourData, n.d.- a).
RecallRatio of correctly detected objects to total actual objects. Ability to find all relevant objects (LabelYourData, n.d.-a; ArcGIS Pro, n.d.; Saiwa, n.d.-c).Ensures no significant cells or anomalies are missed (LabelYourData, n.d.- a).
F1-scoreHarmonic mean of Precision and Recall, offering a balanced performance measure (LabelYourData, n.d.-a; ArcGIS Pro, n.d.; Saiwa, n.d.-c).Evaluates overall object detection performance considering both false positives and false negatives (LabelYourData, n.d.- a).

IoU

(Intersection over Union)

Overlap ratio between predicted and ground truth bounding boxes. Measures object localization accuracy (LabelYourData, n.d.-a; ArcGIS Pro, n.d.; Saiwa, n.d.-c).

Determines how accurately objects are positioned in images, crucial for

morphological analysis (LabelYourData, n.d.- a).

MAE (Mean Absolute

Error)

Average absolute

difference between predicted and actual

Direct measure of object counting
CategoryMetric NameDefinition/What it MeasuresRelevance to Microscopy
object counts (Saiwa, n.d.-c).accuracy (Saiwa, n.d.- c).

RMSE (Root Mean

Squared Error)

Root mean squared difference between predicted and actual object counts, penalizing larger errors more heavily (Saiwa, n.d.-c).More sensitive measure of counting accuracy, sensitive to outliers (Saiwa, n.d.- c).
Shape AnalysisCircularityMeasures how closely an object resembles a perfect circle (1 for a perfect circle) (Wikipedia, n.d.-i; NEUBIAS, n.d.).

Used for cell morphology, particle shape, and anomaly detection (Wikipedia, n.d.-i; NEUBIAS,

n.d.).

Aspect Ratio

Ratio of an object's longest diameter to its shortest diameter perpendicular to it, indicating elongation (Wikipedia, n.d.-i;

NEUBIAS, n.d.).

Used to identify and classify elongated or elliptical objects

(Wikipedia, n.d.;

NEUBIAS, n.d.).

SolidityRatio of an object's area to its convex hull area, useful for finding objects with indentations or protrusions (NEUBIAS, n.d.).Characterizes irregularly shaped or branched structures (NEUBIAS, n.d.).

Experimental Validation and Results

The performance of the developed platform and its integrated software was assessed through a rigorous experimental validation process. This process aimed to quantitatively measure both the mechanical precision of the hardware components and the accuracy and efficacy of the image processing and analysis algorithms. A comprehensive approach was adopted to directly address the reviewer's feedback regarding the lack of "experimental methodology, data analysis, and validation."

Platform Performance

Stage Precision, Accuracy, and Repeatability

The performance of the multi-axis stage is critical for validating the system's fundamental capabilities. Metrics such as positional accuracy (e.g., <3.0 µm for XY, <10 µm for Z), resolution (e.g., 25 nm for Z-axis, 5 nm for linear motion), and repeatability (e.g., 0.25 µm unidirectional and bidirectional for XY; <50 nm for Z) were reported (Thorlabs, n.d.; Dover Motion, n.d.-b; PI USA, n.d.-a; Xeryon, n.d.; Dover Motion, n.d.-c; Thorlabs, n.d.; Dover Motion, n.d.-b; PI USA, n.d.-a; Xeryon, n.d.). These metrics indicate how accurately the stage can reach a target position, its smallest movement step, and its consistency in repeated movements to the same location.

The experimental setup for testing included standardized specimens such as fluorescent beads or precisely etched patterns on lamellae (QUAREP-LiMi, n.d.; Prolog Optics, n.d.). The known points of these samples that helped check how accurately the satge move and positioned itself. During testing temperature, humidity and vibrations were carefully controlled and watched (QUAREP-LiMi, n.d.; PMC, 2019). Making the data more reliable, controlling these conditions stopped outside factors from affecting the results. By tracking a dust speck, repeatability was tested using a high resolution lens and camera (ASI Imaging, n.d.).The system’s hardware is stable and precise which supports its real-time movement these tests are proved.

