Making Live Animals Transparent (Tartrazine-Assisted Optical Clearing)
Biological tissues scatter light due to refractive-index mismatch, limiting imaging depth. We applied a food-safe dye (Tartrazine) to temporarily reduce scattering in live mouse tissue, improving visibility of internal structures during in-vivo imaging.
- Led the image-processing pipeline: contrast enhancement, denoising, organ segmentation to reveal fine structures.
- Demonstrated how physics-aware preprocessing + ML improve dynamic bio-imaging readouts.
Computational ImagingImage ProcessingSegmentationIn-vivo ImagingPythonOpenCV
Automated Inspection of IC Markings
End-to-end vision pipeline to classify chip markings as Good/Defective (missing/unclear/distorted characters). Combined classical methods for detection/segmentation with learned recognition.
- Preprocessing: grayscale + Gaussian blur; Canny + hybrid Otsu/Sauvola binarization.
- Layout: contour/projection profiles; Fourier & Hough for orientation correction.
- Text extraction: MSER, SWT, watershed, CRAFT; recognition via custom YOLOv8.
- Rule-based final classification; highlighted impact of dataset size on overall accuracy.
OpenCVscikit-imageNumPyYOLOv8CRAFTTensorFlow/KerasManufacturing QA
Shortest-Path Timing in Circuit Graphs (ISCAS’85)
C++ tool that models ISCAS’85 logic circuits as weighted directed graphs and computes minimum-delay paths between chosen IOs.
- Bench parser preserving canonical names (e.g., G1gat) and robust CLI checks.
- Graph from gate connectivity; edge weights from fan-out; Dijkstra + path reconstruction.
- Useful for identifying critical timing paths that drive VLSI performance.
C++STLGraph AlgorithmsDijkstraVLSI TimingMakefileUnix
Deep Learning for Malaria Detection from Blood Smears
CNN-based classifier for red-blood-cell images (infected vs. uninfected) using the NIH malaria dataset (~27k images).
- Keras ImageDataGenerator for augmentation (rotation, rescaling, zoom, flips).
- Custom CNN with dropout; early stopping; evaluation via confusion matrix & reports.
- Model exported for reuse/deployment (malaria_detector.h5).
PythonTensorFlow/Kerasscikit-learnData AugmentationMedical Imaging
