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
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.
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.
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.
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.).
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.).
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).
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.
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.
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.
The key step is image fusion that integrate overlapping microscope images into one clear, high- detail, wide review image.
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.
| Algorithm | Key Characteristics | Typical Use Cases | Performance Metrics (from Experimental Evaluation) | Optimized Parameters |
|---|---|---|---|---|
| Robust to scale, rotation, and illumination | General stitching, applications | Feature Repeatability, Match | (Application- specific, e.g., contrast |
| Algorithm | Key Characteristics | Typical Use Cases | Performance 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) | |
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). | |
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) |
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.).
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.
controls the degree of blurring (Evident Scientific, n.d.-b; Micro.Magnet, n.d.; Evident Scientific, n.d.-e).
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).
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.
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.).
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.).
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.
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).
| Category | Metric Name | Definition/What it Measures | Relevance to Microscopy |
|---|---|---|---|
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). |
| Category | Metric Name | Definition/What it Measures | Relevance to Microscopy |
|---|---|---|---|
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). | ||
| 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.). |
||
| Clarity 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). |
||
| Ability 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.). |
| Category | Metric Name | Definition/What it Measures | Relevance to Microscopy |
|---|---|---|---|
| Ratio 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). | ||
| Ratio 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). | ||
| Harmonic 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). | ||
| 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). |
||
Average absolute difference between predicted and actual |
Direct measure of object counting |
| Category | Metric Name | Definition/What it Measures | Relevance to Microscopy |
|---|---|---|---|
| object counts (Saiwa, n.d.-c). | accuracy (Saiwa, n.d.- c). | ||
| 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). | ||
| Measures 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.). |
||
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.). |
||
| Ratio 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.). |
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."
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.
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.
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.
The effectiveness of image enhancement techniques (CLAHE, noise reduction, super- resolution) was quantitatively evaluated using appropriate image quality metrics.
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 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.
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.
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.
Author-uploaded manuscript from Zenodo (https://doi.org/10.5281/zenodo.21915976, CC BY 4.0), converted by JOS from the Word file.