sml2h3/ddddocr

ddddocr is a Python library for automated image analysis, focused on extracting text and detecting objects from visual content. Its core capabilities include character recognition that can handle alphanumeric, Chinese, and special characters, as well as object detection that returns bounding box coordinates for targets within images.

The library provides specialized support for solving slider CAPTCHAs by identifying the position of missing pieces using edge matching or image comparison algorithms. It also offers image preprocessing through color-based filtering to reduce noise from complex backgrounds, and allows users to constrain OCR output to specific character subsets for improved accuracy on targeted inputs.

For extensibility, ddddocr supports loading user-trained ONNX models with custom character sets, enabling recognition of specialized or proprietary CAPTCHA types. Performance can be enhanced through GPU acceleration, which offloads model inference to a GPU device for faster batch or high-volume processing. The library also includes a REST API server that exposes recognition, detection, and slider-matching functions through HTTP endpoints for remote access.