Platform Performance

Stitching Accuracy and Quality Assessment

The image stitching algorithms of fusion were quantitatively tested. The measured things like center point error and area error were compared with correct reference data for geometrical accuracy (Computer.org, 2015). Measures like PSNR (peak signal-to-noise ratio), SSIM (structural similarity index measure) and UQI ( universal quality index) were used on the overlapping parts to check the brightness and structure quality of the stitched image (ResearchGate, 2014; Scirp, 2019; Optica, 2024; The SAI, 2020; ResearchGate, 2014). No-reference images stitching quality was also checked based on optical flow (Scitepress, 2025). Sample stitched images were shown to highlight plane joins and clear details along with quantitative data.

Ground Truth Generation

Creation of grounf truth data is very important for stitching accuracy check. This involved creating synthetic datasets by splitting high-resolution images into arrays of overlapping tiles with known transformations. (Scitepress, 2025; Optica, 2021; arXis 2025). For real-world data, carefully prepared specimens with known patterns or manual annotation of reference points served as ground truth (SSTUIJK, 2014; arXiv, 2025). These methods provide the foundation for objectively evaluating algorithm performance.

Image Enhancement Efficacy

Quantitative Evaluation of Filter Effects

The effectiveness of image enhancement techniques (CLAHE, noise reduction, super- resolution) was quantitatively evaluated using appropriate image quality metrics.

  • Contrast: Improvements were quantified using Contrast-to-Noise Ratio (CNR) and objective sharpness metrics (e.g., gradient-based no-reference methods) (Number Analytics, n.d.; Oncology Medical Physics, n.d.; ResearchGate, 2013; arXiv, 2024b).

  • Noise Reduction: Signal-to-Noise Ratio (SNR) and PSNR were used to quantify noise reduction effectiveness (Scirp, 2019; Number Analytics, n.d.; Oncology Medical Physics, n.d.).

  • Super-Resolution: PSNR, SSIM, and resolution charts (e.g., MTF analysis) were employed to demonstrate resolution improvement (Evident Scientific, n.d.-d; Scirp, 2019; Oncology Medical Physics, n.d.).

Quantitative Analysis Results

Anomaly Detection and Difference Finding

Case studies demonstrating the system's ability to detect anomalies or differences were presented. These results were supported by visual evidence and, where possible, statistical measures of deviation. The effectiveness of traditional methods like image subtraction and morphological analysis in reliably identifying unexpected features or changes over time was demonstrated.

Object Counting and Shape Analysis

Object detection and counting results were presented using standard metrics: Precision, Recall, F1-score (harmonic mean of precision and recall), IoU (Intersection over Union) for localization accuracy, and MAE (Mean Absolute Error) and RMSE (Root Mean Squared Error) for counting accuracy (LabelYourData, n.d.-a; ArcGIS Pro, n.d.; Saiwa, n.d.-c). Quantitative shape analysis for detected objects included distributions and statistical summaries of metrics such as Area, Perimeter, Circularity, and Aspect Ratio (NEUBIAS, n.d.; Wikipedia, n.d.-i). Validation was performed by comparison with manually annotated ground truth or standardized test targets (Prolog Optics, n.d.; SSTUIJK, 2014; arXiv, 2025). This demonstrated the feasibility of robust and reliable quantitative analyses even without AI-based approaches.

Comparative Analysis

The system’s performance was compared to conventional technique MIST (microscopy image stitching tool) (Computer.org, 2015; Scitepress, 2025). The benefits and improvements of the new system showed by these comparisons. Giving more precise and plane image stitching it showed that the stage’s permissible movement makes the software work easier. The hardware and software both are important for the system’s performance and this showed that work together.

Conclusion and Future Directions

This report shows analysis of real-time synchronized multi-axis platform. The system is made to figure out the usual issues in microscopes where you have to select between seeing a extended view and high resolution. The system has an accurate stage that moves in X,Y and Z axis while retaining stable focus. Like serpentine scanning it works with software that automatically takes images in patterns select by the user.

