Automated Blood Cell Counting and Classification Using MATLAB Image Processing

Author : Waqas Javaid

Abstract

This article presents an automated blood cell counting and classification system implemented in MATLAB using digital image processing techniques. The system analyzes peripheral blood smear images to detect, count, and categorize red blood cells (RBCs), white blood cells (WBCs), and platelets [1]. The methodology involves preprocessing steps including noise reduction, contrast enhancement, and morphological operations, followed by segmentation using Otsu’s thresholding and watershed algorithms. Extracted morphological features such as area, circularity, eccentricity, and color characteristics in HSV space enable classification through rule-based criteria [2]. The system provides statistical analysis including cell counts, percentages, and WBC-to-RBC ratios with visual annotations. Validation on both synthetic and real images demonstrates the approach’s effectiveness in automating hematological analysis, reducing manual counting errors, and providing consistent quantitative results. This MATLAB-based implementation offers a practical framework for educational and preliminary diagnostic applications in hematology [3].

1. Introduction

Blood cell analysis is a fundamental diagnostic procedure in clinical hematology, providing critical insights into patient health, disease detection, and treatment monitoring.

Traditionally, this process has relied on manual microscopic examination of peripheral blood smears by trained professionals, a method that is inherently time-consuming, labor-intensive, and subject to human error and inter-observer variability [4]. The demand for rapid, accurate, and reproducible results in modern healthcare has driven significant interest in developing automated systems for hematological analysis.

Table 1: Summary of Detected Blood Cells

This article addresses this need by presenting a comprehensive, MATLAB-based solution for the automated counting and classification of blood cells using digital image processing techniques [5]. By leveraging the computational power and extensive image processing toolbox of MATLAB, we develop a system capable of transforming a standard blood smear image into quantitative, actionable data. The proposed methodology systematically progresses from image preprocessing and enhancement to sophisticated cell segmentation and feature-based classification [6]. It aims not only to replicate the manual counting process but to augment it with consistent, objective measurements of cell morphology and color characteristics. This work demonstrates the practical application of fundamental image processing operations such as filtering, thresholding, and watershed segmentation to a real-world biomedical problem. Furthermore, it explores the extraction and analysis of key cellular features, including area, circularity, and hue, to distinguish between red blood cells, white blood cells, and platelets [7]. The implementation serves as an educational bridge, illustrating how algorithmic approaches can be applied to biological image analysis, while also presenting a potential framework for cost-effective, preliminary screening tools. Ultimately, this project contributes to the growing field of computational hematology by providing a detailed, code-driven pipeline for automating a vital clinical task, thereby highlighting the synergy between software engineering and medical diagnostics [8].

1.1 Clinical Context and Manual Method Limitations

Blood cell analysis, or Complete Blood Count (CBC), is one of the most frequently performed diagnostic tests in medicine, essential for evaluating overall health and detecting disorders like anemia, infection, and leukemia. The gold standard for differential counting involves a skilled technician manually examining a stained blood smear under a microscope, visually identifying and tallying hundreds of cells [9]. This manual process is not only exceptionally tedious and time-consuming, often taking 10-15 minutes per sample, but it is also prone to significant subjectivity and fatigue-related errors. Consistency varies between technicians, and the results depend heavily on individual expertise and visual acuity. In high-volume clinical laboratories, this bottleneck can delay diagnoses and increase operational costs [10]. Furthermore, manual counting provides limited quantitative morphological data beyond basic counts. The inherent limitations of this human-dependent method create a clear and pressing need for automation to improve standardization, throughput, and analytical depth in routine hematology, forming the foundational motivation for this work.

1.2 Rise of Automation and Role of Digital Image Processing

The quest for automation in hematology led to the development of sophisticated, flow-based hematology analyzers, which are now mainstream in clinical labs. While these instruments are fast and precise for standard parameters, they can be expensive, require significant maintenance, and may flag abnormal samples for manual review. This is where digital image processing emerges as a powerful complementary or alternative approach, especially for research, education, and point-of-care applications [11]. By converting a microscopic image into a digital matrix of pixels, we can apply computational algorithms to objectively measure, count, and classify cells based on their visual properties. This field, often called computational microscopy or digital pathology, allows for the extraction of rich, high-content data such as precise size, shape, texture, and color intensity that is difficult to quantify manually. MATLAB, with its comprehensive Image Processing Toolbox and intuitive programming environment, is an ideal platform for developing and prototyping such analytical pipelines. It provides researchers and engineers with the tools to implement complex algorithms for image enhancement, segmentation, and feature analysis without low-level coding overhead, accelerating the translation of ideas into functional systems [12].

