A TensorFlow-based image recognition system for captchas that works without image segmentation.
Captcha Recognize is an open-source machine learning project that uses TensorFlow to automatically recognize and solve text-based captcha images. It is designed to bypass captcha challenges without requiring image segmentation, using a convolutional neural network trained on labeled captcha datasets. The project solves the problem of automated captcha solving for testing, research, or accessibility purposes.
Developers and researchers working on automation, security testing, or machine learning projects involving image recognition and captcha solving.
It offers a simplified, segmentation-free approach to captcha recognition with high reported accuracy, built on a widely-used deep learning framework (TensorFlow) for reliability and extensibility.
Image Recognition captcha without image segmentation 无需图片分割的验证码识别
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Eliminates the need for complex image segmentation by recognizing characters directly from whole captcha images, simplifying the processing pipeline as highlighted in the project description.
Achieves up to 99.7% accuracy on certain captcha datasets after training, as demonstrated in the README with specific evaluation scripts like captcha_eval.py.
Includes dedicated scripts for training on multiple GPUs, accelerating model development and making it efficient for large-scale datasets.
Allows training on user-provided captcha images with specific naming conventions, enabling adaptation to various captcha formats without code modifications.
Relies on Python 2.7, TensorFlow 1.1, and Ubuntu 16.04, which are no longer supported and pose significant compatibility and security challenges for modern development.
Accuracy drops sharply to 52.1% on captchas from different generators, as shown in the README, indicating poor generalization to non-standard or more complex captcha types.
Requires manual dataset preparation, TFRecords conversion, and specific environment configuration, which can be cumbersome and error-prone for users without deep learning expertise.