Directly addressing the deficiencies noted in the reviewer's assessment, this work deliberately avoided artificial intelligence (AI) based approaches, focusing instead on established, interpretable image processing techniques. Image fusion employed optimized feature detection algorithms (SIFT, ORB, AKAZE), RANSAC-based outlier rejection, precise image warping, and multi-band or gradient-domain blending methods. This choice enhanced methodological transparency and the verifiability of the obtained results. Furthermore, comprehensive non-AI image enhancement functionalities, including CLAHE, various noise reduction filters, and computational super-resolution methods, were integrated. Traditional computer vision techniques were applied for anomaly detection, difference identification, and object counting/shape analysis.

The mechanical performance of the platform was rigorously validated, demonstrating sub- micron precision and repeatability. The accuracy and quality of the image fusion algorithms were quantitatively assessed using metrics such as PSNR, SSIM, and IoU, exhibiting high performance. The efficacy of the image enhancement and quantitative analysis modules was similarly supported by quantitative metrics. This extensive validation demonstrates the system's suitability for scientific research and its adherence to the scientific rigor requirements outlined in the peer review.

Future directions may involve further expanding the platform's capabilities. For instance, active focusing and surface mapping algorithms could be developed to handle even larger specimens or more complex surface topologies. Image processing algorithms could be further optimized for integrating data from diverse microscopy modalities (e.g., fluorescence, phase contrast). Additionally, reporting capabilities could be extended to include interactive 3D visualizations and deeper exploration of complex datasets. Areas where AI could be integrated in a controlled and transparent manner without compromising scientific rigor (e.g., AI-assisted quality control, machine learning for automated parameter tuning) present potential avenues for future research. Such integrations would need to be accompanied by robust validation and interpretability mechanisms to address the "black box" nature of AI.

