"Project Vision AI team on AMD AI League Hackathon"

Uncooperative images in real-world scenarios—such as low-light conditions, heavy blur, glare, and off-axis camera angles—pose significant challenges to standard optical character recognition systems. This application builds a specialized computer vision and OCR pipeline capable of extracting structured text from US license plates (including state names, slogans, and alphanumeric characters), speed limit postings, stop signs, and multi-line work zone warning signs. The solution integrates advanced image preprocessing techniques using OpenCV and PIL, including Adaptive Histogram Equalization (CLAHE) for glare/contrast correction, Fast Non-Local Means Denoising to eliminate sensor noise, and Gaussian Adaptive Thresholding to isolate foreground characters from complex backgrounds. The processed images are passed through a custom Tesseract OCR configuration optimized for uniform alphanumeric block extraction on AMD GPU-accelerated JupyterLab instances. The entire pipeline is structured to ensure high text-recognition accuracy under severe adverse conditions.