Implemented core image processing techniques including color space conversions (RGB, HSV, Grayscale), histogram analysis and equalization, custom convolution filtering, noise injection and removal, edge enhancement, and video frame processing using OpenCV.
Built a complete OCR system without machine learning by generating CAPTCHA datasets, applying image denoising and deblurring, performing character segmentation, and recognizing characters using template matching, achieving over 90% recognition accuracy.
Developed applications using pre-trained computer vision models for cat face detection (Haar Cascades), face/age/gender estimation (OpenCV DNN), and automatic image colorization (CNN Autoencoder), while analyzing model robustness under rotation, blur, brightness changes, and noise.
Designed, trained, and optimized a Convolutional Neural Network for multi-class scene classification on the Intel Image Classification dataset, incorporating data augmentation, performance evaluation (confusion matrix & classification report), and architecture improvements to enhance accuracy.
Implemented a deep learning pipeline for restoring damaged historical photographs using encoder-decoder architectures. The project includes dataset preprocessing, model training, quantitative evaluation with PSNR and SSIM, and qualitative comparison of restored images against ground truth.