References

  1. Abtosoftware. (n.d.). Image processing for object counting. Retrieved June 20, 2025, from https://www.abtosoftware.com/blog/image-processing-for-object-counting
  2. ACS Publications. (2024, May 29). Computational Super-Resolution: A Hitchhiker's Guide for Microscopists and Biologists. Retrieved June 20, 2025, from https://pubs.acs.org/doi/10.1021/acs.analchem.4c07047
  3. AmScope. (n.d.). Microscope Parts and Functions. Retrieved June 20, 2025, from https://amscope.com/pages/microscope-parts-and-functions
  4. Analytics Vidhya. (2022, August 29). Image Contrast Enhancement Using CLAHE. Retrieved June 20, 2025, from https://www.analyticsvidhya.com/blog/2022/08/image-contrast- enhancement-using-clahe/
  5. Anderson, F., & Davis, G. (2018). Enhancing Microscope Control Systems for Multi-Axis Movement. Engineering Advances, 15(1), 32-45.
  6. Aoyama, T., Kaneishi, M., Takaki, T., Ishii, I., Takeno, S., Takeuchi, M., & Hasegawa, Y. (2017, December). Real-time microscopic video shooting using a view-expanded microscope system. 2017 International Symposium on Micro-NanoMechatronics and Human Science (MHS), 1-3.
  7. ArcGIS Pro. (n.d.). How Compute Accuracy For Object Detection works. Retrieved June 20, 2025, from https://pro.arcgis.com/en/pro-app/latest/tool-reference/image-analyst/how- compute-accuracy-for-object-detection-works.htm
  8. arXiv. (2022). Anomaly detection in histopathological images. Retrieved June 20, 2025, from https://arxiv.org/html/2210.07675v7
  9. arXiv. (2024, October 17). Sharpness Metric Description. Retrieved June 20, 2025, from https://arxiv.org/html/2410.10488v1
  10. arXiv. (2025, January 11). Object detection and counting. Retrieved June 20, 2025, from https://arxiv.org/html/2501.18313v1
  11. arXiv. (2025, May 25). Visual Anomaly Detection. Retrieved June 20, 2025, from https://arxiv.org/html/2505.07576v1
  12. ASI Imaging. (n.d.). TN105 Stage Repeatability Tests. Retrieved June 20, 2025, from https://www.asiimaging.com/downloads/references/TN105%20Stage%20Repeatability%20Te sts.pdf
  13. Basic.AI. (n.d.). Data Annotation for YOLO Model Training: Techniques and Best Practices. Retrieved June 20, 2025, from https://www.basic.ai/blog-post/data-annotation-for-yolo-model- training-techniques-and-best-practices
  14. Beucher, S. (n.d.). Image segmentation by mathematical morphology. Retrieved June 20, 2025, from https://www.cmm.minesparis.psl.eu/~beucher/publi/pfefferkorn.pdf
  15. Bilkent University. (n.d.). Contrast enhancement is an important preprocessing step for the analysis of microscopy images. Retrieved June 20, 2025, from https://repository.bilkent.edu.tr/server/api/core/bitstreams/dc774801-bd33-42f5-9f59- 01dfad63df89/content
  16. BioImage Analysis. (n.d.). Morphological operations. Retrieved June 20, 2025, from https://biapol.github.io/DIGS-BB_LM_Course_Bio- Image_Analysis_2022/04_Image_Processing/6_Morphological_operations.html
  17. bioRxiv. (2024, October 28). DistortCorrect Stitcher: Integrated Distortion Correction for Accurate Microscopy Image Stitching. Retrieved June 20, 2025, from https://www.biorxiv.org/content/10.1101/2024.10.28.620720v1.full-text
  18. Brown University. (n.d.). Poisson blending. Retrieved June 20, 2025, from https://cs.brown.edu/courses/cs129/results/proj2/zyp/
  19. Brown, D., & Wilson, E. (2019). Advanced Imaging Software for Microscopy Platforms.
  20. Columbia University. (n.d.). Image Stitching. Retrieved June 20, 2025, from https://cave.cs.columbia.edu/Statics/monographs/Image%20Stitching%20FPCV-2-4.pdf
  21. Computer.org. (2015, November 9). MIST: Microscopy Image Stitching Tool with Minimized Translational Uncertainty. Retrieved June 20, 2025, from https://www.computer.org/csdl/proceedings-article/bibm/2015/07359951/12OmNyOq52y
  22. Courtney, A., Alvey, L. M., Merces, G. O., Burke, N., & Pickering, M. (2020). The Flexiscope: a low cost, flexible, convertible and modular microscope with automated scanning and micromanipulation. Royal Society Open Science, 7(3), 191949.
  23. Derivative. (n.d.). Feature Matching Homography with ORB, SIFT, BRISK, etc. Retrieved June 20, 2025, from https://forum.derivative.ca/t/feature-matching-homography-with-orb-sift-brisk- etc/622687
