Abstract
The netdisk not only supplies users with plenty of storage, but also provide the facility to share data. Therefore it has brought great convenience to users, but it also becomes the important channel for malicious and infringing applications to spread. Due to the extraction of netdisk information will get into trouble with the input of CAPTCHA, it's difficult to detect the applications. In this paper, we propose a method to extract downloading links of netdisk and another method based on the Drop Fall Algorithm to undertake the recognition of merged characters. We combine the numbers of characters and minimum values to confirm the initial points. The performance analysis shows that the method of extracting link is effective, and the recognition of characters is suitable for the netdisk environment.
Introduction
Cloud storage platforms, or netdisks, have revolutionized file sharing. However, their anonymity and ease of use have made them a primary vector for distributing malicious software and infringing content. For security researchers and search engine developers, indexing these files is technically challenging. Traditional web crawlers are often blocked by dynamic link generation and sophisticated anti-crawler mechanisms, such as CAPTCHAs with merged characters.
Methodology
Efficient Link Extraction
To avoid the high computational cost of headless browsers like Selenium, this scheme utilizes a reverse-engineered network traffic approach. By identifying the underlying API logic, the system performs a three-step handshake to retrieve direct download links:
- POST Request: Sends necessary parameters including sign, timestamp, and bdstoken.
- GET Request: Triggers a 302 redirect from the server.
- Final GET Request: Extracts the final download URL from the "Location" field in the HTTP response header.
CAPTCHA Recognition via Modified Drop Fall
The most significant barrier to automation is the CAPTCHA system, which often features distorted and overlapping characters.
The proposed recognition pipeline involves:
- Pre-processing: Images are binarized using Otsu’s Thresholding. A Color Filling Strategy (CFS) is then applied to identify distinct character blocks.
- Modified Drop Fall Algorithm: Standard Drop Fall algorithms simulate a path between characters but struggle with high variability. This scheme improves the process by seeding "starting drops" at calculated intervals ($i \times W/N$, where $W$ is width and $N$ is character count). The algorithm then scans for the nearest minimum projection point to initiate segmentation, making the process more predictable.
- Character Recognition: Segmented characters are processed using the Tesseract-OCR engine.
Experimental Results
The system was evaluated using 100 shared links and 100 CAPTCHA images from a live netdisk environment.
- Link Extraction: The system achieved a 92% success rate in downloading shared files, with an average transfer speed of 184KB/s.
- Segmentation and OCR: For human-readable CAPTCHAs (61% of the sample), the system accurately segmented individual characters at a rate of 46.72%. Once segmented, Tesseract-OCR achieved a 57.89% recognition accuracy.
- Overall Success: The full-string recognition rate for 4-character CAPTCHAs was 22.95%. While seemingly low, this is sufficient for automated systems that can perform multiple retries to gain access.
Conclusion
This research demonstrates an effective scheme for penetrating the automated defenses of netdisk services. By focusing on API logic rather than UI rendering and enhancing character segmentation heuristics, the system provides a viable path for security indexing. Future work will focus on improving segmentation accuracy and de-obfuscating client-side JavaScript to handle evolving API parameters like "Logid."
Summary updated