Portfolio

Face Tracking

Face tracking in this algorithm is performed using OpenCV’s Haar cascade classifier for initial face detection, followed by Dlib-based landmark detection to extract detailed facial features. The function begins by converting the input grayscale image into an 8-bit format, as OpenCV processes images in 8-bit depth. The grayscale version is then passed through OpenCV’s face detection model, which identifies bounding boxes around faces based on pre-trained features. These detected faces are further processed with Dlib’s shape predictor, which maps landmark points on facial regions such as the eyes, nose, and lips.

Once faces are detected, the algorithm updates the tracked face list, ensuring real-time tracking across frames. The system then draws bounding boxes around the detected faces and overlays landmark points, visualizing key facial features. If a face region is too large, it is resized to a fixed 150×150 resolution before passing it to Dlib, preventing memory issues. Finally, the processed frame is converted into a WriteableBitmap and displayed in the UI using Dispatcher, enabling real-time facial recognition and tracking. This method is effective for biometric authentication, facial expression analysis, emotion detection, and AR/VR applications, where detailed face tracking is crucial.

Quick GUI by Tkinter

Tkinter is Python’s standard library for creating Graphical User Interfaces (GUI). It provides a simple yet powerful way to develop desktop applications with interactive elements such as buttons, labels, text inputs, and more. Since Tkinter is included with Python, it doesn’t require additional installation for building small to medium-sized desktop applications, from data visualization tools to hardware control interfaces

Cardiovascular Diseases Prediction from ECG Image.

Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide, making early detection critical for effective intervention. Electrocardiograms (ECGs) serve as a widely accessible and noninvasive diagnostic tool for identifying cardiac abnormalities. This study implements a deep learning-based approach to classify ECG images into four categories: abnormal heartbeat, myocardial infarction, history of myocardial infarction, and normal heart conditions. Unlike traditional handcrafted feature extraction techniques, we leverage transfer learning by employing the MobileNet architecture, pretrained on ImageNet, to extract relevant features from ECG images.

The dataset used in this study is the ECG Images Dataset of Cardiac Patients, created under the auspices of Ch. Pervaiz Elahi Institute of Cardiology, Multan, Pakistan. This dataset aims to support the scientific community in conducting research on cardiovascular diseases. The dataset consists of ECG images from different patient groups, with a total size of 194 MB, and contains:

• Myocardial Infarction Patients: (240 × 12) = 2,880 images

• Patients with Abnormal Heartbeat: (233 × 12) = 2,796 images

• Patients with a History of Myocardial Infarction: (172 × 12) = 2,064 images

• Normal Person ECG Images: (284 × 12) = 3,408 images

By fine-tuning MobileNet and incorporating a custom classification layer, we achieve efficient feature extraction and classification without the need for extensive computational resources. Experimental results demonstrate the effectiveness of transfer learning in ECG classification, yielding a perfect classification performance with an accuracy more than 99 %. ​

During training, we observed fluctuations in accuracy at specific epochs (e.g., 30, 50, and 72), where accuracy momentarily dropped to 0.6–0.7, deviating from the expected trend of ~0.9. These fluctuations may result from batch variation, learning rate dynamics, or the model temporarily overfitting to specific patterns in the training data. However, the model recovered and ultimately converged to high accuracy. Such fluctuations emphasize the importance of using techniques like learning rate scheduling, adaptive optimizers, and increased batch sizes to stabilize training and improve convergence.

These results highlight the reliability of MobileNet in ECG classification, demonstrating its potential for real-time heart disease detection. Additionally, the integration of this approach into IoT-enabled healthcare systems can enhance accessibility and accuracy in cardiovascular disease diagnosis.

Curved Ceiling Ribs

Now there is more fashion. There is no so-called trends. Now chase after anything not necessary — nor for fashionable color nor the shape, nor for style. Think about the content that you want to invest in a created object, and only then will form. The thing is your spirit. A spirit unlike forms hard copy.

• Now there is more fashion. There is no so-called trends.
• Now chase after anything not necessary — nor for fashionable color nor the shape, nor for style.
• Think about the content that you want to invest in a created object, and only then will form.
• The thing is your spirit. A spirit unlike forms hard copy.

Now there is more fashion. There is no so-called trends. Now chase after anything not necessary — nor for fashionable color nor the shape, nor for style. Think about the content that you want to invest in a created object, and only then will form. The thing is your spirit. A spirit unlike forms hard copy.