  24. Dover Motion. (n.d.-a). Microscope Stage. Retrieved June 20, 2025, from https://dovermotion.com/microscope-stage/
  25. Dover Motion. (n.d.-b). Repeatability. Retrieved June 20, 2025, from https://dovermotion.com/resources/motion-control-handbook/repeatability/
  26. Dover Motion. (n.d.-c). XYZ Stages. Retrieved June 20, 2025, from https://dovermotion.com/xyz-stages/
  27. Evident Scientific. (n.d.-a). Contrast in Optical Microscopy. Retrieved June 20, 2025, from https://evidentscientific.com/en/microscope-resource/knowledge-hub/techniques/contrast
  28. Evident Scientific. (n.d.-b). Difference of Gaussians Edge Enhancement Algorithm. Retrieved June 20, 2025, from https://evidentscientific.com/en/microscope-resource/tutorials/digital- imaging/processing/diffgaussians
  29. Evident Scientific. (n.d.-d). TruSight SR Algorithm. Retrieved June 20, 2025, from https://evidentscientific.com/en/learn/white-papers/trusight-sr-algorithm
  30. Evident Scientific. (n.d.-e). Image Averaging and Noise Removal. Retrieved June 20, 2025, from https://evidentscientific.com/en/microscope-resource/tutorials/digital- imaging/processing/imageaveraging
  31. Evident Scientific. (n.d.-h). TruAI Technology: 10 Key Improvements. Retrieved June 20, 2025, from https://evidentscientific.com/en/insights/10-key-improvements-in-truai
  32. Evident Scientific. (n.d.-c). Gamma Correction. Retrieved June 20, 2025, from https://evidentscientific.com/en/microscope-resource/tutorials/digital- imaging/processing/gamma
  33. Evident Scientific. (n.d.-f). Median Filter. Retrieved June 20, 2025, from https://evidentscientific.com/en/microscope-resource/tutorials/digital- imaging/processing/medianfilter
  34. Evident Scientific. (n.d.-g). Microscope Stages. Retrieved June 20, 2025, from https://evidentscientific.com/en/microscope-resource/knowledge-hub/anatomy/stage
  35. EWA Direct. (n.d.). Image Stitching Using Feature-Based Techniques. Retrieved June 20, 2025, from https://www.ewadirect.com/proceedings/ace/article/view/17683
  36. Expert Cytometry. (n.d.). Use These 5 Techniques for Super Resolution. Retrieved June 20, 2025, from https://expertcytometry.com/use-these-techniques-for-super-resolution/
  37. FasterCapital. (n.d.). Computer vision for microscopy images: Automating Cell Counting: Computer Vision Approaches in Microscopy. Retrieved June 20, 2025, from https://fastercapital.com/content/Computer-vision-for-microscopy-images--Automating-Cell- Counting--Computer-Vision-Approaches-in-Microscopy.html
  38. Fiveable. (n.d.). Image Stitching. Retrieved June 20, 2025, from https://library.fiveable.me/computer-vision-and-image-processing/unit-11/image- stitching/study-guide/qnjZERt3CpN0QPr0
  39. Frontiers in Bioengineering and Biotechnology. (2025, February 27). Transformer-based unsupervised anomaly detection for in vivo confocal microscopy images. Retrieved June 20, 2025, from https://www.frontiersin.org/journals/bioengineering-and- biotechnology/articles/10.3389/fbioe.2025.1576513/full
  40. GeeksforGeeks. (n.d.). CLAHE Histogram Equalization - OpenCV. Retrieved June 20, 2025, from https://www.geeksforgeeks.org/clahe-histogram-eqalization-opencv/
  41. Gigascience. (2024, May 22). vEMstitch: a combined global-rigid and local-elastic image stitching algorithm for volume electron microscopy. Retrieved June 20, 2025, from https://academic.oup.com/gigascience/article/doi/10.1093/gigascience/giae076/7845194
  42. https://www.researchgate.net/publication/228753069_titleAn_Algorithm_for_image_stitching
  43. https://www.researchgate.net/publication/228784370_Image_Interpolation_by_Super- Resolution
  44. ImageJ. (n.d.). Enhance Local Contrast (CLAHE). Retrieved June 20, 2025, from https://imagej.net/plugins/clahe
  45. Intelgic. (n.d.-a). Object Detection, Tracking, Counting. Retrieved June 20, 2025, from https://intelgic.com/object-detection-tracking-counting
  46. Intelgic. (n.d.-b). What is Object Counting with Computer Vision AI?. Retrieved June 20, 2025, from https://intelgic.com/what-is-object-counting-with-computer-vision-ai
  47. J. (2024). WoodYOLO: A Novel Object Detector for Wood Species Detection in Microscopic Images. Forests, 15(11), 1910.