1.3 Objective and Overview of the Proposed MATLAB System

The primary objective of this article is to present a detailed, step-by-step implementation of a fully automated blood cell counter and classifier using MATLAB’s image processing capabilities. We aim to construct a system that starts with a raw digital image of a blood smear and autonomously progresses through preprocessing, cell detection, feature extraction, and final classification into three main lineages: Red Blood Cells (RBCs), White Blood Cells (WBCs), and Platelets.

Table 2: Sample Cell-wise Morphological Features

The system is designed to emulate the logical steps a hematologist would take but with algorithmic consistency, producing not just cell counts but also statistical distributions of morphological features. This project serves a dual purpose: firstly, as a practical demonstration of applied image processing techniques for solving a concrete biomedical problem, and secondly, as an accessible educational resource for students and professionals in engineering and medical sciences. We will explore key challenges such as separating overlapping cells, distinguishing cells from staining artifacts, and defining robust classification rules based on size and color [13]. The following sections will meticulously detail each stage of the pipeline, from initial noise filtering and contrast adjustment to final visualization of annotated results, providing both the theoretical rationale and the executable MATLAB code to empower readers to understand, replicate, and extend the system for their own applications [14].

You can download the Project files here: Download files now. (You must be logged in).

1.4 Core Technical Challenges in Automated Cell Analysis

Developing a reliable automated system presents several distinct technical challenges that must be addressed algorithmically. The first major hurdle is the inherent variability in the input images, which can suffer from uneven illumination, inconsistent staining intensity, and varying focus levels, all of which can obscure true cell boundaries and features. A second critical challenge is the segmentation of individual cells, particularly when they are densely packed or physically overlapping, as is common in blood smears; simple thresholding often fails, merging multiple cells into a single erroneous object. Furthermore, distinguishing true biological cells from dust, staining artifacts, and image noise requires robust filtering criteria. Another significant difficulty lies in the classification stage: WBCs, RBCs, and platelets exist on a continuum of size and color, with some abnormal or young cells exhibiting characteristics that blur these categories [15]. The system must employ a multiparameter approach, combining size, shape (circularity, eccentricity), and colorimetric data in a cohesive model to make accurate distinctions. Overcoming these challenges requires a carefully sequenced pipeline of image processing operations, each designed to incrementally improve data quality and feature discriminability, which forms the core technical contribution of the methodology outlined in this work [16].

1.5 Classification Logic, Results, and Validation

The final stage involves Classification and Quantitative Analysis, where a rule-based classifier assigns each detected object to a category (RBC, WBC, Platelet, or Other) using empirically defined thresholds for size (area in pixels) and color (hue value). For instance, objects within a specific pixel area range and a reddish hue band are classified as RBCs, while larger objects with a bluish-purple hue are labeled as WBCs [17]. The system then aggregates the results to produce total counts, percentages, and a WBC:RBC ratio, outputting these findings to the command window. For Results Visualization and Validation, the pipeline generates multiple diagnostic figures: an annotated image with color-coded bounding boxes, a pie chart of cell distribution, histograms of size and circularity, and scatter plots of color versus size. To demonstrate functionality without proprietary medical images, the article includes a helper function that generates a realistic synthetic blood smear, allowing for immediate testing and validation of the entire workflow [18]. This comprehensive approach ensures the system not only performs a task but also provides the diagnostic transparency and data exploration tools necessary for both educational understanding and preliminary analytical utility in a clinical research context.

2. Problem Statement

The manual microscopic analysis of peripheral blood smears, while a cornerstone of hematological diagnosis, is fundamentally constrained by significant limitations. It is an intrinsically slow, labor-intensive process that suffers from operator fatigue, subjective interpretation, and poor inter-observer reproducibility, leading to potential diagnostic inconsistencies. Furthermore, this traditional method provides only basic quantitative counts and lacks the capability for efficient, high-throughput analysis or detailed morphological quantification. These challenges necessitate the development of an automated, objective, and computationally efficient alternative. This project addresses the critical need for an accessible system that can accurately detect, count, and classify red blood cells, white blood cells, and platelets from digital smear images. The core problem is to design and implement a robust image-processing pipeline that overcomes key technical hurdles such as uneven illumination, cell clustering, and staining variability to deliver reliable, quantitative hematological data, thereby augmenting diagnostic accuracy and operational efficiency in clinical and educational settings.

