Main Article Content
A new technique of watermarking for medical data protection using a neural network is explained in this research. The neural network is used in conjunction with the Levenberg-Marquard method to enhance the control of watermarking strength as well as the performance of traditional watermarking techniques. When compared to earlier data security solutions, the simulation results show that the suggested scheme is more effective. The watermark extraction approach used in the suggested method produced an image that was identical to the original. A method for identifying clinically significant motion characteristics from an ordinary video of a person is presented here. Our approaches for measuring gait pathology with common cameras make quantitative motion estimation more accessible in hospitals and at home, allowing researchers to perform large-scale investigations of musculoskeletal and neurological problems. Digital watermarking has been extensively researched as a way of preserving the intellectual property rights ofhigh-value-added digital photographs. Machine learning and Deep Neural Network (DNN) have recently driven usto robust applications in the huge advancement of Artificial Intelligence. DNN Algorithms have overcome severaldifficulties in areas such as image processing, audio recognition, and natural language processing; in particular,trained DNN models have made it simple for researchers to achieve state-of-the-art findings. However, distributingthese trained models is always a difficult process, especially in terms of security and protection. The experimentaltest was conducted extensivein order to present some watermark analysis in DNN. This paper suggests a DNNmodel for digital watermarking that examines Deep Neural Network intellectual property, embedding watermarks,and owner verification. This model can generate watermarks to protect against potential attacks (fine-tuning andtrain to embed). The standard dataset is used to test this method. As a result, this model is resistant to counter-watermark attacks. Our approach checks the ownership of all remotely extended deep learning models reliably andpromptly without altering the model's accuracy for standard information inputs. Digital information may now beeasily changed, copied, distributed, andstored, result inginarequirementforsecureinformationownership.