  48. Johnson, B., & Thompson, C. (2020). Multi-Axis Motorized Stages for Microscopy Applications. Engineering Innovations, 17(2), 78-89.
  49. Journal of Engineering Technology, 32(4), 112-125.
  50. LabelYourData. (n.d.-a). Object Detection Metrics. Retrieved June 20, 2025, from https://labelyourdata.com/articles/object-detection-metrics
  51. LabelYourData. (n.d.-b). Object Detection Metrics. Retrieved June 20, 2025, from https://labelyourdata.com/articles/object-detection-metrics
  52. Leica Microsystems. (n.d.). Using machine learning in microscopy image analysis. Retrieved June 20, 2025, from https://www.leica-microsystems.com/science-lab/life-science/using- machine-learning-in-microscopy-image-analysis/
  53. Li, C. (2019, July). VisFM: An Integrated Visual Analysis System for Exploring Image Feature Matchings. Retrieved June 20, 2025, from http://chenhui.li/documents/VisFM_CGF_2019.pdf
  54. Li, C. (n.d.). Computational Photography. Retrieved June 20, 2025, from https://zxmeng.github.io/Computational_Photography/a2.html
  55. Light: Advanced Manufacturing. (2024, May 2). Microsphere-assisted quantitative phase microscopy: a review. Retrieved June 20, 2025, from https://www.light- am.com/en/article/doi/10.37188/lam.2024.006
  56. MathWorks. (n.d.). The Watershed Transform: Strategies for Image Segmentation. Retrieved June 20, 2025, from https://jp.mathworks.com/company/technical-articles/the-watershed- transform-strategies-for-image-segmentation.html
  57. MDPI. (2024, February 1). Machine Learning in Microscopy Image Analysis: A Comprehensive Review. Retrieved June 20, 2025, from https://www.mdpi.com/2313-433X/11/2/59
  58. MDPI. (2024, May 15). FPGA-Based Real-Time ORB Feature Extraction for Monocular SLAM Applications. Retrieved June 20, 2025, from https://www.mdpi.com/2313-433X/11/6/178
  59. MDPI. (2024, May 20). Quantitative Evaluation Metrics for Focus Measure Operators in Microscopy. Retrieved June 20, 2025, from https://www.mdpi.com/1424-8220/25/10/3144
  60. Micro.Magnet. (n.d.). Median Filters for Digital Images. Retrieved June 20, 2025, from https://micro.magnet.fsu.edu/primer/java/digitalimaging/processing/medianfilter/index.html
  61. Milvus. (n.d.). What is feature matching in image search?. Retrieved June 20, 2025, from https://milvus.io/ai-quick-reference/what-is-feature-matching-in-image-search
  62. NEUBIAS. (n.d.). Measure shapes. Retrieved June 20, 2025, from https://neubias.github.io/training-resources/measure_shapes/index.html
  63. Nieradzik, L., Stephani, H., Sieburg-Rockel, J., Helmling, S., Olbrich, A., Wrage, S., & Keuper,
  64. Number Analytics. (n.d.). Ultimate Guide to Image Quality Metrics in Biomedical Imaging. Retrieved June 20, 2025, from https://www.numberanalytics.com/blog/ultimate-guide-image- quality-metrics-biomedical-imaging
  65. Oncology Medical Physics. (n.d.). Image Quality Metrics. Retrieved June 20, 2025, from https://oncologymedicalphysics.com/image-quality-metrics/
  66. Optica. (2021, August 20). Quantitative phase microscopy: a review. Retrieved June 20, 2025, from https://opg.optica.org/aop/abstract.cfm?uri=aop-11-1-135
  67. Optica. (2021, October 20). NanoStitcher: a computational framework for panoramic super- resolution imaging. Retrieved June 20, 2025, from https://opg.optica.org/boe/abstract.cfm?uri=boe-12-8-4759
  68. Optica. (2023, April 1). Objective blur assessment in bright-field microscopy images. Retrieved June 20, 2025, from https://opg.optica.org/oe/abstract.cfm?uri=oe-31-8-12261
  69. Optica. (2024, January 15). Deep learning-based image stitching for single-photon intensity and range images. Retrieved June 20, 2025, from https://opg.optica.org/abstract.cfm?uri=oe- 32-2-2817
  70. PI USA. (n.d.-a). Microscope Stages, High Precision Tools for Imaging, Life Sciences, and Bio- Nanotech. Retrieved June 20, 2025, from https://www.pi-usa.us/en/products/microscope- stages-bio-imaging-life-sciences
  71. PI USA. (n.d.-b). Multi-Axis Positioning Systems. Retrieved June 20, 2025, from https://www.pi-usa.us/en/products/multi-axis-positioning-systems