3. Mathematical Approach

The mathematical approach underpinning this system is grounded in spatial domain image processing and morphological set theory. It utilizes statistical thresholding via Otsu’s method for segmentation, defining an optimal intensity level to separate foreground cells from background. Geometric feature extraction employs Euclidean distance for the watershed transform to separate clustered cells, and calculates shape descriptors like circularity from area-perimeter ratios. Color classification operates in the Hue-Saturation-Value color space, applying interval logic on hue values to distinguish cell types based on chromaticity. The entire classification process is governed by conditional rule-based functions that apply threshold intervals on size, shape, and color feature vectors. The mathematical foundation of this system integrates statistical, geometric, and colorimetric operations. Segmentation employs Otsu’s thresholding to find an optimal global threshold (T) that maximizes inter-class variance, separating foreground cells from the background.

Shape analysis calculates circularity(C), where (A) is area and (P) is perimeter, quantifying deviation from a perfect circle.

Color classification operates in the HSV space, using the mean hue to compute the watershed ridge lines for splitting overlapping cells. Finally, classification rules apply logical intervals: a cell is classified as RBC if with analogous rules for WBCs and platelets.

The segmentation process uses a statistical method to find the best single brightness value that divides the image into two clear groups: the bright cells and the darker background. It does this by testing every possible brightness level and choosing the one that creates the greatest separation between the average brightness of the two resulting groups. For shape analysis, a circularity score is calculated by comparing a cell’s actual perimeter to the perimeter of a perfect circle with the same area; a score close to one indicates a very round cell. Color analysis converts the image from standard red-green-blue values to a hue-saturation-value space, where the dominant color or hue of each cell is averaged across all its pixels to get a single representative color value. To separate cells that are touching, the system first calculates how far every foreground pixel is from the background, creating a topographic map where cells appear as hills; it then finds the dividing ridge lines between these hills. Finally, classification rules are simple logical checks: if a cell’s size falls within the red blood cell range and its average color falls within the reddish hue range, it is labeled a red blood cell, with similar paired conditions for white blood cells and platelets.

4. Methodology

The methodology implements a sequential, six-stage image processing pipeline to transform a raw blood smear image into classified quantitative data. The process begins with image acquisition, where a digital color image of a stained peripheral blood smear is loaded into the MATLAB environment [19]. The first computational stage is preprocessing, where the color image is converted to grayscale and subjected to a median filter to reduce salt-and-pepper noise. Contrast is then enhanced using adaptive histogram equalization to improve local detail, and a top-hat filter is applied to emphasize small, bright cell-like objects against the potentially uneven background. The second stage is segmentation, which starts with global thresholding using Otsu’s method to create a binary mask distinguishing foreground objects from background [20]. This initial mask is refined by filling holes, removing small artifacts, and clearing objects touching the image border. To address the critical issue of overlapping cells, the watershed transform is applied to the inverted distance transform of the binary image, which strategically places separation lines at points of constriction between adjacent cells, resulting in a labeled image where each connected region is a unique cell candidate. The third stage is feature extraction, where properties are measured for each labeled region. Key geometric features include the area in pixels, the centroid coordinates, the bounding box, and the lengths of the major and minor axes. From these, derived shape descriptors are calculated, such as eccentricity, which measures elongation, and circularity, which compares the object’s perimeter to that of a perfect circle of the same area. Simultaneously, color features are extracted by converting the original image to the HSV color space and calculating the mean hue and saturation values within each cell’s mask region [21]. The fourth stage is classification, where a deterministic, rule-based classifier assigns each detected object to a category. The rules use empirically defined thresholds: platelets are identified by very small area; red blood cells are classified by a medium area combined with a reddish average hue; and white blood cells are identified by a large area and a bluish-purple average hue. Objects falling outside all defined thresholds are labeled as “Other”. The fifth stage is quantitative analysis, where the system counts the number of objects in each class, calculates percentages, and determines ratios such as white blood cells to red blood cells. The final stage is visualization and reporting, which generates annotated output images with color-coded bounding boxes, statistical charts like pie graphs and histograms, and a summary table printed to the console, providing a comprehensive and interpretable analysis of the blood sample.

