Steganography and Pixelation Based Privacy Preserving for Image

The rapid progression of machine learning has significantly bolstered the potential applications of data utilization, thereby augmenting the inherent value of data. However, the current landscape lacks effective regulatory strategies for data usage across various scenarios, resulting in substantial risks of unauthorized privacy breaches in collected data. Consequently, there is a crucial need to identify and extract private elements from collected data to facilitate both privacy protection and efficient data utilization. This paper introduces a desensitization process aimed at recognizing private elements in image data through semantic segmentation. Additionally, a steganography implementation is incorporated to formulate a splitting and indexing protection scheme for private data. Experimental validation illustrates that the computational cost of the desensitization process in this scheme is comparable to random computational costs. Importantly, the proposed approach accurately restores private information post-authorization, enabling the segmented storage and indexing of data.