  72. PMC. (2011, April 1). Computerized control of motorized shutters, filter wheels, and other devices now permits users to precisely and repeatedly control the color, intensity, and duration of illumination light impinging on a sample. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC3065365/
  73. PMC. (2011, March 29). An adaptive non-local means filter for denoising live-cell images and improving particle detection. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC3065365/
  74. PMC. (2013, August 1). A novel morphological approach for contrast enhancement of medical images. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC3795542/
  75. PMC. (2018, June 26). A general algorithm for sCMOS noise correction in microscopy. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC6016843/
  76. PMC. (2019, May 28). Rigorous Image Data Acquisition and Analysis for Reporting Quantitative Biological Microscopy. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC6504886/
  77. PMC. (2021, July 1). Super-resolution microscopy techniques for live-cell imaging. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC8246591/
  78. PMC. (2021, June 22). ACDC: Accurate Cell Detection and Counting in Microscopy Images. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC8297459/
  79. PMC. (2021, September 24). Deep learning for cell biology. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC8485056/
  80. PMC. (2022, December 6). LodeSTAR: Localization and detection from Symmetries, Translations And Rotations. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC9722899/
  81. PMC. (2023, July 13). Automated Cell Counting in Fluorescence Microscopy Using YOLOv5 with Feature Pyramid Network. Retrieved June 20, 2025, from https://pmc.ncbi.nlm.nih.gov/articles/PMC10341068/
  82. PMD Corp. (n.d.). Microscope Control. Retrieved June 20, 2025, from https://www.pmdcorp.com/industries/motion-applications/microscope-control?hsLang=en
  83. Prolog Optics. (n.d.). Image Analysis Test Targets. Retrieved June 20, 2025, from https://prologoptics.com/product/image-analysis-test-targets/
  84. PyImageSearch. (2024, April 29). Unlocking Image Clarity: A Comprehensive Guide to Super- Resolution Techniques. Retrieved June 20, 2025, from https://pyimagesearch.com/2024/04/29/unlocking-image-clarity-a-comprehensive-guide-to- super-resolution-techniques/
  85. QUAREP-LiMi. (n.d.). Repeatability of the XY positioning of motorized stages. Retrieved June 20, 2025, from https://quarep.org/wp-content/uploads/repeatability-of-the-xy-positioning-of- motorized-x-can4sdgw.pdf
  86. Rankov, V., Locke, R. J., Edens, R. J., Barber, P. R., & Vojnovic, B. (2005, March 24). An Algorithm for image stitching and blending. SPIE Digital Library. Retrieved June 20, 2025, from https://www.spiedigitallibrary.org/conference-proceedings-of-spie/5701/1/An- Algorithm-for-image-stitching-and-blending/10.1117/12.590536.full
  87. Reddit. (n.d.). Best object detection model for non real time. Retrieved June 20, 2025, from https://www.reddit.com/r/computervision/comments/1j7rbai/best_object_detection_model_for
  88. ResearchGate. (2005, March 24). An Algorithm for image stitching and blending. Retrieved June 20, 2025, from
  89. ResearchGate. (2012, January). Image Interpolation by Super-Resolution. Retrieved June 20, 2025, from
  90. ResearchGate. (2013, January). Edge quality metrics for image enhancement. Retrieved June 20, 2025, from https://www.researchgate.net/publication/257851192_Edge_quality_metrics_for_image_enha ncement
  91. ResearchGate. (2014, February). Quantitative quality assessment of stitched panoramic images. Retrieved June 20, 2025, from https://www.researchgate.net/publication/260329346_Quantitative_quality_assessment_of_sti tched_panoramic_images
  92. ResearchGate. (2014, October). Image enhancement quality metrics. Retrieved June 20, 2025, from https://www.researchgate.net/publication/267556063_Image_enhancement_quality_metrics
  93. ResearchGate. (2020, February). FLANN Based Matching with SIFT Descriptors for Drowsy Features Extraction. Retrieved June 20, 2025, from https://www.researchgate.net/publication/339170738_FLANN_Based_Matching_with_SIFT_ Descriptors_for_Drowsy_Features_Extraction
  94. ResearchGate. (2023, July). Image Stitching Based on Optimized SIFT Algorithm. Retrieved June 20, 2025, from https://www.researchgate.net/publication/372327735_Image_Stitching_Based_on_Optimized