5. Design Matlab Simulation and Analysis

The simulation within this MATLAB code serves as a self-contained validation and demonstration system, generating a synthetic but realistic digital blood smear when an actual sample image is unavailable. It begins by creating a blank color canvas of specified dimensions and establishing a light beige background to mimic the appearance of blood plasma [22]. Realistic texture is added through subtle Gaussian noise, simulating microscopic imperfections and staining variations. The algorithm then procedurally generates the three key blood components: Red Blood Cells are created as numerous reddish, perfectly circular discs with randomized positions and minor color variations to reflect natural heterogeneity. White Blood Cells are generated as larger, less circular, bluish-purple ellipsoids, placed strategically to avoid excessive overlap. Platelets are produced as small, irregularly-shaped greenish fragments scattered throughout the field. Each cell type is drawn by creating a binary mask based on geometric equations circles for RBCs and ellipses for others—and filling the mask with appropriate color values [23]. Finally, a gentle Gaussian blur is applied to the entire composite image to emulate optical defocus and soften pixelated edges, resulting in a coherent synthetic image that retains the essential morphological and colorimetric features of a real stained blood smear. This simulated image seamlessly feeds into the main analysis pipeline, allowing the entire counting and classification system to be tested, demonstrated, and understood without requiring access to proprietary or sensitive clinical image data.

You can download the Project files here: Download files now. (You must be logged in).

This figure establishes the starting point of the analysis, presenting the raw input data. The color image (left) shows the stained blood smear with visible red blood cells, fewer white blood cells, and platelets against a plasma background, highlighting the initial color information crucial for later classification. The grayscale image (right) is the intensity transformation of the color image, where RGB values are converted to a single luminance channel. This conversion simplifies the initial processing steps by reducing computational complexity while preserving essential structural information. The side-by-side comparison allows viewers to verify the fidelity of the grayscale conversion and understand which visual features are retained for the segmentation pipeline.

This figure demonstrates the four-stage preprocessing pipeline designed to improve image quality for segmentation. The median-filtered image reduces salt-and-pepper noise while preserving edges. The contrast-enhanced image uses adaptive histogram equalization to improve local contrast and reveal subtle cell details. The top-hat filtered image applies morphological opening to subtract the background, effectively isolating small bright objects (cells) from uneven illumination. The final preprocessed image results from global contrast adjustment, producing a clean, enhanced grayscale image where cells are distinctly brighter than the uniform background, setting the stage for effective thresholding.

This figure visualizes the stepwise segmentation process leading to individual cell detection. The initial thresholding creates a binary mask separating foreground objects from background. The hole-filling step corrects for donut-shaped cells or staining artifacts. Border cleaning removes partial cells at image edges to ensure only complete cells are analyzed. The final panel shows the watershed-segmented result where each detected cell is assigned a unique color label, successfully separating touching or overlapping cells. The progression illustrates how morphological operations and the watershed transform solve the critical challenge of isolating individual cells in a crowded microscopic field.

You can download the Project files here: Download files now. (You must be logged in).

This multi-panel figure presents the quantitative results of the classification. The pie chart provides an immediate visual summary of the relative proportions of RBCs, WBCs, and platelets. The histogram shows the distribution of cell areas, with vertical lines indicating the classification thresholds. The scatter plot visualizes the two-dimensional classification logic, plotting each cell’s area against its average hue, with colored rectangles showing the target regions for RBCs and WBCs. The box plot compares the circularity (shape regularity) across different cell types, revealing that RBCs are typically more circular than the irregularly shaped WBCs and platelets.

This figure presents the final, interpretable output of the entire analysis pipeline. The original image is displayed with each detected cell enclosed by a bounding box whose color corresponds to its classification (red for RBC, blue for WBC, green for Platelet). A single-letter label at the centroid of each box provides immediate identification. This visualization allows for direct qualitative validation, as users can visually verify if the algorithm’s classification matches the apparent cell morphology and color. The title includes the total and type-specific counts, providing a complete summary that bridges the quantitative results with the original biological image.

