Non-Destructive Optical Document Highlighter & AI Engine

Physical highlighting permanently marks books, library archives, and legal documents, while digital notes remain disconnected from the physical source material.

Designed an optical capture workflow that isolates color-coded digital highlights without altering physical paper, pairing computer vision segmentation with text extraction.

Optical sensor capture workflow feeding an OpenCV image processing pipeline with HSV thresholding, contour detection, OCR bounding box extraction, and structured note sync.

• Formulated adaptive HSV color filtering thresholds to reliably detect transparent highlighter tints across varying illumination
• Engineered spatial bounding box merging algorithm to reconstruct multiline paragraphs from fragmented character contours
• Authored and submitted formal patent specification with the Indian Patent Office (Application No. 202511052205)

def extract_highlighted_regions(image_bgr, lower_hsv, upper_hsv):
    hsv = cv2.cvtColor(image_bgr, cv2.COLOR_BGR2HSV)
    mask = cv2.inRange(hsv, lower_hsv, upper_hsv)
    kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))
    cleaned = cv2.morphologyEx(mask, cv2.MORPH_CLOSE, kernel)
    contours, _ = cv2.findContours(cleaned, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    return [cv2.boundingRect(c) for c in contours if cv2.contourArea(c) > 100]