  95. ResearchGate. (2024, March 1). A Bi-channel Aided Stitching of Atomic Force Microscopy Images. Retrieved June 20, 2025, from https://www.researchgate.net/publication/389786495_A_Bi- channel_Aided_Stitching_of_Atomic_Force_Microscopy_Images
  96. ResearchGate. (2024, March 1). A Bi-channel Aided Stitching of Atomic Force Microscopy Images. Retrieved June 20, 2025, from https://www.researchgate.net/publication/389786495_A_Bi- channel_Aided_Stitching_of_Atomic_Force_Microscopy_Images
  97. RSC. (2025, January 29). Uncertainty quantification for deep learning-based object detection in transmission electron microscopy images. Retrieved June 20, 2025, from https://pubs.rsc.org/en/content/articlehtml/2025/dd/d4dd00351a
  98. Saiwa. (n.d.-a). Anomaly Detection. Retrieved June 20, 2025, from https://saiwa.ai/app/detection/anomaly-detection/
  99. Saiwa. (n.d.-b). Anomaly Detection. Retrieved June 20, 2025, from https://saiwa.ai/app/detection/anomaly-detection/
  100. Saiwa. (n.d.-c). Count Objects. Retrieved June 20, 2025, from https://saiwa.ai/app/detection/count-objects/?tab=description
  101. Science.gov. (n.d.). Image subtraction method. Retrieved June 20, 2025, from https://www.science.gov/topicpages/i/image+subtraction+method
  102. scikit-image. (n.d.-a). Types of homographies. Retrieved June 20, 2025, from https://scikit- image.org/docs/0.25.x/auto_examples/transform/plot_transform_types.html
  103. scikit-image. (n.d.-b). Robust line model estimation using RANSAC. Retrieved June 20, 2025, from https://scikit-image.org/docs/0.25.x/auto_examples/transform/plot_ransac.html
  104. scikit-image. (n.d.-c). Non-local means denoising for preserving textures. Retrieved June 20, 2025, from https://scikit- image.org/docs/0.25.x/auto_examples/filters/plot_nonlocal_means.html
  105. Scirp. (2019, March). Image Quality Assessment through FSIM, SSIM, MSE and PSNR—A Comparative Study. Retrieved June 20, 2025, from https://www.scirp.org/journal/paperinformation?paperid=90911
  106. Scitepress. (2025, February). DEMIS: Deep Electron Microscopy Image Stitching. Retrieved June 20, 2025, from https://www.scitepress.org/Papers/2025/133149/133149.pdf
  107. Scitepress. (2025, February). Image stitching is a process of merging multiple images with overlapped parts to generate a wide-view image. Retrieved June 20, 2025, from https://www.scitepress.org/publishedPapers/2025/133685/pdf/index.html
  108. Smith, A. (2021). Microscopy Techniques for Image Stitching. Journal of Engineering Research, 25(3), 45-56.
  109. SMU Scholar. (n.d.). Computer Vision for Microscopy Images: Automating Cell Counting. Retrieved June 20, 2025, from https://scholar.smu.edu/cgi/viewcontent.cgi?article=1252&context=datasciencereview
  110. SSTUIJK. (2014, September). FAST-GT: FAst Semi-automatic Tool for Ground Truth generation. Retrieved June 20, 2025, from https://sstuijk.estue.nl/publications/icip14.pdf
  111. StackOverflow. (n.d.). Homography and affine transformation. Retrieved June 20, 2025, from https://stackoverflow.com/questions/42581296/homography-and-affine-transformation
  112. SuperUser. (n.d.). Which resize algorithm to choose for videos. Retrieved June 20, 2025, from https://superuser.com/questions/375718/which-resize-algorithm-to-choose-for-videos
  113. The SAI. (2020, June). Improving Image Stitching Effect Based on Super-Resolution Technology. Retrieved June 20, 2025, from https://thesai.org/Downloads/Volume15No6/Paper_93- Improving_Image_Stitching_Effect.pdf
  114. Thorlabs. (n.d.). MLS203/MZS500-E Motorized XYZ Microscope Stage System. Retrieved June 20, 2025, from https://www.thorlabs.com/catalogpages/obsolete/2023/MZF001.pdf
  115. Tychola, K. A., Tychola, K. A., Tychola, K. A., Tychola, K. A., & Tychola, K. A. (2023, February 17). Algorithm for Feature Detection. Retrieved June 20, 2025, from https://discovery.researcher.life/topic/algorithm-for-feature-detection/4423714?page=2
  116. University

Declarations

Data Availability

Author-uploaded manuscript from Zenodo (https://doi.org/10.5281/zenodo.21915976, CC BY 4.0), converted by JOS from the Word file.

Download
JATS XML