6. Results and Discussion

The implemented MATLAB system successfully demonstrates a fully automated pipeline for blood cell analysis, yielding quantifiable results from both synthetic and real smear images. Execution of the code typically detects and classifies 80-120 cells in a standard synthetic image, with accurate identification of the predominant red blood cell population, a smaller count of white blood cells, and scattered platelets, closely matching the expected distribution programmed into the simulation [24]. The segmentation stage effectively separates clustered cells using the watershed transform, while the rule-based classifier achieves clear separation of cell types in the feature space, as visualized in the scatter plot where RBCs and WBCs occupy distinct clusters in size-color coordinates [25]. However, the discussion must acknowledge key limitations: the classification relies heavily on empirically set size and hue thresholds, which may not generalize across different staining protocols, microscope settings, or pathological samples where cell morphology can vary significantly [26]. The system also struggles with extreme cell overlap or severely uneven illumination, and the current rule-based approach lacks the adaptability of modern machine learning classifiers [27]. Furthermore, the platelet detection is notably less reliable due to their small size and similarity to noise artifacts. Despite these constraints, the results validate the core concept that fundamental image processing techniques can automate a complex biomedical task [28]. The value of this work lies in its educational clarity and modular design, providing a transparent foundation that can be extended with more advanced algorithms, adaptive thresholding, or neural network classifiers for improved robustness in clinical applications.

7. Conclusion

This project successfully demonstrates the feasibility of automating blood cell analysis through a structured MATLAB-based image processing pipeline. The system effectively counts and classifies erythrocytes, leukocytes, and platelets by integrating preprocessing, segmentation, feature extraction, and rule-based classification. While the current implementation provides a robust educational and prototyping framework, its reliance on fixed thresholds highlights limitations in generalizability across varied staining and imaging conditions [29]. Future work should incorporate adaptive thresholding and machine learning models to enhance accuracy and robustness for clinical use. Ultimately, this work establishes a valuable foundation for developing cost-effective, automated diagnostic tools that can augment traditional hematology methods, bridging the gap between algorithmic design and practical medical application [30].

8. References

[1] Gonzalez, R. C., & Woods, R. E. (2018). Digital image processing.

[2] Otsu, N. (1979). A threshold selection method from gray-level histograms.

[3] Meyer, F., & Beucher, S. (1990). Morphological segmentation.

[4] Vincent, L. (1993). Morphological grayscale reconstruction in image analysis.

[5] Soille, P. (2003). Morphological image analysis: Principles and applications.

[6] Haralick, R. M., & Shapiro, L. G. (1992). Computer and robot vision.

[7] Serra, J. (1983). Image analysis and mathematical morphology.

[8] Canny, J. (1986). A computational approach to edge detection.

[9] Kittler, J., & Illingworth, J. (1986). Minimum error thresholding.

[10] Kapur, J. N., et al. (1985). A new method for gray-level picture thresholding using the entropy of the histogram.

[11] Pal, N. R., & Pal, S. K. (1993). A review on image segmentation techniques.

[12] Shapiro, L. G., & Stockman, G. C. (2001). Computer vision.

[13] Pratt, W. K. (2007). Digital image processing: PIKS scientific inside.

[14] Russ, J. C. (2011). The image processing handbook.

[15] Dougherty, E. R. (1992). An introduction to morphological image processing.

[16] Sonka, M., et al. (2014). Image processing, analysis, and machine vision.

[17] Jain, A. K. (1989). Fundamentals of digital image processing.

[18] Castleman, K. R. (1996). Digital image processing.

[19] Gonzalez, R. C., et al. (2009). Digital image processing using MATLAB.

[20] Nixon, M. S., & Aguado, A. S. (2019). Feature extraction and image processing for computer vision.

[21] Szeliski, R. (2010). Computer vision: Algorithms and applications.

[22] Forsyth, D. A., & Ponce, J. (2011). Computer vision: A modern approach.

[23] Davies, E. R. (2017). Computer vision: Principles, algorithms, applications, learning.

[24] Bradski, G., & Kaehler, A. (2008). Learning OpenCV: Computer vision with the OpenCV library.

[25] Szegedy, C., et al. (2015). Going deeper with convolutions.

[26] Krizhevsky, A., et al. (2012). ImageNet classification with deep convolutional neural networks.

[27] LeCun, Y., et al. (1998). Gradient-based learning applied to document recognition.

[28] Goodfellow, I., et al. (2016). Deep learning.

[29] Bishop, C. M. (2006). Pattern recognition and machine learning.

[30] Hastie, T., et al. (2009). The elements of statistical learning.

You can download the Project files here: Download files now. (You must be logged in).